711 comments

[ 3.6 ms ] story [ 206 ms ] thread
In March, Sam Altman said that Pachocki's "overall leadership and technical vision" was essential for pre-training GPT-4.

https://twitter.com/sama/status/1635700851619819520

He also brought Pachocki to capitol hill soon after.
Yep, this and the capital hill stuff is what got the plug pulled. Thank God somebody finally recognized what an enormous threat this guy and his mad scientist friends are.
Aleksander in particular is deeply invested in AI safety as a mission. It's a very confusing departure, since most of the reporting so far indicates that Ilya and the board fired Sam to prioritize safety and non-profit objectives. A huge loss for OpenAI nonetheless.
The way Sam and Greg were fired maybe led him to no longer have faith in the company and so he quit?
More like the guy who engineered this situation is an asshole and they don't want to work for him.
Who's the situation-engineer for some of us duller but curious folks?
It's been confirmed to be Ilya.
Who was that? How are they an asshole?
Perhaps you could argue that he wants to stick with Sam and the others because if they start a company that competes with OpenAI, there’s a real chance they catch up and surpass OpenAI. If you really want to be a voice for safety, it’ll be most effective if you’re on the winning team.
Depends how much research is driven by Ilya…
I dunno, the moat Sam tried to build might make it hard to make a competitor.
We are about to find out if the moats are indeed that strong.

xAI recently showed that training a decent-ish model is now a multi-month effort. Granted GPT-4 is still farther along than others but curious how many months/resources does that add up when you have the team that built it in the first place

But also, starting another LLM company might be too obvious a thing to do. Maybe Sam has another trick up his sleeve? Though I suspect he is sticking with AI one way or the other

One funny detail is that the OpenAI charter states that, if this happens, they will stop their own work and help the organisation that is closest to achieving OpenAI's stated goal.
Maybe Sam wants to build something for profit?
But now it may be the regulations they've gotten in place will make it harder for any new upstarts to approach them.
> If you really want to be a voice for safety, it’ll be most effective if you’re on the winning team.

If an AI said that, we'd be calling it "capability gain" and think it's a huge risk.

> since most of the reporting so far indicates that Ilya and the board fired Sam to prioritize safety and non-profit objectives

With evidence, or is this the kind of pure speculation that media indulges in when they have no information and have to appear knowledgeable?

Twitter rumors from “insiders”
No. Statements from Ilya himself.
> most of the reporting so far indicates that Ilya and the board fired Sam to prioritize safety and non profit objectives

Maybe Ilya discovered something as head of AI safety research, something bad, and they had to act on it. From the outside it looks as if they are desperately trying to gain control. Maybe he got confirmation that LLMs are a little bit conscious, LOL. No, I am not making this up: https://twitter.com/ilyasut/status/1491554478243258368

lol sorry if this is clearly a joke but who cares if it's a little bit conscious. So are fucking pigeons.
it would be funny if Ilya followed the ranks of Blake Lemoine and went off for AI consciousness
A rudder only works as long as you are moving faster than the current. I can imagine (some) people concerned with safety also feeling a sense of urgency, because their ability to steer the AI toward the good is limited by their organization's engine of progress.
I'm going to predict more by tomorrow as the AGI secures its position.

In other news, Tesla FSD has been rebranded Saneer-Weeksbooth Autobot. So be careful out there folks.

Well it begins. OpenAI will be a shell of itself in very short time.

Advice: you can't win over a narrative, which is what Sam has become. People and resources will come to him, by themselves.

The advice seems off-point: overall it seems like Sam has operated opposite of the narrative he built (which aggrandized a vision of the future and downplayed the importance of making money) and got fired for it.

If I was OpenAI I'd want to quit today not because I want to follow Sam, but because the same bullshit that people left Google Brain et al. for has managed to catch up with them at OpenAI. It's a shame honestly, it was so exciting to see a company finally free itself of the shackles of navel gazing and just build things, but it seems like that's over today.

You misunderstood my point.

The narrative I am referring to is a simple one: take back what should be his. A.k.a, revenge. That is for sure a strong word.

People live on stories (collective imagination, if fancier), and what they like most is a wronged prince/princess took back his/her crown. It is the same with Taylor Swift rerecording her albums. The story potential will feed itself, until it is fully realized. The OpenAI board had committed a historical misstep, but maybe it is indeed what it is designed for: they hold no stakes in the business, so their view can't be held accountable in a business microscope. But money will really dislike it.

> If I was OpenAI I'd want to quit today not because I want to follow Sam

Sounds like I did? I think this is kind of a bad take, no one is quitting OpenAI before Sam has even had 24 hours to process being fired to help him take on his revenge arc.

Instead it sounds like people are angered by the process and powers that lead to him being fired like that, which is extremely understandable given the history of OpenAI. People forget half of OpenAI's competitive advantage was just not letting themselves be mired in self-sabtoage, the exact kind their board just demonstrated today.

I mean yes, but also no.

the dataset is the crucial bit of openAI. that takes a lot of time and money to make. So its perfectly possible for openAI to carry on innovating without these people.

But equally, it could turn to shit.

However, Sam isn't jesus, so its not like he's going to magically make another successful company.

Fair, it is still developing.

But I think one thing is certain, he WILL create another AI company. It seems very unlikely he would quit the business.

> the dataset is the crucial bit of openAI

I bet he'll train models on copious amounts of synthetic data made with GPT-4. There are lots of datasets in the open. That makes catching up easier.

No public facing model can be protected from data exfiltration and distillation. All deployed skills leak, your competition will replicate with less effort. And they only need to leak once and every subsequent model can inherit the skill. I think the first movers paid a high price for being first, and will quickly see their advantage erode. Latecomers will catch up and find AI easier to work with. The difference is made by the great fine-tuning datasets that are in the open, a growing lake of distilled abilities.

Another latecomer advantage is benefiting from significant innovation in the engineering part: flash attention, quantization, continuous batching, KV caching, LoRA, and more.

The new AI era will be more equalitarian. Catching up is much easier than discovering, and we can run AI privately, unlike search engines and social networks. You can't exploit SOTA advantage at scale. Being first is a fleeting advantage, the moment you go in the open everyone replicates.

Maybe one reason this is happening is because AI skills are very composable. Any addition to the skill repertoire already fits with other skills. This makes open sourcing skills very attractive. Of course, the datasets are what is being open sourced.

Sam should lead a group of outstanding engineers to rebuild another AI company. Maybe in the long run, leaving openai's naive board of directors might not be a bad thing for him.
> group of outstanding engineers to rebuild another AI company

openai was that backed by a nonprofit structure. and it still caused sam to be michael dell'ed/steve job'ed.

seems like the issue was with having a board/didnt have a majority on it. Zuckerberg having 53% voting power on the board is probably the greatest thing he managed. Anything sam does from now should follow the same.

/disclaimer - i have no idea how voting shares work.

2024 for OpenAI is looking bleak.
The world needs more truly open source AI models where the success and outcomes do not depend on a single figurehead or a corporation.
Who grey-texted this comment? So confused who could disagree with it. Is that the problem, this comment is so obviously true, it's just redundant?
The problem is that you cannot finance the training of a competitive AI model and then turn around and give it all away for free.

Who's supposed to pay for that?

Open and paid are not mutually exclusive. Someone is paying for Linux kernel development as well.
Universities usually do that, they work on open problems and publish their findings for free.
Since there is enough private investment available, why spend public money?
Good lord, is this the 'hacker' mindset now?
The truth or falsity of this statement turns entirely on whether more AI is good or not. You are in agreement with doomers that single figureheads are a weakpoint. That's the point. The disagreement is on whether having massive uncontrollable power widely dispersed is beneficial.
(comment deleted)
Are those the first cracks in the AI market bubble?
Is it a bubble if it’s useful and I use it dozens of time a day?
I agree AI is useful, but not to that extent to what it is valued on the market. I do not think that AI companies can deliver as much as they promise. With the driving core at OpenAI basically gone, I bet they will soon implode under the weight of their promises. Which means, investors will start pulling out their stakes. boom
What's the use of a newborn baby?

AI is as real as the mobile/internet/pc revolution of the past.

So many use it obsessively every single day.

I paid the 20 dollar subscription. I don't even subscribe to netflix.
Even if OpenAI implodes it will hardly impact other LLM-focused startups. In fact it would probably be a boon for them as people search for GPT alternatives.

Sam & Greg could start a new AI company by Monday and instantly achieve unicorn valuation. Hardly a burst.

Honestly this is exciting. Are they going to be the first company to achieve a $1 Billion evaluation within 3 days? Would they file the incorporation papers on Monday meaning they get that valuation within 24 hours?
This is almost certain to happen if they can snag the talent, I bet his phone is blowing up with VCs right now, revenge move and now unshackled from a non-profit nature of OpenAI
Where are they going to get the compute or the data?
Speaking for my own n of 1, ChatGPT Pro has almost entirely (>90%) replaced the Google search engine in my daily life. The results from ChatGPT are just so much better and faster.

That's got to be worth something, since Alphabet is a $1.7T company mostly on the strength of ads associated with Google search.

Chat gpt is not a good source of truth so can’t be used for information retrieval at scale. You might have a specific usage pattern that is very different to the majority of Google Search users so it works for you
Searching Google is not a good source of truth either; especially their infoboxes which have been infamously and dangerously wrong. And if you follow a random search result link - well, who knows if the content on that site is trustworthy either!
But you’re in control of your information retrieval, you didn’t have an unreliable agent synthesise bits in the middle.

Again - to each their own. But what people use google for GPT doesn’t replicate anyway (and what the Google business was built on) - which is commercial info retrieval.

As of a recent update, ChatGPT can do an internet search to answer "find a Thai restaurant near me." Of course, it uses Bing, not Google.

And for my single query above, ChatGPT searched multiple sources, aggregated the results, and offered a summary and recommendations, which is a lot more than Google would have done.

ChatGPT's major current limitation is that it just refuses to answer certain questions [what is the email address for person.name?] or gets very woke with some other answers.

the ad ridden shitware website filled with seo buzzwords with 100% opacity keywords also isn't a good source of truth. i'll take chatgpt over that.
Sure, use a different search engine then.

To each their own.

Personally, I don't have use case for comparing Google and ChatGPT that has truth as a requirement in the output.

For the majority of my use of ChatGPT and Google, I need to be able to get useful answers to vague questions - answers that I can confirm for myself through other means - and I need to iterate on those questions to hone in on the problem at hand. ChatGPT is undoubtedly superior to Google in that regard.

Agreed, but this will probably be limited to domains where there’s better products
Google is not a good source of truth at all, for anything other than hard facts. And nowadays, even the concept of "hard fact" is getting a bit fuzzy.

Google search reminds me of Amazon reviews. Years ago, basically trustworthy, very helpful. Now ... take them with a tablespoon of salt and another of MSG.

And this is separate from the time-efficiency issue: "how quickly can I answer my complex question which requires several logical joins?", which is where ChatGPT really shines.

I used it in the beginning, but now I am back to google... I don't think the results were better with ChatGPT.
Google ads span much more farther than in search - it's all over internet, on all websites, mobile etc.
For the past week or so I have been typing my search queries into Open AI, Bard and Duck Duck Go to compare them.

I haven't finished making up my mind, the the AI's are doing OK. I have only been asking it for code snippets that are easily verifiable.

Google doesn’t care if you’re going elsewhere to ask deep questions about Rust or whatever. They care way more that people go to them to look for the best bread mixer, or find a good restaurant, or a local massage therapist. In that regard I think Amazon is still a much bigger threat to them.

GPT is very useful as a knowledge tool, but I don’t see people going there to make purchasing decisions. It replaces stackoverflow and quora, not Google. For shopping, I need to see the top X raw results, with reviews, so I can come to my own conclusion. Many people even find shopping fun (I don’t) and wouldn’t want to replace the experience with a chatbot even if it were somehow objectively better.

Fair enough. My questions are more likely to be about Fast.ai, but I get your point.

Did you see the recent article about a restaurant changing its name to "Thai Food near me"?

Yea. No.

People stopping to use google for the small stuff will be the beginning of the end of google being the mental default for searches.

There is a wide variety of services available to people for specific use cases. When stack overflow came along, I used that for programming questions instead of google. But I still use google for most other searches.

I go to Amazon if I want to find a book or a specific product.

For the latest news, I come here, or Reddit, or sometimes twitter.

If I want to look up information about a famous person or topic, I go to Wikipedia (usually via google search). I know I can ask ChatGPT, but Wikipedia is generally more up to date, well-written and highly scrutinized by humans.

The jury’s still out on exactly what role ChatGPT will serve in the long term, but we’ve seen this kind of unbundling many times before and Google is still just as popular and useful as ever.

It seems like GPT’s killer app is helping guide your learning of a new topic, like having a personal tutor. I don’t see that replacing all aspects of a general purpose search engine though.

Your last paragraph - yes! Many people haven't realized this yet.

She/he/it/them is an amazing programming tutor.

> I agree AI is useful, but not to that extent to what it is valued on the market.

I agree, it's greatly undervalued!

Dotcom bubble...was/is the internet not useful?
It doesn’t feel the same since there are a handful of players in the space. I see your point though.
Was the housing market a bubble if millions of people lived in it and spent 75% of their time in it?
It can be useful in certain contexts, most certainly as a code co-pilot, but that and yours/others' usage doesn't change the fundamental mismatch between the limits of this tech and what Sam and others have hyped it up to do.

We've already trained it on all the data there is, it's not going to get "smarter" and it'll always lack true subjective understanding, so the overhype has been real, indeed to bubble levels as per OP.

> it's not going to get "smarter" and it'll always lack true subjective understanding

What is your basis for those claims? Especially the first one; I would think it's obvious that it will get smarter; the only questions are how much and how quickly. As far as subjective understanding, we're getting into the nature of consciousness territory, but if it can perform the same tasks, it doesn't really impact the value.

My basis for these claims is from my research career, work described so far at aolabs.ai; still very much in progress, but form what I've learned I can respond to the 2 claims you're poking at--

1) we should agree on what we mean by smart or intelligent. That's really hard to do so let's narrow it down to "does not hallucinate" the way GPT does, or more high level has a subjective understanding of its own that another agent can reliably come to trust. I can tell you that AI/deep learning/LLM hallucination is a technically unsolvable problem, so it'll never get "smarter" in that way.

2) This connects to number 2. Humans and animals of course aren't infinitely "smart;" we fuck up and hallucinate in ways of our own, but that's just it, we have a grounded truth of our own, born of a body and emotional experience that grounds our rational experience, or the consciousness you talk about.

So my claim is really one claim, that AI cannot perform the same tasks or "true" intelligence level of a human in the sense of not hallucinating like GPT without having a subjective experience of its own.

There is no answer or understanding "out there;" it's all what we experience and come to understand.

This is my favorite topic. I have much more to share on it including working code, though at a level of an extremely simple organism (thinking we can skip to human level and even jump exponentially beyond that is what I'm calling out as BS).

Then why can't the the grounded truth of ChatGPT be born of a body of silicon and emotional experience of zillions of lines of linguistic corpus?
Those zillions of lines are given to ChatGPT in the form of weights and biases through backprop during pre-training. The data does not map to any experience of ChatGPT itself, so it's performance involves associations between data, not associations between data and its own experience of that data.

Compare ChatGPT to a dog-- a dog's experience of an audible "sit" command maps to that particular dog's history of experience, manipulated through pain or pleasure (i.e. if you associate treat + "sit", you'll have a dog with its own grounded definition of sit). A human also learns words like "sit," and we always have our own understanding of those words, even if we can agree on them together too certain degrees in lines of linguistic corpora. In fact, the linguistic corpora is borne out of our experiences, our individual understandings, and that's a one way arrow, so something trained purely on that resultant data is always an abstraction level away from experience, and therefore from true grounded understanding or truth. Hence GPT (and all deep learning) unsolvable hallucination or grounding problems.

But I'm not seeing an explicit reason why experience is needed for intelligence. You're repeating this point over and over again but not actually explaining why, you're just assuming that it's a kind of given.
I don't see why "does not hallucinate" is a viable definition for "intelligent." Humans hallucinate, both literally, and in the sense of confabulating the same way that LLMs do. Are humans not intelligent?
I would appreciate another example where a major new communications technology peaks in its implementation within the first year after it is introduced to the market.
FTX / crypto, which just imploded last year.

Look, I'm an AGI/AI researcher myself. I believe and bleed this stuff. AI is here to stay and is forever a part of computing in many ways. Sam Altman and others bastardized it by overhyping it to current levels, derailing real work. All the traction OpenAI has accumulated, outside of github copoilot / codex, is itself so far away from product-market fit that people are playing off the novelty of AGI / the GPT/AI being on its way to "smarter" than human rather than any real usage.

Hype in tech is real. Overhype and bubbles are real. In AI in particular, there's been AI winters because of the overhype.

It seems we are talking about multiple different things. I never denied hype was a thing.

You’re talking about hype cycles now. Previously it seemed like you said AI was not going to be advancing.

LLMs are maybe headed into oversold territory, but LLMs are not the end of AI, even in the near term. They are just the UI front end.

> major new communications technology peaks in its implementation within the first year after it is introduced to the market.

Peak is perhaps the wrong word, local maximum before falling into the Trough of Disillusionment.

Yes, a bubble is a mismatch between price and value. Saying value > 0 is not disproving a bubble.
It is a bubble if it is overvalued. I don't think it is, but nothing prevents something useful from being a bubble, if the valuation is extreme.
Could be. Housing bubble happened even though most people lived in houses and still do. It's all in price vs utility. If the former gets way ahead of the latter, and people start trading just on future price raises, you've got a bubble.
There have been bubbles in the housing market in the past - houses are quite useful.

It's a bubble if the valuation is inflated beyond a level compared to a reasonable expected future value. Usefulness isn't part of that. The important bit is 'reasonable', which is also the subjective bit.

Electric cars are useful as well, but still, most electric car startups are -90% down from the peak. A financial bubble does not mean the underlying product is bad.
If the compute is paid with imaginary hype money? Doesn’t matter how useful a service is if providing it turns out to be unsustainable.
You must not use AI if you think this. AI is not the bubble. Everything AI will replace is the bubble.
AI art is already boring. its all a fad, eventually the cracks start to show and you can't unsee those cracks. very similar to bitcoin. tbh.
You're one of the ones who used to call bitcoin "buttcoin" didn't you? Very sure of yourself, very wrong.
Seeing btc is still quite useless apart from a few legitimate reasons, otherwise mostly being used for illegal purposes: yes, btc is a failure.
after i lost everything in the mtgox scandal, i used my last bitcoin to buy a lenovo thinkpad, which i used to learn computer programming and network engineering.
Oh, you mean because Bitcoin didn’t turn out as a stupidly-high risk speculation object without any value rooted in reality, wasting a sizeable portion of the power production of a world threatened by the climate crisis, swinging between orders of magnitude in valuation on Elon‘s whims? The Bitcoin that is irrelevant everywhere else but the crypto bubble itself?
Come on man, we're all on the AI hype train now. Get with the program, you're one bubble behind!
The interesting to boring scale depends on novelty. Given the ai oversaturation it works exactly as expected.

AI Art is currently in very early stage. In the real art space (3d modeling, sculpting, animation, vfx, animation, rigging, retargeting), it could make huge breakthroughs and multiply true artists' productivity in significant ways

I've noticed that those who worship productivity tend to appreciate "content creators" not artists :)
During my brief venture into 3d graphics I often found that 1 hour of truly creating is spaced between ten hours of fixing up UV's, redoing topology, finding cause for the seam, trying to make a watertight mesh for photon tracing, fighting the subdivision algorithm to retain details, diverting a edge loop, where it causes the less distress.

I call the first hour true productivity. The last part is, from the perspective of the end product, simply a wasted time. That's very similar to the boilerplate code everybody agrees is a necessary evil in the programming.

If AI allows to reduce the #2 it truly will have positive impact

It's interesting how some terms we use reveal our world view. What does referring to art as "product" imply?
"This radio thing is a dead end. I usually just get a bunch of static when I turn it on, or maybe an electrical shock. And even if it works, it's only a question of whether a tube burns out before the battery dies." - lazystar's grandpa, circa 1923
as i mentioned earlier, i was one of the first generation of bitcoin users. the problems with it have yet to be fixed.
AI today is in a weird position. What I can do today with AI was _inimaginable_ just 1 year ago. However, for a lot of people, the concept has worn out so fast, that they don't realize anymore what has happened... Some very intelligent people said some very silly things, such as that it was a bad search engine (it isn't a search engine) that it was a glorified word guesser (it doesn't exactly work as your telephone word suggestion). And so on and so forth. People always try to understand new technology through the lense of older technology. I do it, you certainly do it. This is how we grasp novelty. But AI is in a different dimension. I have been working in the domain for 30 years and I really didn't think we would reach this level in my life time. Talking to a computer to bring it to make some quite complicated task is INCREDIBLE... However, since communication is really ubiquitous for Humans, we tend to forget that it is an incredible achievement... In less than a year, we went for Scify ("Her") to reality, and in less than a year people have become blasé for something so fantastic... This is what consummerism did to people... They can't wonder more than a year...
It's a wonder. But if capex and opex are insane, and go up with every generation, the magic fades quickly.

Like, GitHub Copilot may be amazing, but if it looses money for every added user, if power users loose the company 4 or 8 times what they already pay, then maybe it's not an efficient use of compute resources.

Yes, this. What made tech companies valued at higher revenue multiples than other industries is that new users could start using a product at near zero marginal cost once the tech was built. New revenue at zero marginal cost. AI is great but expensive to operate and the expense grows in direct proportion to usage. New users come in and you have to stand up a new data center full of H100s to serve it.
> Talking to a computer to bring it to make some quite complicated task is INCREDIBLE...

Are there any good examples of this? I struggle to use ChatGPT, maybe I'm using it not cleverly (or deeply) enough.

Recently I had to share a documentation written in Word, which I had to format in Markdown to put it on Github. I transformed my document into a text file for each chapter and then I asked chatGPT to transform each of these files into a Markdown page. And I also asked it to improve the English. Then I asked chatGPT to translate each of these files into different languages. The result is here: https://github.com/naver/tamgu/tree/master/documentations Basically, I did in a couple of hours, something that would have taken weeks of tedious work
OK then that's tedious work, not complex work.

I know that lots of people have personal stories of using ChatGPT but I was hoping something publicly reported on or like a showcase of truly impressive usages somewhere.

You are kidding right. Taking a raw text file and detecting every single header, sub-headers, keywords and pieces of code to add the right markdown tags is a simple task to you?

Have you ever tried to make a Python program to do exactly that?

I only used couple of sentences to build my prompt...

No, what is the computational complexity of the task? The computational complexity of the problem is not about how long your code is or how long you took to write it.

I don't know your CS background but perhaps I do not view the terms "complex" and "tedious" the way you assume. A tedious parser is certainly tedious to write, but it is not (necessarily) complex. And from an engineering standpoint it is questionable that you lost all the formatting information from Word, which would have already demarcated what things were headers, code, and so forth. So, you had to use a roundabout way—an LLM—to recover that information from the semantics.

If what you're really arguing is that ChatGPT works well for language translation tesks, in this case translating mixed prose, code, and foreign languages--sure I guess that's great at productivity and removing tedium, but it's not that surprising a usage given what LLMs are. They are language translators.

In other words you're saying it's complex but your argument reduces a task that is straightforward but tedious for humans, to the problem complexity of natural language processing.

I think something terrible happened at OpenAI and we just don't know what it is yet, we will eventually. But I believe something unethical happened, something that might be even illegal. For the board to remove the CEO, it means that the board doesn't trust him anymore.
$MSFT might be down but all others are probably going to be up
We need to know why he was let go. All else is speculation. I doubt people would follow Sam if he was let go for a good reason.
Unless they don't know the reason. (But of course, it's hard to imagine what reason could be so secret that it's better to let people start jumping ship than share it with employees.)
We don't know if these people left to follow Sam. There's so few details surrounding what the actual catalyst of the firing was and neither said are revealing the details.
I think you can safely assume the reason they left is related to the firing. The question is,are they following Altman or are they just disgusted by the power struggle.
As I said in the other thread, this is 100% a AI doomerist hijack.
Doomerism when it comes to technology has always been such a weird mindset to me. "oh no, they're gonna take the horses/horse buggies/telegram/landlines away!"

"Goddamn, how dare they invent the bigger cannons!?" - Romans, 1453, in Constantinople probably. (One incident where I can use my exempt powers).

its not even that, they have literally one argument and its nanobots
The more realistic argument is that AI will be used as a power amplifier by the already powerful.
actually is the opposite, it would democratize power at an unprecedented scale, that's why corporations are funding these NGO's (useful idiots)
Through what mechanism would it democratize power? I thought the GPTs were already limited to regular end users due to computational constraints. Most people can't afford dozens of Nvidia GPUs and the API infrastructure to data mine.
Nah other way around. It would amplify the majority. That's why the powers-that-be consider it a huge potential problem.
Ah yes, because all those normal people will be able to run these powerful models on the devices that they currently own. Such a naive take.

The rich will ALWAYS get their piece of the pie, and once they've had their fill, we'll be left fighting for the crumbs and thanking them for their generosity.

AI won't solve world hunger, it will make millions of people jobless. It won't stop wars, it will be used as a tool for the elite to spread propaganda. The problems that plague society today are ones that technology (that has existed for decades) can fix but greed prevents it.

Don't misrepresent the problem.
what is the problem, in tangible terms?
Machines have so far replaced us in physical tasks, which has forced us to move largely to menial office jobs, typing on a computer, doing things machines are bad at. Over 80% of jobs in the US are office-confined (or from home, but that's not the point). We're actually a poor fit for those monotonic, sedentary jobs, our bodies and minds are not designed for them. And from that the subsequent devastating effects on our physical and mental health. But you gotta have a job, or you can't exist. The system throws out parts that are not useful. It's the nature of the system.

Well here comes AI to take those jobs. What happens, you think? Where do we go next? Do you imagine we'll all just sit idle and give out orders for the AI to fulfill? Recall: the system throws away parts that are not useful. And we're not better at orchestrating this system than we are at implementing it. Most people already struggle to handle the complexity of modern life. So they'll be thrown out.

Now think what happens with a society where most people are unemployed, unhappy and hungry, and businesses are mostly, not ENTIRELY mind you, but most self-sufficient machinery that does the thinking and does the footwork?

But even that doesn't describe the problem alone. It's more of an end game. Before this we'll see not-so-superior AI pollute our web, media, public space with quickly generated content, as actual artists and thinkers are displaced, unable to compete. Our culture will die first. And then, eventually, we'll start dying.

As I'm describing this, note I don't say this from place of fear. I don't fear this. I see it more as an obvious place for our civilization to go. We can't help it, because we don't decide where this civilization goes any more than your cells decide where you go, or any more than the atoms of your cells "decide" where the cell goes.

We're not in control. That's just evolution.

Say, when you're sick and you have cancer, those cells are part of you, but they harm you, so you cut them out, apply chemotherapy, and then if there's a prosthesis to substitute the organ you removed with a machine, you do it, and you don't think twice about it.

What makes you think our society as a whole is different? If humans are not good at what society needs, it cuts those people out, and replaces them with working machines. It's so plainly obvious. We pay lip service to human rights and the value of an individual, but clearly that's not what we end up doing. A politician is chasing money and power, and they don't mind starting wars to get them if they can. A business chases profit, so they don't mind automating away any employee they can. It's always been this way. So now that you can replace the human thinkers, businesses won't need human thinkers. And since there's nothing left humans are good at, society won't need humans.

Nanobots are easy and convenient (for a superintelligence). But it's not like they're necessary. ASI can take over the world the old fashioned way, it just takes longer and is harder to explain.
Most of the time the technology is not building an artificial intelligent computer that is capable of superhuman reasoning.

Have you used gpt4? it's reasoning capabilities match human ability. The more you think about it, the more scary the reality becomes. GPT 4 can reason through any mental exercise as well as a human. The rest of the work to make it autonomous is simple in comparison.

> GPT 4 can reason through any mental exercise as well as a human.

So can I.

And yet, people don't consider me an existential threat.

Mostly because I do not have nukes.

Yes. This point is almost always missed. 'Human Level' itself is not very high. And, 'Human Level' is still like Homer "doh, they unplugged me".

Current GPT doesn't have a physical threat.

But, take something like the movie "Colossus". Where they did give control of nukes. That was scary.

Now, go watch the Netflix show about AI. This GPT stuff is so far just fun apps.

The military already has AI that can out pilot a human in a F-16, you think it will stop there ? That is probably already old news.

Yes but you can't scale your brain by building a bigger womb.
Yann LeCun disagrees with you, and I take his word on it
gpt4 is capable of reasoning “in distribution”. Its reasoning drops when you go outside the goldilocks zone
> it's reasoning capabilities match human ability

All research that I have seen disagrees with this take. Ask GPT-4 a few basic block world problems and see for yourself if it can match human ability.

Please. Next year maybe but have a little respect please
>Have you used gpt4? it's reasoning capabilities match human ability.

I use it every day, and I have to often guide it like a 5 year old to come to the conclusion to help me the way I want it to.

>GPT 4 can reason through any mental exercise as well as a human.

So can my alcoholic neighbor. That should not be a benchmark of anything.

I agree. Critics will say something like "but don't you think it's sad that a machine was trained on the work of artists without their permission, and is now decimating the lives of the very artists who made its existence possible in the first place?". And, I agree, on its face it does sound sad. Very sad. But what these critics need to remember is that progress has always been a good thing, all throughout history. If something was true in the past, it will continue to be true in the future, and while it may seem difficult to envision how this will create a better world for all of us right now, it's important that we have faith and keep marching forward regardless.
Counterpoint: dogma often led to dark times.

We could try to think a little more deeply about things than "let jesus take the wheel"

Well, shit, maybe Altman shouldn't have stoked that by signing letters how about AI was an extinction risk then.
altman is a secret accelerationist and was just playing a ruse
(comment deleted)
As a doomer: yeah right, as if for-profit OpenAI wasn't an accelerationist hijack to begin with.
I just want to point out that it was exactly a week ago that OpenAI was openly announcing $10M pay packages to pull AI researchers from other major companies, like Alphabet. These other companies are well established and if OpenAI were a normal corporation might look at various hostile takeover strategies.

How much more cost effective to just fool half the board of a nonprofit into taking unnecessarily aggressive action.

(comment deleted)
Makes me wonder whether to keep building upon OpenAI? Given that they have an API and it takes effort to build on that vs. something else. I am small fry but maybe other people are wondering the same? Can they give reassurances about their products going into the future?
(comment deleted)
I am wondering the same. It’s a PR desaster to their dev community and i’m not even sure if Sutskever isn’t secretly happy about this.
I am at lost. Not fear, just lost.

Don't know what to do. Is my investment into their API still worth it? It feels very unstable at this moment.

Microsoft owns it. I honestly can not imagine trying to build a business on an API owned by Microsoft of all companies.
Would like to elaborate a bit?

Though my investment will be still tiny at the moment, but not other multi-modal model on the market right is as good.

I actually wouldn't be worried in the short term for exactly this reason. Microsoft has legal access to GPT-4 and is allowed to host and serve it via Azure. If OpenAI somehow tanks its API in the near term, MS is sitting on a gold mine and will make use of that by continuing to serve it. In the long term I am worried, but less worried if Sam and Greg form a competing co to continue to build.
What makes you say that?

Microsoft seems like one of the more reliable partners to build on compared to Google etc. just for the simple reason that their customers are large businesses and not breaking things for them is in their blood. Just like Windows backwards compatibility.

People have been successfully building on MS APIs for four decades now. I've been for nearly three.

What exactly are you saying?

Why would you build a business that Microsoft could sherlock in a second?
Isn't Microsoft famous for their insane API stability and backwards compatibility?
While I think the sentiment is overblown, if you compare Azure to AWS, Azure's stability is like Google's.
Microsoft is famous for a lack of adoption DESPITE backwards compatibility.
No I would not say that at all. Microsoft has gone through how many Desktop APIs for Windows at this point? At least a dozen I'd think.
If you're just using their completions/chat API, you're gonna be ok. As an ultimate fallback you can spin up H100s in the cloud and run VLLM atop a high param open model like Llama 70B. Such models will catch up and their param counts will increase.. eventually. But initially expect gpt-3.5-esque performance. VLLM will give you an OpenAI-like REST API atop a range of models. Keep making things :))
Thx. I will. My current interests mainly lies in benchmarking their vision model.

That being said, I might not go further relying on their APIs for something more serious

By that logic you could never use a third party API.
If you are building something that is end-user facing that relies on ChatGPT then that was always a huge and risky bet on the future of OpenAI.

In addition, it would likely be some time, possibly years, before it would be ready for production.

Perhaps recent events have just brought that more clearly into focus for you.

(comment deleted)
Also same here. Actually currently staying up late Friday night hacking on OpenAI API projects (while waiting for SpaceX Starship launch, it's quite a day for high-tech news!) - and wondering if I should even bother. Of course I will keep hacking, but...still it makes you think. Which is a very unexpected feeling.

Hugely more interested in the open source models now, even if they are not as good at present. Because at least there is a near-100% guarantee that they will continue to have community support no matter what; the missing problem I suppose is GPUs to run them.

Totally. I'll keep going too. I am just putting a nice GUI wrapper around the new Assistant stuff which looks damn cool. Project is half "might make some bucks" and half "see if this is good to use in the day job".
Yeah, the assistants api is pretty great. Curious if you’ve faced issues with certain things not working out of the blue forcing you to re-run threads?

For example, i have an assistant which is supposed to parse an uploaded file and extract useful info from it. To use this assistant, I create a thread and a run and attach it to the assistant with a different file-id. About half the time, the assistant simply throws up its hands and says it can’t parse the file I supplied with the thread. Retrying a few times seems to do the trick.

The assistants API is fantastic; I was just getting started with it. This news makes me reconsider -- but I also think it's inevitable that a compatible API will be released with open source underlying LLMs. I've deployed the OpenAI-compatible completions API over Llama2 in production with vLLM, and it works perfectly.

Do you know if there are any projects working on this? Even something like a high quality json-tuned base model would go a huge way toward replicating OpenAI's current product.

Sorry, no idea if what you’re looking for exists. For now, I was looking at integrating with OpenAI and productising some th ing but this situation is making me nervous.
I'm in this boat. Not for my startup but for side projects I was absolutely pinning my hopes on them unlocking access to tools and relaxing some of their restrictions in the near future.. a future which now seem unlikely.
I’d recommend trying to build out your systems to work across LLMs where you can. Create an interface layer and for now maybe use OpenAI and Vertex as a couple of options. Vertex is handy as while not always as good you may find it works well for some tasks and it can be a lot cheaper for those.

If you build out this way then when the next greatest LLM comes out you can plug that into your interface and switch the tasks it’s best at over.

The problem is swapping LLMs can require rework of all your prompts, and you may be relying on specific features of OpenAI. If you don't then you are at a disadvantage or at least slowing down your work.
Just ask the LLM to rewrite your prompts for the new model.
Does it really have that kind of self awareness to be able to do that successfully? I feel very sceptical.
I doubt self awareness has anything to do with it..
What else would you call the ability for it to adapt a task for its own capabilities?
Language modelling, token prediction. It's not much different from generating code in a particular programming language; given examples, learn the patterns and repeat them. There's no self-awareness or consciousness or understanding or even the concept of capabilities, just predicting text.
Sure but that kind of sounds like it is building a theory of mind of itself.

If it does have considerable training data including prompt and response when people are interacting with itself then I suppose it isn't that surprising.

That does sound like self awareness, in the non magical sense. It is aware of its own behaviour because it has been trained on it.

Just have it write 10 and bench them against your own.
I have a hierarchy of templates, where I can automatically swap out parts of the prompt based on which LLM I am using. And also have a set of benchmarking tests to compare relative performance. I treat LLMs like a commodity and keep switching between them to compare performance.
Just curious are you using something specific for the tests?
Isn’t the expectation that “prompt engineering” is going to become unnecessary as models continue to improve? Other models may be lagging behind GPT4 but not by much.
The dream maybe. You still have to instruct these natural language agents somehow, and they all have personalities.
Definitely, just like with games development, the key is to master how things work, not specific APIs.

AI tools will need a similar plugin like approach.

I have a good idea how transformers work and have written Python code and trained toy ones, but end of the day right now calling OpenAI nothing I can build can beat it.
That would go as well as trying to write a universal android iOS app or write ansi sql to work across database platforms. A bad idea in every dimension.
>Can they give reassurances about their products going into the future

They wouldn't have been able to do that even before Sam's dismissal

it's business and systems design 101 to actually worry about your dependencies. regardless of this drama, you should have thought about what you'd do if OpenAI shut down, or become your competitor, or gets worse, or is bought by MS or something.

> Can they give reassurances about their products going into the future?

emotional comfort is not the thing you should be looking for mate.

Indeed. It was a big enough battle to convince execs that building on top of OpenAI was ok. Now that conversation is pretty much impossible. You have the Microsoft offering, but to most muggles that just looks like them reselling OpenAI.

The board of OpenAI should have been replaced by adults a long time ago.

Take my opinion with some skepticism because I am retired and the massive amount of time I put into LLMs (and deep learning in general) is only for my own understanding and enjoyment:

In all three languages I frequently use (Common Lisp, Python, and Racket) it is easy to switch between APIs. You can also use a library like LangChain to make switching easier.

For people building startups on OpenAI specific APIs, they can certainly protect themselves by using Azure as an intermediary. Microsoft is in the “stability business.”

people can have multiple overlapping reasons for things, it doesn't have to be they are on "sam's side" or the "board's side" or whatever. you can agree w/ a firing and for whatever reason disagree w/ the way it was done or future direction of a company, etc. all just guessing anyway.
If you're an OpenAI engineer, now would be a good time to ask for a raise.
Google and Microsoft are todays big winners.
Microsoft not very much
They have the GPT tech, now they can poach the talent from openai without guilt
Why should they feel guilt for giving workers a chance to get paid what they really are worth?
Because capitalism is a bad thing for rich people when poor people also get to do it like they do.
Microsoft has a huge stake in OpenAI, so they're very much a loser.
IdK Satya might sense there's a wounded gazette in the midst, and starve OpenAI for funds, and pick up the pieces.
I don't think microsoft has the culture to be competitive on the cutting edge of AI.

If they try to integrate OpenAI, they will suffocate it.

So did Google some with DeepMind. All the big innovations were at DeepMind, and then they got bought, stifled and absorbed. Big companies have a hard time innovating, so buy it.
And then, 80% of the time, they suffocate what they bought.
(comment deleted)
The board is like a who's who of privilege and nepotism - https://sfstandard.com/2023/11/17/openai-sam-altman-firing-b...
(comment deleted)
Why nepotism? The article you linked to has no mention of particular family ties that got them there? If anything it’s a remarkably meritocratic board, when compared with most other industries

The fact that it’s all well connected wealthy people is kind of the point, the board is there (among others) to bring advice and experience

What credentials does Tasha McCauley have to be on the board of a company like OpenAI? I can think of 100s or perhaps 1000s of people more qualified to be there.
i think its more about nepotism being the child of someone important.
If you really want to nitpick you could say it's cronyism, but we all know what nepotism means in this context, and there's no requirement nepotism be based on family ties anyways.
On a side note, what is the with the ownership structure diagram?

OpenAI non-profit --owns---> Holding Company for Open AI non-profit

That makes no sense. How does a non-profit "own" a holding company, that own the non-profit?

(comment deleted)
(comment deleted)
(comment deleted)
While I don’t know what is happening in OpenAI, from my experience(long enough to have an opinion), top people often leave when the employer is doing well or passing the last hurdle because, by that time the employer has gained adequate presence and brand identity that anyone leaving can immediately end up in top FAANG or equivalent Corps with much better benefits and of course pay/equity and other things.

Only time employees leaving is a bad sign had been when the company had been in zombie mode too long(few years) and haven’t delivered anything yet or still in alpha/beta stage with crappy deliveries no one is using or has no significant uptake.

Also, if there are financial difficulties, then a lot of people leave in droves and some actually come back after finances get better.

FAANG would probably be viewed as a step down from OpenAI at this point (or at least until today)
Nothing wrong with greener pastures/lower steppes
Which is exactly why you can expect a good position and lots of benefits.
does not matter, even CEO of Facebook role is a step down. This is a different game. I doubt these people have to work anymore. they are after something else.
If you are a bright mid-career FAANG engineer, you probably don’t have to work. You can just move to Vietnam and live comfortably on your savings.

And people who are invested into “social impact” still care about their wealth and well-being despite living luxurious lives by the standards even of the developed world, let alone all other countries.

Err, how do you just “move to Vietnam” as a non-Vietnamese citizen? You’d need a visa and thus a job surely?

Like, sure you’d be able to live off your savings if you were allowed to stay in the country, but most countries have a short limit on a tourist visa and then you’d need to leave…

yea you missed the thing that happened lol
I don’t know if people realise that there really isn’t any more prestigious place to work at right now than OpenAI. You don’t leave there for a better opportunity.
They can ask for huge salary bumps which may provide better opportunities for thier families.
OpenAI pay is competitive enough (remember the articles last week about $10M offers?) and there's still plenty of room for their stock to go up. IPO or full acquisition by MS would both translate to a huge payday.
> I don’t know if people realise that there really isn’t any more prestigious place to work at right now than OpenAI. You don’t leave there for a better opportunity.

My guess is they know something we don't Or they assume sama being fired means the trajectory of OpenAI as a company has changed or is likely to change significantly?

People who work at places of high prestige, especially in senior roles, are often the kind of people who take part in creating those "better opportunities". :)
>there really isn’t any more prestigious place to work at right now than OpenAI. You don’t leave there for a better opportunity.

it's like you wrote this comment yesterday, and not in the context of what has happened

It’s like we’re talking about the motivation of people who left a company and thereby created the very context that happened?
> While I don’t know what is happening in OpenAI

Would it really have been so hard to find out Sam Altman just got fired without notice?

GPT5 pre-training just ended I believe. Brock, Pachocki, Szymon Sidor, would have likely all been involved.

These are huge losses. Pachocki led pre-training for GPT-4, and probably GPT-5. Brockman is the major engineer responsible for the efficiency improvements that enabled ChatGPT and GPT-4 to be even remotely cost-effective. That is a piece that is often overlooked, but OpenAI's advantage over the competition in compute efficiency is probably even larger than the model itself.

"Greg Brockman works 60 to 100 hours per week, and spends around 80% of the time coding. Former colleagues have described him as the hardest-working person at OpenAI."

https://time.com/collection/time100-ai/6309033/greg-brockman...

This is like a supervillain origin story, Greg and Sam are 100 % going to start something new now, even if it's just out of spite with how much both seem to have liked their work at OpenAI
If they were popular at OpenAI, I would say they have a good chance of succeeding too. They could offer excellent equity packages to all the best engineers and researchers and due to the non-profit nature of OpenAI (and hence no equity), these people might be very tempted to leave.
OpenAI is supposedly “capped profit” so employees do get equity with limited upside.

But yeah, since Sam and Greg were apparently pushed out because they were building too good of a business any OpenAI employees that were aligned with them are likely to jump ship and join them, and OpenAI will revert to the non-profit research lab it started out as.

Can they do that though? With all the obligations they have to MS.
Greg seems to be much loved by OpenAI employees, and generally inspiring person.
Would they even be able to compete with OpenAI at this point? Even without Greg and Sam they have Ilya, the models they've trained so far, institutional knowledge, datasets and billions from Microsoft. Could OpenAI be at escape velocity anyway to AGI just continuing on the track it's been on?
There are plenty of investors who would pour billions into these 2. MSFT got it's worth others want too.
Microsoft got the models. And their compute infrastructure. They’re powerful enough to bleed openai out if it’s in their interest. I’m not sure if there is much trust between the new leadership and microsoft.
> Could OpenAI be at escape velocity anyway to AGI just continuing on the track it's been on?

Could a nuclear energy company be at escape velocity to fusion because they are the best at fission? I wouldn't think so

I don't think its possible, at least for the foreseable future, they were heavily over indexed on Azure offering them discounted compute, not like they're gonna buy that amount of GPU elsewhere
On the other hand, Anthropic exists, so I am not sure.
I can imagine both Apple and Google happily positioning themselves in the same type of relationship as OpenAI-Microsoft.
Google had internal efforts, and Anthropic already. They definitely don’t have the compute to spare to split with another organization.

Apple surely doesn’t have a cluster that at all compares with the big cloud giants.

Oracle and AWS are really the only cloud left, and oracle is already renting to Microsoft for GPU compute.

>Apple surely doesn’t have a cluster that at all compares with the big cloud giants.

Apple has a lot of cash to throw at it. Question would be if Apple is even interested in it.

They should be. The slight improvements in messaging autocomplete in iOS 17 have made a noticeable difference in my texting. To have an iPhone that understands me, and a Siri that doesn’t say, “Here’s what I found on the web” is extremely valuable.
Idk. It looks like Apple is happy to sell you the hardware to do the image recognition etc on your own device and receive the result. They can claim privacy and save on computation.
Then they can charge.
The problem is that a significant amount of the already-made GPUs are in use. Even if they can afford to throw money at it, where is that money going to go?

The best they can do is out-bid their competitors, for the competitors hardware. I'm sure apple doesn't want to pay Google for GCP resource to train an AI. Again, there may not be enough companies renting out GPUs at all.

Apple is burning through more than a million USD per day for its research on Ajax and Co, they definitely have an interest in building big imo.
That's .1% of their annual expenditures
ALDI vs. LIDL
ALDI Nord vs Aldi Süd; Adidas vs Puma?
Yes you're correct, thanks!

Edit: apparently ALDI VS LIDL is an urban myth. It's ALDI that was split in two ..

Aldi and Lidl are each other's biggest competitors. Aldi Süd and Nord don't really clash because there is only two markets where both are present: Germany, where they aren't competing though (split in North and South) and the US (Aldi vs. Trader Joe's). Every other country only has one of the two Aldis. Lidl on the other hand is present in most large markets alongside one of the Aldis.
Seems to me that most of that is spent on twitter.
I am either skeptical or envious of such claims. Someone coding so much would quickly be launched into meetings to communicate one's results and to coordinate with others.

It would be my life's dream to spend 80 hours per week coding without having to communicate with others... but no one is an island...

It's possible, but harder than almost any other role. There are people at Google/Meta like this. Usually E7/E8 levels, "coding machines". It's much easier to go into a pseudo PM/TL/Director role though to hit those levels and income, so it's uncommon.

You really have to have a passion for coding to put in the hours and be very good at it. Incredibly rare, believe it or not. Lots of people think they are good coders but this is another level. Proof is in your commit/code review count/async comms being 10x-100x of everyone else in your org, and it's clear you're single-handedly enabling delivery of major projects earlier than anyone else could. Think of the pressure of doing this continuously.

> and it's clear you're single-handedly enabling delivery of major projects earlier than anyone else could.

You have to watch out with that.. I've seen whole projects pushed through by management where no one else was involved enough to review normally, but everyone had an interaction that implied they had only seen the top of the iceberg of problems with it.

Maybe Google and Meta are different than my company, or maybe I am not in the league of such star coder, but in my experience as soon as a demo of my code is delivered I am immediately launched into managerial mode coordinating other devs working on my code. I came to just accept it.
It's not about being rockstar or 10x. He was the chairman of the board (and President of the LLC). Practically speaking, he can work however he wishes within the company. Seeing that he went from CTO role to President role, it's fairly obvious that he got the opportunity to structure the role and the work to best fit him (and probably the company, too).
There’s always a hoarde of people second guessing the 10x engineer. Of course it looks impossible to regular folks. I have seen a few people like this. They’re real. Sometimes it’s even worth the dysfunction they cause to see this in action.
Same, I've seen it in practice and the numbers didn't lie, week on week on week. But you know, some people are very uncomfortable with someone else being called smart. Worse yet what if they're called smarter than what they actually are? Like an injustice in the universe, but comes from defensiveness I think.

I don't know this guy in particular so I have no clue though.

Not commenting about the ppl who are subject of this thread but talking in general. I have been lucky enough to have seen some of these 10x engineers but what is much more common is a 1x engineer feeling and treated like a 10x engineer because they are surrounded by 0.1x engineers.
Haha, that reminds me a lot of this quote from an Atlantic article on Freeman Dyson:

I asked him whether as a boy he had speculated much about his gift. Had he asked himself why he had this special power? Why he was so bright?

Dyson is almost infallibly a modest and self-effacing man, but tonight his eyes were blank with fatigue, and his answer was uncharacteristic.

“That’s not how the question phrases itself,” he said. “The question is: why is everyone else so stupid?”

That’s still 10x. If you think that’s worth mentioning, you should see the 10x engineers swoon over the 100x unicorn.
I am not second guessing the 10x engineer, that topic is not the one under discussion.

The topic at hand is “how did a high level engineer got to focus on programming”. And I am saying that the reason has to do more with his influence and role within the organization, rather than other reason.

I've seen the bugs of multiple 10x engineers multiply together for 10^n x bugs
In my experience I have encountered two 10x engineers:

1) Moves fast, flexes their authority to sweep small stuff under the rug until it is out of scope and can be "fixed real quick" later. Often leverages many subject matter experts through effective and persistent communication and learns quick enough to get PRs through the door (that sometimes need "quick" fixes later). Enjoys selecting items that benefit their career the most, at the expense of others on their team. Mentors only enough to onboard and increase his team's yield, not to aid their careers. Fueled by the recognition and validation of peers through PR/project completion.

2) Gets shit done, is the SMI themself. Solo code cannon, but PRs go in clean, beautiful to look at. May not get along well with some but not necessarily abrasive to work with especially being part of their direct team. Can be a great altruistic mentor if they spare 5% of their time. Enjoys what they do, and the technologies they work with. Fueled by personal satisfaction in their achievements, and in uplifting their team.

Typically when you see the type 2 engineer, they are also an architect. It is very rare that they don't seem to have knowledge of nearly the entire system and all its interactions.
Sorry, but what is SMI?
I just engaged in the kind of acronym abuse I don't enjoy receiving! It stands for Subject Matter Expert.
>There’s always a hoarde of people second guessing the 10x engineer.

Because most 10x engineers recognized by management as such are characterized chiefly by building out shoddy software extremely quickly that only they can understand.

In a similar dynamic, Doctors that are scored highly by patients often have pretty bad medical outcomes.

It could be also a sign of dependency hoarding and making you the bottleneck of the whole project. Bad architectural decisions, narcissistic need of importance or both. With those hours your partner starts to date with your friend. With experience I can assure you that position is not worth it. Not for you and not for the project. You end up draining your imagination. Over fitting is emerging in programming like it is emerging in the machine learning.
> Over fitting is emerging in programming like it is emerging in the machine learning.

That's a nice insight. I have been in that place many times, I was overfitting on my own imagination.

Filling up that mana bar is not easy.
This is a well debunked myth. You can commit a lot of code and commit better quality code, but there is an upper bound on productivity. If you don't get enough rest the quality diminishes.

Management and leadership of a team has a way bigger impact than any single individual contributor could ever have. Humans are generally limited not by intelligence but by motivation and vision. Directing people to achieve what you want is what allows the scaling of innovation.

Hero worship is a very human thing, but unscientific.

I'm not discounting management or leadership. These are also very critical roles that can make or break organisations. But I'd challenge your assertion that management has a "way bigger impact" than a single IC can have. Both are critical at companies doing internet scale products.
For the most part you would run out of things to code surely? Unless you really are a one man-band with a full understanding of the commercials, user feedback, support etc.
They’re doing groundbreaking research, there’s always something new to try.
OpenAI is an absolute unicorn, and not in the bullshit-1-mrd-vc-money-dollar sense but in being truly outstanding. Since all they do is software, that is solely because of the people involved, being able to do things and doing things that other people won't and achieving things that other people don't.

When it comes to sports it's fairly obvious what outliers look like and well accepted that they exist. I don't see a single reason to believe, that the same would not be true in every other walk of life or thinking that OpenAI just got lucky (considering how many people are trying to get lucky right now with less success in this space).

There are extraordinarily effective people in this world, and they are sparse and it's probably not you or me (but that's completely fine with me, I am happy to stretch myself to the best of my abilities).

> Since all they do is software...

For a certain definition of "software": when only doing one training run costs an 8 digits sum (requiring hardware one order of magnitude more expensive than that to run) I kinda dispute the "all they do is software".

It's definitely not "all software": a big part of their advantage compared to actually free and open models is the insane hardware they have access to.

The free and open LLMs are doing very well compared to OpenAI once you take into account that the cost to train them is 1/100th to 1/1000th what it costs to train the OpenAI models.

This can be seen with StableDiffusion: once the money is poured in training the models and then the model made free, suddenly the edge of proprietary solutions is tiny (if it even exists at all).

I'd like to see the actually open and free models trained on the hardware used to train OpenAI: then we'd see how much of a "software edge" OpenAI has.

And my guess is it'd be way less impressive than you make it out to be.

> I'd like to see the actually open and free models trained on the hardware used to train OpenAI: then we'd see how much of a "software edge" OpenAI has.

It would seem like you're talking about what "software edge" OpenAI has in the future, when others have caught up, while parent is talking about the existing "software edge" OpenAI has today, which you seem to implicitly agree with, as you're talking about OpenAI maybe not having any edge in the future.

They are using hardware, yes, but they are not creating (which is what I mean by "doing") the hardware. Anyone else with funding could have access to the same hardware for running their software, and other people did do that, and do do that (now, of course, in a drastically tighter supply/demand situation).

I do not wanna be flippant here: Obviously having easy access to money and a good standing with the right people is making things A LOT simpler, but other people could have reasonably convinced someone to give them money to built the same software. That's what VCs do, after all.

Regarding the rest: Feels very much like a different topic. I'll pass.

I can imagine this type of person to abide to their normal obligations during business hours, and code full time the rest of their wake-up time.

In my company, 80% coding for a senior SWE is rare. But if they deliver, management will give them some slack on the other evaluation axis. I have colleagues who work almost by themselves on new high impact projects. This has many benefits. No need to argue about designs, code reviews (people just approve blindly their code). The downside is that you need to deliver.

This is very true everywhere I've looked.

What also happens is regular developers (like me) want the same treatment as if they could end-to-end deliver "if they only let me", but many times can't, and actually need the structure and processes of a team. I've seen this freedom not working at all.

Indeed so ... the "structure" (call it bureaucracy of you like) is all of:

- an equalizer (entire team treated the same)

- a confidence booster (approval of others gives feeling of having done well)

- a way of distributing information (everyone is aware of all other team work)

You can run a team as a form of "competitive sport", and race everyone against each other; who churns out most "wins", and helpfulness, non-code-work, cross-team work are "distractors" to that objective hence undesirable and definitely not rewarded.

If the personalities in your team are "right" then this can work and by striving to best each other, all achieve highly. Have a single non-competitive person in there though... and it'll grate. Forcing a collaborative element into the work (whether by approval/review procedures, or by things like mentoring/coaching, or even just to force briefings to the team on project completion) creates a balance between the "lone crusaders" and the "power of the masses". Make the loners aware of, and contribute to, the concept of "team success", and give the "masses" insight into contributing factors of high individual performance.

yes there must be strong accountability for this to work (e.g. a self financed open source project or bootstrapped startup), not only do mid devs overestimate their appetite, motivation to grind and delivery, but also face the Curse of Development wrt communicating to the money people their value. Why should the rockstar grind away 50x harder than their coasting peers for 30% more salary? What happens when the bean counters reorg you or a manager labels you not a team player? Equity is the right form of comp to motivate this level of delivery and at that point it’s not about 50x skills but about sales and overcoming the communication gaps to establish a nonzero price for your equity. Which is why so many amazing niche projects languish and starve and the founder-engineer eventually breaks and goes and ships react apps for whatever empty startup has startup-investor fit that year
No one wants to be another woz
I think Woz got the absolute best deal but that's coming from a tinkerer's POV. People underestimate how much it sucks to be under scrutiny 100% of the time as a face of the company (i.e. Steve #2)
Why not?? Being another Woz would be amazing.
I just saw this. Glad to see I’m not the only one.

https://www.cnbc.com/2017/04/21/why-apple-co-founder-steve-w...

My take: He’s the Keanu Reeves of tech (or Keanu is the Woz of the film industry). The world can use more of this.

My attitude about money is pretty much the same as Woz's. There's a lot of us out there, but the worldview is so alien to the modern computer industry that it just doesn't register.
There's probably people whose main job was to read the code and then communicate it more broadly. This is also cutting edge ML where a ton of code is basically thrown away due to not panning out so possibly the amount that needs to be communicated is fairly small.
I'm trying to read and reread this over and over again to make sense of this but to me it sounds like in the comments people speak as if Greg brockman resigned while in the article he is not amongst the three names who resigned. What am I missing here?
Seriously, do we honestly believe 80 hours a week coding is a good thing. What, is he this bad at coding.

What about he spends 4 hours a week coding cause he’s so good at coding.

Way more impressive.

Might as well add LOC as a metric. Both can mean the person is extremely inefficient, over-engineering everything, and their eyes are begging for a break.

However! The best engineers I've been around do work a lot and they like it.

I’m not sure why it’s seen as ok to work 100 hours per week, or even glorified.

If instead of work it was something else it would be seen as a problem. 100 hours per week doesn’t leave room for anything else other than basic human needs.

“They like it”, well all addicts like what they’re addicted to, it doesn’t mean it’s healthy.

When you are a founder and working in a company doing bleeding edge it’s easy to work lots of hours. Maybe it’s not the kind of environment for you but others thrive in it. Lots of high demand type roles out unrelated to engineering that also have large hour workloads and compensate exceptionally well.
What do those people want the compensation for? To sleep on a mansion and back to work? Assuming they leave the office.

What I’m trying to say is that it is an addiction like any other and should be treated as such, not glorified.

Yes we understood your idea the first time around. And you still miss the point. It might not be for you but many individuals genuinely love their work. Either because they founded it, like the area of work, the people, or some combination.

It’s ok to not enjoy it yourself. Different strokes for different folks.

I don’t think it should be culturally championed but I don’t see it as an immediate red flag especially in the case of a bleeding edge company like OpenAI.

And I think you might be the one missing the point, because you keep saying it’s ok for them because they love it.

Surely all addicts love the thing they’re addicted to, but that doesn’t make it ok, even in the case where their addiction doesn’t ruin their lives short or mid term.

We understand. You don’t agree with it. Thank you for sharing.
Agreed. "I like it" might also just mean "I can't bear being alone with my thoughts" or "I can't deal with life and need the distraction". Not that that's always the case, it's probably more often than not in these situations and should be seen as a hint that there might be more going on.
I think a lot of addicts despise their addiction. Exceptions are the few highly acceptable addictions in society such as coffee. Nonetheless, doing anything in excess can be detrimental to one's health and livelihood and should be kept in checked, monitored.

OK now back to my 12 hour day. Not burnt out yet so I'm going to keep going. And yes, I LIKE IT!

80h a week doing stuff does not prove your level at it, it proves your work capacity.

Being good at something lies in the result and/or appreciation of your work by skilled pairs, which also seem to be there.

So why even claim he did 80 hours a week coding, while also being the chairman of the board?

Can we get a pllleeeeeeaaaase????

He’s clearly a terrible programmer and/or a terrible chairman and to be honest this news says he’s at least 1 of 2 on the above.

Or perhaps he's just a really exceptional person.

I don't see any comments here claiming that it's something that most people could do well.

You really think they'd let him anywhere near any of those two roles if he was pumping out tire fire code into production and constantly spewing erratic BS at strategy meetings the last eight years? Please.
Occasionally you meet people who shock you with how talented they are. I watched a couple of his presentations and he immediately reminded me of some of those people I’ve met before.
Realistically the odds of gdb being a terrible programmer are very very slim. He has been in the field for at least a decade, published papers, given talks, was the CTO of stripe ( a company generally respected for having sick technical infrastructure). If he works an inordinate amount, then it's probably cause he loves it. I would guess he is much more likely to be world class than terrible
Dumb people don't like other peoples success. It reminds them that they're dumb nobodies.
I guess I just don’t like BS in all its forms.

I don’t like nonsense PR stories or myths about people’s extraordinary prowess.

I just respond badly to BS and these statement have obvious BS if you stop for even a second to think about them.

On some level too, it offends me when I see right minded intelligent people in my community lapping it up.

So a couple of things.

Say I were to tell you that he was the President of openAI but he also did 80 hours of janitorial work per week.

Would you say that was a good use of his time?

Would you say that maybe he should be spending his time on being president of the company and not mopping up? You would be right.

Now substitute programming for janitorial work.

Now be a little more critical about things you see online.

What Brockman tweets is from technical standpoint the most mundane, boring, and obvious stuff I’ve read from a programmer. My reads of this guy have been he’s not working on any problems that are technically difficult (or interesting). It’s much easier to work long hours on easy problems. He also has a managerial vibe in all communications which supports my feeling.

Most programming work in any project and company is mundane, so I do agree someone taking care of all that without whining is actually extremely valuable. I couldn’t do it.

Still doesn’t really make sense to put him on such a pedestal like many in this thread. It seems like a cultural thing in the US to overvalue individuals, and downplay the importance of good teams.

I know nothing about the guy but judging his work and assuming you can guess the type of problems he works on based off his tweets is ludicrous.
Ah, the reverse of the old "he uses difficult terms so he must be super smart".
he is on a pedestal because everyone who’s worked with him said he’s amazing and effective.

But it is your right to assume what he works on from reading his tweets and leap from that to how this is an American cultural thing tho.

They said "senior researchers", and I would say a programmer is not a researcher if they spend all their time on programming.
This type of research happens in teams
Lots of research is mostly programming, e.g. my applied maths PhD. The way to try out software-based ideas is with programming.
I think the point was more than a lot of programming is not actually programming. Working in the industry, most somewhat complex systems require mostly work on paper, planning, research, reading documentation etc. and in the end some writing of code. Too often though, that is dismissed because it's "less agile" and a few years down the road the technical debt is huge.
There's no point building a hypothetical system. You have no idea if it works until you try it. And lol, documentation? For a system that doesn't exist?
I disagree.

Personally, I find it much easier to get lost in time and focused when I am working on something challenging. Time just flies by.

If I have to work on something boring / routine / repetitive I find it much hard to focus and time goes by so slowly.

Then my brain decides to look for ways to automate what I am doing. Perhaps a DSL or .. or .. o .. No work, remember work, but I could hmm if I write a Perl script i, No work you need to work, but it woud be work if i cold only

(I am diagnosed with ADHD)

I'm not diagnosed as you are, but I'm the same in that terms. Boring needs to be automated, challenging tasks need to be automated too.

So basically, my brain is lazy and try to find a way to keep it in that state.

I would rather have one of you than ten people who dig the ditch in front of them.

You will see the distance to be travelled and say let's build a airplane.

but incentives in most companies demand "progess" hence most projects start by piling the car high and driving off. it's when they are attaching floats to the car and paddling across the atlantic shouting progess reports back to shore that the value of automation comes to mind

don't worry about the ADHD - embrace it. (my hint - of the boring has to be done, make it the only thing, have nothing else).

Impressive analysis. Can’t wait for someone to tell you what kind of a programmer you are based on your HN comments.
How nice would be him tweeting about the real stuff that is their competitive advantage
By that definition Elizer should be working on really hardcore stuff, right? And yet his explanations about actual technical stuff come across as a guy that barely understands how matmul works.
It's definitely a Silicon Valley thing at least to overvalue individuals and downplay good teams. Keep in mind, there are a lot of young people in this site. It's pretty fun, you get boom-bust cycles like what happened with Musk and plenty of grifters trying to take advantage of this mindset.
Coding for 50 to 80 hours a week? Well, I call bullshit on that. Never seen anyone do that consistently and with high quality and quantity output.

Let's first define 'coding' before we jump into the details: 'coding' for me is sitting at your computer doing the work. It's not getting a coffee, chatting with a colleague, going to toilet or reading hacker news. So if you're reading this and claiming to do 100 hours per week of productive time, I call bullshit on that.

Being at the office for 60 to 100 hours, sure, I believe that.

When I was studying for exams at University, I did more than half of the work before noon. The rest was spread out over the afternoon and evening. At 20:00 my brain was dead. I could read a sentence, and nothing would stick. Read it again, impossible to process it.

So I always wondered how these other students could study until 2am in the morning. Well, turned out they didn't do shit in the morning. That's how they studied "all the way into the night".

Now back to my programming career: At my best I do 4 to 6 hours of concentrated coding per day. At my best, nobody seriously outperformed me. So if you claim to do more than x2 the work that I'm doing, I would love to see the output of that.

People like Cal Newport basically confirm what I've seen over the years. So do habits of the most famous authors.

Now, I can be convinced that it's actually possible. Take a look at Carmack, who claims to do 12 hours a day. He doesn't seem to be a bullshitter to me. So either he's counting time that I wouldn't count, like dungeon mastering a D&D game, or playtesting, or whatever. Or he's actually a super human work machine. Now he worked with Abrash, who seemed to do more sane hours. And in the end Carmack had high respect for the output of Abrash.

So yeah, if you know people who can actually do 14 hours of high concentrated coding 7 days out of 7, I would love to hear it and get some kind of confirmation that they're not browsing reddit and HN 50% of that time. And if you're reading this and claim to do 14 hours a day of concentrated work, I call bullshit on that you HN addict!

While I'm sure 100 hours a week is impossible, my dad did 6x11 work days which were dominated by coding pre-internet. You wouldn't know him, he burned out. I have personally witnessed him do more than 60 hours in a week coding. That said, his coding work isn't necessarilly creative. He can do a 12 hour stint of step through debugging stopping only to microwave frozen food. Or 12 hours of data analytic work. Since it is said this man is into optimization I'd say it is possible he's like my dad, he gets into this numb zombie state "change something, run again, look at profiler output line by line." Some people doom scroll HN or tiktok 14 hours a day, other's doom scroll a flame graph.
I also notice that the more complex a task is, the less hours of it I can do.

If I know what to write, and I just have to crunch out pretty straightforward code, I can do more hours (nowhere near 12 hours though, maybe 8 at best).

I can imagine the work your dad did, didn't include juggling a big complex system in his head, which seems to require a lot of mental energy.

That's basically also what Carmack states, that you can reach 12 hours if you plan your work to include some easier tasks for that day. But then again, I was never able to really apply that strategy.

Thanks for you take on it! :)

I think it mostly takes mental energy to addapt to change. I think you can focus on a large complex mature codebase and make small improvements so long as you're not radically chaging everything. Maybe it comes down to synapses firing versus synapses rewiring but that's just a laymans guess.
I dunno man. I feel like reading this was definitely work. It surely wasn’t leisure. Have you ever examined code output by gpt where it looks impressive at first glance? That’s what reading this was like. If you’re reading this, I’m just kidding. If you’re not reading this, I’m not kidding.
This is Time Magazine...Once wrote a piece about Gates and Warren Buffett spending the weekends on Math quizzes...
Coding for 10/12hs daily isn’t healthy, is not sustainable and is not the way to live. I love coding and I being in front of a computer all my life, since I was 8. All the meaningful life moments in retrospective weren’t at the screen. Life occurs outside the screen.

I understand that sometimes is worth it, to create a great product, solve something important or just for fun. But beware

Meaning of life is highly subjective. Just because your personal beliefs don't value coding highly doesn't mean that is universal.
I agree with you. Just bear in mind that later on in life, you can’t go back.
(comment deleted)
That sounds spectacularly unhealthy
Brockman did not get fired from his job at the company, only from his position as chairman of the board.

Did nobody read this damn article? It's 5 paragraphs long, take 1 minute.

It’s so juicy statements like this become stale in hours. Exciting times.
Greg Brockman hasn't quit OpenAI. He quit the board[1].

   As a part of this transition, Greg Brockman will be stepping down as chairman   
   of the board and will remain in his role at the company, reporting to the CEO.  
[1] https://openai.com/blog/openai-announces-leadership-transiti...
he tweeted he's leaving the company a few hours after the press release
Do you have a source that they just finished GPT-5 pre training or is that just speculation.
What is the truth about OpenAI achieving some sort of "true" AGI?
Probably almost guaranteed to be false.
(comment deleted)
My prediction: https://news.ycombinator.com/item?id=38309611&p=7#38311488

So here's my theory which might sound crazy. Sam planned to open a new AI company and taking away openAI's top talents to his company. And breaking up openAI into non profit and his for profit company.

Sam's first tweet after all this has, just hours after this article:

> will have more to say about what’s next later.

So either he knew that he was about to be fired or at least was prepared.

Also based on the wording of the press release, Sam did something that the board absolutely hated. Because most of the time even if he did something illegal it doesn't make sense to risk defamation by accusing him publically. Also based on his video of yesterday at the APEC summit, he repeated the similar lines few times:

> I am super excited. I can't imagine anything more exciting to work on.

So here if we assume he knew he was about to get fired, the conclusion is clear.

(comment deleted)
Some Microsoft bullshit. I've read an article where they said Greg to be a really hard worker at Open AI,so I can't possibly think of a reason other than this. Maybe they'll do something with Google? Who knows
they seems to finally achieve AGI, they realized they don't need as many people anymore :D