Watson had a ton of promise. I feel like IBM was desperate for revenue and to reinvent itself as a younger more innovative startup-type so the marketing and sales people took over what was VERY clearly an immature technology and tried to sell it everywhere.
They were good at that, but then the tech didn't meet the lofty expectations that had been created.
ChatGPT is entirely different - people understand instantly how useful it is and don't need marketing and sales people to peddle it.
Part of it (and I was briefly at IBM after the Watson craze so I heard some complaining) was also that they'd rebranded a lot of other things as Watson to try to cash in on it.
IBM tried to sell the company I worked at Watson Weather or something, following their acquisition of The Weather Network I believe. Everything was Watson something. It was such evident bullshit.
They convinced executives to do a pilot. It was a shit-show. Quick and dirty API endpoints running on bluemix and spewing 500 internal errors at every gust of wind. Nothing remotely intelligent.
I was pretty glad I didn't have to deal with Bluemix. Instead I was told we'd be using AWS, had that change after I started, waited four months for a vSphere setup to be rolled out for me, and quit two weeks later.
ChatGPT is useful, but I’ve seen people get more wary about using it over time. It’s very good at making stuff that looks correct without being correct. That’s fine for a school assignment, but people hate looking dumb by having silly things slip through.
The hallucination meme has far exceeded the reality, at least for GPT4.
That said, it would be insane to use it at work to generate content in a domain you're not an expert in. It is an amazing productivity tool, but it does not change the reality that faking expertise at work is always going to bite you.
A screwdriver is really dangerous. If you stick it into your eye you’ll go blind. It would be insane to do that.
Such risks are probably not things that we as a society should let prevent us from having screwdrivers. Ditto for gpt.
Rather than copying its output for publishing, I assume most people are using it to have conversations with, and ask questions, probe the answers, discover new information, research or use that information, and follow up with the conversation over time. That's how I use it and it's now my first step before Google, for certain types of questions or thoughts. I don't see that utility really going away.
There are lots of people paying for Plus by now, even more using the API. With their prices as relatively steep as they are they should be rolling in income by now. As long as Microsoft keeps throwing money at them it's irrelevant anyway.
If we follow the money, the valuation, on either low or high end is absurd. It's starting to seem like weWork at a larger scale.
Revenue for 2022? $3M
No trace of financials for 2023, other than some expected 100M or 1B depending on the wind for this year. we can be sure of one thing, when numbers can be used to boost interest and claim a success, we get numbers release faster than accounting even settled.
My guess is that they are burning at least hundreds of millions by poaching the best AI talent from Google.
Infrastructure wise, about the same. And no, it doesn't help the bottom line to be owned by a massive cloud provider, it only distracts since there is a senior VP somewhere in there now who's non negotiable OKR is complete migration to MS Azure. But openAI will still be charged like any other client.
Welcome to vampirizing VC at its best. Put billion in to inflat the bubble, if it wins great, if it doesn't then you would have sold plenty as a prefered must have no choice parter for services, and by then insider info would have given you a head start to sell of a good chunk of your bags anyway.
OpenAI may win the prize for the most hyped business and fastest skyrocketing company in the entire history, it risks to not even explode but worse: to disappear in the deep sky with backers pulling their hair off or each others when they finally get to receive the in and out reports. Everyone else will have moved on, subscribing to a plethora of dedicated services built by tiny companies improving on moderate size models and building catered solutions.
It will be a great happy ending after all the spams we've seen everywhere over the internet about this gpt genius tech.
some businesses might incorporate chatGPT and some new businesses might form. as the technology (chatGPT) advances we might find ways around it's ability to "confidently lie" in subtle ways.
i've had a bad experience with it generating code and config, it's wrong of inefficient in everything it does.
it's kind of like a glorified search engine, it gets you only the information you need. It can generate examples too.
Yes. ChatGPT will be the Netscape Navigator of LLMs. The first one most people were exposed to at the time but quickly forgotten in a few years, but not because either the Web nor LLMs were a fad.
I don’t get this argument - every other type of software, whether bare servers, Office 365 or Google docs, image editing, etc. is moving to SaaS or cloud-hosted but LLMs are going to go the opposite direction and be more desirable to run locally?
To be clear: I'm not saying there won't be hosted options. It's just rare for the first mover to survive. The first movers in web-based office suites got bought out...by Google. OpenAI will inevitably have some cruft in its technology or unexpected burden in its business model that lets someone swoop in and steal or buy its thunder.
The crucial difference here is that people never got to use Watson. They just watched it play Jeopardy, and then went on with their lives. No one was able to actually use Watson in their daily life, so of course we all forgot about it.
IBM has ruined everything its touched for the past 20 years. The out of touch executives, insane amount of marketing/sales hype, and cycles of consolidation and cancelation of products turns people off. Why trust IBM with a critical part of your infrastructure or stack? If it weren't for the extremely lucrative legacy deals they get thru ERP sales to governments and F500 companies or z/os stuff, they'd be 6 feet under, but somehow, they are still a go-to vendor for all sorts of legacy trash. Don't get me wrong, they've produced alot of great technologists and contributions to software/hardware, but not a company I'd trust.
Not sure the analogy holds up, if we are looking for an analogy, AlphaGo seems to work better: it was an impressive and unexpected stunt, for sure, but it wasn't really clear how that would translate into business applications. ChatGPT is different, it's a business of its own right from the start, and it is its users who are falling over themselves thinking up new business applications, not just its creator. The bigger risk to OpenAI is that their moat looks increasingly shallow, more and more people are having success training their own models.
The “moat” for Open AI is their relationship with Microsoft. The secret weapon for Microsoft is their association with ChatGPT. The two Will feed on each other.
The “Open Source” ecosystem will drive the innovation and keep the technology in the news cycle and public mindshare. Many businesses will not or cannot pull technology out of the chaos of that world and will look to Microsoft.
Google and Meta can incorporate this stuff into their other products but it is going to be harder for them to sell it directly.
People are perfectly able to install and run their own PostgreSQL databases, or Kafka brokers. The software is free and doesn't need much resources.
Still, MSFT is selling its PostgreSQL-as-a-service and Not-quite-Kafka-but-similar-enough-as-a-service thing to a shit-ton of corporations which maybe once had the capability to host this for themselves, but have lost it while they migrated all of their stuff to the cloud. And they do so at a hefty premium.
The same thing will happen to LLMs. Anyone will be able to run an own, custom-trained variant of sufficiently capable models on their own hardware. But most corporations will not do it. Just like they don't operate their own Kafka broker. They will instead pay a lot for MSFTs cloud service offering, which of course comes with the crucial promise that their data is safe and secured and handled in a way that is compliant with all privacy laws. Which of course isn't true, but that doesn't matter, the promise is what matters. And currently businesses largely believe that promise, which is why the MSFT/Azure connection is indeed the actual "moat" of OpenAI.
> They will instead pay a lot for MSFTs cloud service offering, which of course comes with the crucial promise that their data is safe and secured and handled in a way that is compliant with all privacy laws. Which of course isn't true, but that doesn't matter, the promise is what matters.
In what way is this not true? Obviously there is no perfection here, only degrees of risk. But this is literally why people pick MSFT over others. They have by far the strongest culture around maintaining trust in the enterprise space.
I agree that MS and Google in particular have incredible adoption levers by integrating them with their existing services. But there will still be a huge, probably bigger, market for integrating AI with other apps. Think about Copilot, in principle not even MS has a moat there, other than restricting VS Code plugins. And that's not even talking about building your own counseling/... app.
Back during the Watson hype our company had a partnership with another company who was betting big on Watson. So, I was loosely around to watch it unfold.
It started off rosy, the CEO of the company we were partnered with had fully bought onto the hype train. Every meeting we had he couldn't help but mention Watson (by name). I tried to gently warn him that maybe, just maybe, he had fallen for the marketing hype and the idea wasn't going to be a blockbuster like he thought. It fell on deaf ears.
Over the course of the next year or so, they pumped big $$$ into the project. Eventually, the project was quite delayed and over budget and the CEO stopped bringing it up. I asked about the status of that Watson thing and the look on his face said everything. Not long after the entire leadership team was shaken up.
This story is a standard inside tech leadership discussions.
I love how acknowledgment of misinvestments are always brushed off, thrown down into the abysses of managerial due to incompetence losses, with little to no written traces of what happened. that look in their eye is also a warning to not bring this up again.
We also found ourselves "competing" with Watson. Watson promised Big Improvements For Free, while we were up front that you put in work for easier time down the road.
At least Watson was an unknown. We got mostly annoyed at other things where we could say "You already tried that and it didn't work, which is why you're talking to us". Even though they kept living in hope that something would turn big masses of noise into clean data with no effort.
A decade later, they might actually be getting that. I'm skeptical, but at least it's a real thing.
The public never had an opportunity to use Watson. They should have forgotten; it wasn't for them.
IBM's marketing department, at least at one time, understood that public recognition drove sales in the boardroom. Stripping the company of any and every product that was consumer or consumer adjacent hurt their recognition and reputation. Of course, this is just one of many reasons IBM is less of a company today.
Thinkpads are still designed in america, and its still basically an american company, just with a different owner.
Same thing applies to Motorola and other lenovo acquisitions.
End of an era for sure. To rhyme with another response on this post, another phase in IBM's retreat from public consciousness.
I do think maintaining some brand lustre in consumer-facing products for lines of business where it at all makes sense is a potent recipe for excitement in enterprise etc.
Well, there is also the issue that Watson isn't an actual product. They used the jeopardy game to get brand recognition, but the customers don't get that. They get a team of contractors to hobble together whatever open source software or IBM products are out there to build you a super expensive "AI" product with what seems like a significant rate of failure (conjecture based on internet comments over numerous failures). So there was never anything for the public to interact with in the same manner as they can with ChatGPT.
This is dead on. IBM pulls publicity stunts, often achieved by brute force. Then, it uses that marketing to sell contracts to customers whom they promise the world then, typically, fail to deliver on. Or as noted deliver on in the most basic way often by contractors and/or fellows (fancy word for contractors). But the money is paid, and the contract signed, so IBM got theirs. Rins repeat.
Pretty easy to look back and see their amazing "innovation" every 5-10 years, which never results in any lasting product.
In the in-between time, they farm employee IP via contracts taking all past and future rights, and profit on that IP ownership for no ones benefit but the IBM management. Disgusting company. Pour one out for Redhat.
"Follow the money" generally works as a proxy [0], and I can't get over the fact that Altman sold 49%, and effectively ceded control to MSFT for $29 B.
If ChatGPT was all that, you'd imagine at least 10x with OpenAIs founders keeping control being the baseline
It is all that. The problem is that there isn’t much of a moat. You can get GPT-3 level performance on a high end laptop and GPT-4 equivalent on high end hardware seems like it’s coming. A lot of advancement is happening in quantization and pruning and training can be done by anyone with access to enough compute.
This is going to just become a ubiquitous tool and part of the computing toolbox. I am increasingly thinking the winner take all dynamics of social media and search may not apply. There’s going to be many cloud hosted AIs and lots you can run yourself if you feel like spending a few thousand dollars on hardware. That cost will fall as more special purpose NPU hardware enters the market and acceleration even becomes a standard part of CPUs.
Is a GPT3-like model currently available for discussion type use and works on a laptop? How might I go about getting started with locally running a model of this quality?
The large Alpaca models can run on a M1 Pro MacBook or similar performance Linux PC with llama.cpp. You will need 32-64G of RAM and fast SSD for the large models. They show performance that reminds me of GPT-3 (not quite 3.5).
There are some newer models out there I have not tried yet like the open llama, GPT4all, etc. so I’m not sure how good they are. I get the sense they are still GPT-3 level but are achieving that with less RAM.
There’s a race on both for raw capability and optimization via pruning and quantization. The latter is important to make these things runnable locally without gigantic hardware. Lots of people have stuff with GPUs, fast CPUs, and 32-64G RAM. Few have huge workstations with hundreds of gigs of RAM.
Unless progress stagnates I can see something approaching GPT-4 that can run on under $5k worth of hardware in a year or so.
Open model progress seems to be lagging only 1-2 years behind big cloud hosted models.
The quality does not yet measure up exactly to ChatGPT (even 3.5), but yes it is possible
Probably the fastest way to get started is to look into [0] - this only requires a beta chromium browser with WebGPU. For a more integrated setup, I am under the impression [1] is the main tool used.
If you want to take a look at the quality possible before getting started, [2] is an online service by Hugging Face that hosts one of the best of the current generation of open models (OpenAssistant w/ 30B LLaMa)
I downloaded a version of that openassistant model for llama.cpp and it’s at least on par with GPT-3 or a little beyond. It’s to the level of being generally useful.
That’s the tech aspect, AI is essentially the new Linux
On top of that, given Metcalfe’s law, communication and coordination cost of a team grows exponentially with the size of the team, which means small teams have a huge advantage over big teams
Given that neither Google or Meta or any other company or open source solution is even coming close to the performance of merely the publicly released handicapped version of gpt-4 (no image interpretation, plug-ins is not fully released, etc), OpenAI might not have a moat (I’d argue there is a moat: the moat seems to be only a handful of people in the planet seem to be smart enough to innovate at this level), they have a year or more headstart and are clearly exploiting it and not slowing down. So yeah if I could buy openai shares I’ll buy it at a $500B valuation at this point.
It’s a lead but not a moat. If they slow down they are not likely to maintain that lead.
Similar things could be said for example about SpaceX with reusable rockets, but with a capital intensive industry like that it might take a decade or more for others to catch up.
Software iteration time is very fast, so if OpenAI slows down others could catch up in a year or two.
OpenAI offers several services: completions (i.e. chat GPT), embeddings, moderation, whisper, etc. Of these, I am convinced that embeddings are the true vendor lock in of our time. Once you base everything on one company's embeddings engine, it will be very difficult to switch to another without significant cost / effort.
As a concrete example, assume that embedding vectors are just two dimensional (in reality OpenAI's are 1536D). "cat" might map to [0.1, 0.9] in OpenAI while the same term maps to [0.7,0.3] in another company's engine. The mapping is completely non linear so matrix multiplication or other basic tools cannot be use to find a mapping. The "Rosetta stone" in this case is another massive LLM which would be exceedingly expensive to create. I'd posit that this will create a moat and OpenAI/MSFT are in a good position in this regard.
Completions have such a simple API I can see this becoming almost as ubiquitous as s3. It will be trivial for companies to switch this aspect.
There are open source AI models which can approach GPT-3 on an academic comparison done by the people who created these OSS models. There's however zero progress towards OSS products using these AIs, while Microsoft and various startups are already working on products using OpenAI's tools.
OpenAI/Microsoft don't need superior performance to have a moat (though they do have it, at least for now). They could just use the existing moats and make them unassailable. Or add new moats via API or hardware.
You are correct that typically OSS is unable or unwilling to do “product,” largely because OSS in its raw form is in fact enough of a product for its main developer-user demographic.
That being said I still see a much shallower moat here than there was in say Internet search. I can download a model in a few minutes. One could not download an entire web crawl index in any reasonable amount of time, nor could one update the index in real time continuously without enormous amounts of bandwidth and compute. DIY self hosted Google was to my knowledge barely even attempted due to the intrinsic difficulties.
This stuff is also not about communication or sharing, areas where there are very strong network effect moats. It’s not chat or social media or collaborative office software.
Generative AI is more like a stand alone application software. It can be hosted in the cloud but it’s easy to stand up competitors and it can be self hosted if you are willing to spring for the hardware.
I’m not saying you won’t have big dominant players, just that I see more opportunity for competition and diversity here than for lots of other things.
>This stuff is also not about communication or sharing, areas where there are very strong network effect moats. It’s not chat or social media or collaborative office software.
An example business need: Take all my documents and create an LLM acting as an internal knowledgebase.
Likely eventual Microsoft/Google solution: Press this button to take all your documents from your Office 365/GSuite account into our LLM. The LLM provides answers and links to the original document. We have automatic retraining, but you can remove/add data and retrain with a few more clicks. We have set up authentication and filtering so that unauthorized users can't get at your data.
Likely eventual OSS 'solution': Find your API key, download all your documents, and train them manually on the NVidia GPU cards you bought. Setting up a virtual python environment with CUDA is easy-peasy! Now, host the documents on your curlftpfs host so that our LLM could link to them.... (I could continue but this is too depressing).
I’m not saying there won't be small players, but the Google Research spin of 'no point in investing anything, Open Source will eat all!' was silly in the extreme. There'll be other products, and a smart enough OpenAI has good chances to create its own moats.
I don't think that's quite it. As I understand it their corporate structure basically has two entities a non-profit and OpenAI LP which has been accepting capped-profit investments. Different investors (early vs late) have received different capped-profit deals from 100x to 7x. Once all investors are paid off then all further profits are meant to flow to the non-profit (not sure what they are meant to do with it!).
Not sure about how this related to the founders, other than SamAltamn apparently having zero financial stake. I'm guessing the other founders may have put money in and have capped-profit deals ?
In the same sense everyone stopped being awed by the ability to communicate and share libraries of art, video, and text 100000X greater than every king in recorded human history for free. It's not going away, it's going to become embedded in our lives and taken for granted.
That’s a good thing. Maybe then people will stop fixating on whether it’s sentient and acclimatize themselves to the fact that it’s just an inanimate tool for us to use.
That's kind of the point. There's a pretty long history of humans talking to trees, rivers, tombstones, cool looking mountains, and household shrines. We'll anthropomorphize anything.
This is not a bad thing, but it's a very predictable thing. We're making inanimate objects simulate talking to you, and we're going to see consciousness when it's not there.
I personally think "chatGPT", will be a blip, it'll be remembered as a catalyst, not much more, and I'm not sure openAI really should have the valuation they do, open source is going to eat their dinner, when it takes openai 10 million dollars and an army from Kenya to create the same language a college student can build on a 2k computer for under $100 in cloud fees, it's pretty much game over... Microsoft on the other hand, because of the way they're pouncing on this and implementing it across all their products, they're the ones who really won here.
In 5 years it'll be the stable diffusions and alpaca's, and other open source models ruling and closed source will be mostly for vertical specific custom use cases. Generative AI on the hand --- it's never going back to the way it was before, not unless we have a Carrington event, that is.
Transformer-based LLMs seem a much more fundamental advance towards AI than Watson, but at the same time very primitive. Watson's Jeopardy win was over 12 years ago, and if we project that far ahead (2035) I'm sure today's LLM-based AI is going to be looking pretty dated too !
Connectionist, maybe gradient trained, AI seems here to stay, but no doubt future systems will be more brain-like in terms of capability, and not at all obvious how much of this simplistic pre-trained transformer approach will be retained.
I don’t think they are very comparable. Winning Jeopardy was an amazing feat and impressive, but the general public didn’t get its hands on IBM Watson to use for their own problem solving.
I did help a consulting customer, a long time ago, try to use the Watson NLP APIs, but I was not impressed.
I am 90% retired and in my 70s but I still do a lot of coding and writing for my own enjoyment. Even in my retired lifestyle, the OpenAI APIs (and Hugging Face models) are so useful and easy to use that they have changed the way I work. I feel like most of the busy work is now stripped away and I can be more creative. Bing+ChatGPT search has also been transformative and I look forward to seeing what Google builds Bard into.
ChatGPT lets normies use natural language to do complicated stuff that used to be only accessible to engineers ... so, no.
The winner might not be something called ChatGPT though.
The AutoGPT stuff is pretty ridiculous...
- ask GPT for the steps to perform a task
- then for each task ask GPT, hey you're good at coding, can you code this? or if not can you break it down into more detailed steps? and add those to the queue
as GPT gets smarter this type of agent workflow seems like it might work quite well for a lot of high-level tasks.
Yes, but that doesn't mean it may not also be forgotten, other than as part of computer history. These LLMs have multiple uses, and the use cases aren't going to go way, but LLMs are very primitive as a type of AGI and will soon be surpassed, and "ChatGPT" as a brand may not last too long either. After all, who nowadays really remembers early pioneer computer products like Lotus 1-2-1 or WordPerfect .. the genres persisted, but those specific products are only remembered in historical context.
I never used Watson personally. So I have no personal experiences to forget. ChatGPT on the other hand, even if I never use it again I’ll never forget.
I was part of the Merge Healthcare acquisition by IBM. We pioneered the DICOM standard that governs all versions of radiographs. We were acquired to "feed" Watson data - XRays, CAT scans, mammograms... and it never happened en masse. Apparently no one did their research on first case vs second case use scenarios. We only had access to a fraction of the radiograms we had touched so our prediction's confidence percentages were too low and models died in the queue.
81 comments
[ 2.7 ms ] story [ 123 ms ] threadThey were good at that, but then the tech didn't meet the lofty expectations that had been created.
ChatGPT is entirely different - people understand instantly how useful it is and don't need marketing and sales people to peddle it.
They convinced executives to do a pilot. It was a shit-show. Quick and dirty API endpoints running on bluemix and spewing 500 internal errors at every gust of wind. Nothing remotely intelligent.
That said, it would be insane to use it at work to generate content in a domain you're not an expert in. It is an amazing productivity tool, but it does not change the reality that faking expertise at work is always going to bite you.
I simply don’t recall anything else where most people’s opinions spiked then fell so rapidly before.
"The dirty secret of artificial intelligence" - https://news.ycombinator.com/item?id=35832168
It raised over $11B so far.
If we follow the money, the valuation, on either low or high end is absurd. It's starting to seem like weWork at a larger scale.
Revenue for 2022? $3M
No trace of financials for 2023, other than some expected 100M or 1B depending on the wind for this year. we can be sure of one thing, when numbers can be used to boost interest and claim a success, we get numbers release faster than accounting even settled.
My guess is that they are burning at least hundreds of millions by poaching the best AI talent from Google. Infrastructure wise, about the same. And no, it doesn't help the bottom line to be owned by a massive cloud provider, it only distracts since there is a senior VP somewhere in there now who's non negotiable OKR is complete migration to MS Azure. But openAI will still be charged like any other client.
Welcome to vampirizing VC at its best. Put billion in to inflat the bubble, if it wins great, if it doesn't then you would have sold plenty as a prefered must have no choice parter for services, and by then insider info would have given you a head start to sell of a good chunk of your bags anyway.
OpenAI may win the prize for the most hyped business and fastest skyrocketing company in the entire history, it risks to not even explode but worse: to disappear in the deep sky with backers pulling their hair off or each others when they finally get to receive the in and out reports. Everyone else will have moved on, subscribing to a plethora of dedicated services built by tiny companies improving on moderate size models and building catered solutions.
It will be a great happy ending after all the spams we've seen everywhere over the internet about this gpt genius tech.
some businesses might incorporate chatGPT and some new businesses might form. as the technology (chatGPT) advances we might find ways around it's ability to "confidently lie" in subtle ways.
i've had a bad experience with it generating code and config, it's wrong of inefficient in everything it does.
it's kind of like a glorified search engine, it gets you only the information you need. It can generate examples too.
Increasingly robust local LLMs as models get more efficient and capable and run on weaker hardware, or hardware catches up? Here to stay.
The “Open Source” ecosystem will drive the innovation and keep the technology in the news cycle and public mindshare. Many businesses will not or cannot pull technology out of the chaos of that world and will look to Microsoft.
Google and Meta can incorporate this stuff into their other products but it is going to be harder for them to sell it directly.
People are perfectly able to install and run their own PostgreSQL databases, or Kafka brokers. The software is free and doesn't need much resources.
Still, MSFT is selling its PostgreSQL-as-a-service and Not-quite-Kafka-but-similar-enough-as-a-service thing to a shit-ton of corporations which maybe once had the capability to host this for themselves, but have lost it while they migrated all of their stuff to the cloud. And they do so at a hefty premium.
The same thing will happen to LLMs. Anyone will be able to run an own, custom-trained variant of sufficiently capable models on their own hardware. But most corporations will not do it. Just like they don't operate their own Kafka broker. They will instead pay a lot for MSFTs cloud service offering, which of course comes with the crucial promise that their data is safe and secured and handled in a way that is compliant with all privacy laws. Which of course isn't true, but that doesn't matter, the promise is what matters. And currently businesses largely believe that promise, which is why the MSFT/Azure connection is indeed the actual "moat" of OpenAI.
In what way is this not true? Obviously there is no perfection here, only degrees of risk. But this is literally why people pick MSFT over others. They have by far the strongest culture around maintaining trust in the enterprise space.
It started off rosy, the CEO of the company we were partnered with had fully bought onto the hype train. Every meeting we had he couldn't help but mention Watson (by name). I tried to gently warn him that maybe, just maybe, he had fallen for the marketing hype and the idea wasn't going to be a blockbuster like he thought. It fell on deaf ears.
Over the course of the next year or so, they pumped big $$$ into the project. Eventually, the project was quite delayed and over budget and the CEO stopped bringing it up. I asked about the status of that Watson thing and the look on his face said everything. Not long after the entire leadership team was shaken up.
I love how acknowledgment of misinvestments are always brushed off, thrown down into the abysses of managerial due to incompetence losses, with little to no written traces of what happened. that look in their eye is also a warning to not bring this up again.
At least Watson was an unknown. We got mostly annoyed at other things where we could say "You already tried that and it didn't work, which is why you're talking to us". Even though they kept living in hope that something would turn big masses of noise into clean data with no effort.
A decade later, they might actually be getting that. I'm skeptical, but at least it's a real thing.
IBM's marketing department, at least at one time, understood that public recognition drove sales in the boardroom. Stripping the company of any and every product that was consumer or consumer adjacent hurt their recognition and reputation. Of course, this is just one of many reasons IBM is less of a company today.
https://www.youtube.com/watch?v=5g7WrTuL5AQ
I do think maintaining some brand lustre in consumer-facing products for lines of business where it at all makes sense is a potent recipe for excitement in enterprise etc.
Pretty easy to look back and see their amazing "innovation" every 5-10 years, which never results in any lasting product.
In the in-between time, they farm employee IP via contracts taking all past and future rights, and profit on that IP ownership for no ones benefit but the IBM management. Disgusting company. Pour one out for Redhat.
If ChatGPT was all that, you'd imagine at least 10x with OpenAIs founders keeping control being the baseline
[0] https://garymarcus.substack.com/p/is-microsoft-about-to-get-...
This is going to just become a ubiquitous tool and part of the computing toolbox. I am increasingly thinking the winner take all dynamics of social media and search may not apply. There’s going to be many cloud hosted AIs and lots you can run yourself if you feel like spending a few thousand dollars on hardware. That cost will fall as more special purpose NPU hardware enters the market and acceleration even becomes a standard part of CPUs.
There are some newer models out there I have not tried yet like the open llama, GPT4all, etc. so I’m not sure how good they are. I get the sense they are still GPT-3 level but are achieving that with less RAM.
There’s a race on both for raw capability and optimization via pruning and quantization. The latter is important to make these things runnable locally without gigantic hardware. Lots of people have stuff with GPUs, fast CPUs, and 32-64G RAM. Few have huge workstations with hundreds of gigs of RAM.
Unless progress stagnates I can see something approaching GPT-4 that can run on under $5k worth of hardware in a year or so.
Open model progress seems to be lagging only 1-2 years behind big cloud hosted models.
Probably the fastest way to get started is to look into [0] - this only requires a beta chromium browser with WebGPU. For a more integrated setup, I am under the impression [1] is the main tool used.
If you want to take a look at the quality possible before getting started, [2] is an online service by Hugging Face that hosts one of the best of the current generation of open models (OpenAssistant w/ 30B LLaMa)
[0]: https://mlc.ai/web-llm/ [1]: https://github.com/oobabooga/text-generation-webui [2]: https://huggingface.co/chat
On top of that, given Metcalfe’s law, communication and coordination cost of a team grows exponentially with the size of the team, which means small teams have a huge advantage over big teams
Similar things could be said for example about SpaceX with reusable rockets, but with a capital intensive industry like that it might take a decade or more for others to catch up.
Software iteration time is very fast, so if OpenAI slows down others could catch up in a year or two.
As a concrete example, assume that embedding vectors are just two dimensional (in reality OpenAI's are 1536D). "cat" might map to [0.1, 0.9] in OpenAI while the same term maps to [0.7,0.3] in another company's engine. The mapping is completely non linear so matrix multiplication or other basic tools cannot be use to find a mapping. The "Rosetta stone" in this case is another massive LLM which would be exceedingly expensive to create. I'd posit that this will create a moat and OpenAI/MSFT are in a good position in this regard.
Completions have such a simple API I can see this becoming almost as ubiquitous as s3. It will be trivial for companies to switch this aspect.
OpenAI/Microsoft don't need superior performance to have a moat (though they do have it, at least for now). They could just use the existing moats and make them unassailable. Or add new moats via API or hardware.
That being said I still see a much shallower moat here than there was in say Internet search. I can download a model in a few minutes. One could not download an entire web crawl index in any reasonable amount of time, nor could one update the index in real time continuously without enormous amounts of bandwidth and compute. DIY self hosted Google was to my knowledge barely even attempted due to the intrinsic difficulties.
This stuff is also not about communication or sharing, areas where there are very strong network effect moats. It’s not chat or social media or collaborative office software.
Generative AI is more like a stand alone application software. It can be hosted in the cloud but it’s easy to stand up competitors and it can be self hosted if you are willing to spring for the hardware.
I’m not saying you won’t have big dominant players, just that I see more opportunity for competition and diversity here than for lots of other things.
An example business need: Take all my documents and create an LLM acting as an internal knowledgebase.
Likely eventual Microsoft/Google solution: Press this button to take all your documents from your Office 365/GSuite account into our LLM. The LLM provides answers and links to the original document. We have automatic retraining, but you can remove/add data and retrain with a few more clicks. We have set up authentication and filtering so that unauthorized users can't get at your data.
Likely eventual OSS 'solution': Find your API key, download all your documents, and train them manually on the NVidia GPU cards you bought. Setting up a virtual python environment with CUDA is easy-peasy! Now, host the documents on your curlftpfs host so that our LLM could link to them.... (I could continue but this is too depressing).
I’m not saying there won't be small players, but the Google Research spin of 'no point in investing anything, Open Source will eat all!' was silly in the extreme. There'll be other products, and a smart enough OpenAI has good chances to create its own moats.
Not sure about how this related to the founders, other than SamAltamn apparently having zero financial stake. I'm guessing the other founders may have put money in and have capped-profit deals ?
This is not a bad thing, but it's a very predictable thing. We're making inanimate objects simulate talking to you, and we're going to see consciousness when it's not there.
In 5 years it'll be the stable diffusions and alpaca's, and other open source models ruling and closed source will be mostly for vertical specific custom use cases. Generative AI on the hand --- it's never going back to the way it was before, not unless we have a Carrington event, that is.
Connectionist, maybe gradient trained, AI seems here to stay, but no doubt future systems will be more brain-like in terms of capability, and not at all obvious how much of this simplistic pre-trained transformer approach will be retained.
I did help a consulting customer, a long time ago, try to use the Watson NLP APIs, but I was not impressed.
I am 90% retired and in my 70s but I still do a lot of coding and writing for my own enjoyment. Even in my retired lifestyle, the OpenAI APIs (and Hugging Face models) are so useful and easy to use that they have changed the way I work. I feel like most of the busy work is now stripped away and I can be more creative. Bing+ChatGPT search has also been transformative and I look forward to seeing what Google builds Bard into.
The winner might not be something called ChatGPT though.
The AutoGPT stuff is pretty ridiculous...
- ask GPT for the steps to perform a task
- then for each task ask GPT, hey you're good at coding, can you code this? or if not can you break it down into more detailed steps? and add those to the queue
as GPT gets smarter this type of agent workflow seems like it might work quite well for a lot of high-level tasks.
Chatgpt has many uses.