This guy thought Theranos was a technological revolution and managed to be convinced of it while Holmes' dog was leaving steaming piles of shit in the corner.
>>>"Accustomed to the undomesticated life, Balto frequently urinated and defecated at will throughout Theranos headquarters," Bilton writes. "While Holmes held board meetings with people like Henry Kissinger, Balto could be found in the corner of the room relieving himself while a frenzied assistant was left to clean up the mess."
Suffice to say, I'm not sure this mans opinion is worth the hard drive it's stored on. He was taken in by a charlatan who wasn't even trying very hard.
We have all seen the limitations of ChatGPT at this point.
Also, there's the whole Kissinger is a fucking terrible person who should probably be considered a war criminal thing to top it off. His opinion is gobshite.
All of which may be both true and beside the point, especially since he shares the byline with Eric Schmidt and the dean of MIT's computing school. I'm genuinely not sure who there is cosigning whom.
ChatGPT is today. What's next week, do you think? Next year? But even if it is all bullshit, which I'm far from convinced is the case, how much does that matter if everyone who matters acts on the belief that it's not?
I dropped my trusty Samsung flip phone into the toilet, RAN to 2009 YouTube, learned how to take it apart, took it apart, dried the pieces, and put it into a dessicator over night. (I happened to have one to store infrared transmitting windows.) Worked fine when reassembled the next day.
But Kissinger can never get the sh*t off HIS glasses.
As much as there are problems with HK, I think most of us have realized this is the next chapter of the technological revolution. One that we all thought was supposed to be ushered by crypto/web3 but never really materialized for most.
That said I think this technology will only be exploited to make big tech only bigger.
Who's "we?" Crypto was always a stupid grift. I've never touched the stuff, because although I was born to be a sucker, I am at least wise enough to have realized this about myself.
I think there's something genuinely new in the world, and I think it is very foolish to assume too much about how that's going to play out. Humans will be human, because we always are. Beyond that, the only way anyone can know is by finding out.
The only relevant "crypto" is Bitcoin, when you're actually holding the keys and not delegated to an exchange, and it's primary purpose is not speculation but just transfering money without a trusted third party. I've been using it to move money for over 10 years, and many people I know do too.
Go running down the tracks, money falling out my pockets all the while, just in time to get flattened by the much larger train full of revenuers and regulators coming to clean up all the scams that haven't yet cashed out? No thank you.
For pity's sake, try to have a sense of the possible that can encompass at least a little more than a Mark 82. Computers that kill are most
of a century old by now, and it's easy to kill in any case - if it weren't, you'd need more than gears and lenses to make World War II weapons so fearsomely good at it.
A computer that tries earnestly to save a human life - I'd never heard of that before today, but now I have [1], and now you have too. If that doesn't suffice to excite your imagination at least a little bit, I'm bewildered as to what possibly could.
This has been done since the year 2000, with research demonstrating the ability in the 90s.
-------------
Are you kidding me? We've been putting robots inside of our guns and bullets for decades. Ever since the Nazis flew V1 Flying Bombs automatically during WW2 to automatically hit London, everyone's been putting robots inside of bullets.
USA did Proximity Rounds (Radar "AIs" that automatically blew up our bullets before hitting a target, allowing for a maximum "shotgun" effect to spread our anti-air pellets and maximize damage against Japanese Fighters in WW2).
This isn't even called "AI", its just how weapons work. Guns shoot 3km+ consistently, do you know why? Because we have "fire control computers", which under laymen's terms is an AI gun that calculates distance, movement, and other attributes to land our shots consistently.
ChatGPT and it's future variants completely automates propaganda production. And by enabling the creation of an unlimited on demand stream of fact free opinions targeted to the individual reader, it also undermines the economic basis for independent journalism and the fourth estate.
Perhaps the economic rules need re-writing. Perhaps Twitter will innovate this space and the outcome will be better for the world. I bet Elon M and wellwishers have insight.
If you polled educated laypeople, i.e. people who are successfull in non-STEM fields with little programming experience, I believe that many would say that ChatGTP has achieved spontaneous human intelligence using “the giant network has started to think” type reasoning.
At the liberal arts college Wellesley College in 1971, 80% of the courses were aimed at teaching students to regurgitate the dominant theories in the arts and social sciences, and then to defend these views with university-level paper writing. Not much different than Feynman’s encounter with Physics in Brazil. Indeed, isn’t that what the scripts written for the TV meat puppets are, at a ninth grade level?
Then the indeed, LLM, by regurgitating stuff that it has “read,” is indeed “intelligent,” and a lot of people better get good at using it, lest they be part of the percentage that is not good, and is therefore redundant.
No new information, zero understanding. If that’s you, you’re f*cked, the machines are ready. It’s certainly a revolution.
> What does anyone care what that war criminal Kissinger thinks?
I addressed that point elsewhere. [1] I'll expand on that here by saying that I don't know or care what Kissinger thinks, but as an eternal student of the present moment I do pay attention to what he says, because so do a lot of other people and because he hasn't spoken only for himself in public since well before I was born if ever. One doesn't therefore need to believe what he says, of course, but petulantly refusing even to attend or consider the fact it's him saying it is like disregarding Cicero's place in Roman history because you think he personally was an asshole. (He was an asshole! But an asshole can still need taking seriously.)
The opinion section of a modern paper is written for the rubes. The Journal's is no different - which is why Kissinger is a sometime contributor there, in general. Even so, finding Kissinger there alongside gray eminences drawn from both the academic and the industrial tendencies of computing is novel. The byline is a tour-de-force performance of authoritative knowledge, partly to reassure any rubes who do remember Theranos, mostly on behalf of arguing that AI is a new thing in the world and will produce a new humanity altogether - or, stripping the text of esoteric baggage, that AI is the mother and grandmother of all growth sectors, and can't help but produce enormous returns for anyone who pours funding into it early.
People will act on that, is the point that I'm making, and I wouldn't assume it will be limited only to funding or only to the private sector. More people will be trying to make an AI future happen tomorrow than were trying to do that last week.
In my opinion the thing with chatgpt and similar LLMs is that we have essentially seen Searle's "Chinese Room" thought experiment[1] played out in real life. We now know it is possible to produce a model which definitely doesn't think but which produces a simulation of thought that is utterly convincing to a lot of people. Therefore it is likely the case that we will be unable to determine from simply observing behaviours whether or not an AI is actually thinking in the way we do, and more broadly if we follow Searle, intuitions about AI which are derived from an analogy to our own thinking process are probably not valid because our brains are actually not computer-like. If anything, chatGPT has thrown the cat amoung the AGI pigeons not because it has brought the goal of AGI closer, but because it has shown us we don't currently know how we would be able to tell if we ever get there.
Why do you think it "definitely" doesn't think? If this is about lacking consciousness, then you should make that explicit. But then the question just gets pushed back further, why are you sure it isn't conscious?
The problem with the Chinese room at its core is the problem of attribution. We want to attribute properties like consciousness or understanding to "things" we are familiar with, and the only sufficient thing in the room is the man. But this intuition misdirects us to bad conclusions. The question to ask is what is responding when questions are being asked of the room. The responses are being generated by the algorithm reified into a causally efficacious process. Essentially, the reified algorithm is a set of properties without objecthood. It's so hard for people to see this as a live option because our cognitive makeup is such that we reason based on discrete entities. Considering extant properties without entities to carry them is just an alien notion.
We can go further and conceptualize a virtual objecthood for the process to help us reason about the entity we interface with in the Chinese room. We can then ask what properties does it have, how much does it understand, and so on. But we can't simply say it doesn't have any of these properties because it's merely a mechanical/computational process.
> Why do you think it "definitely" doesn't think? If this is about lacking consciousness, then you should make that explicit. But then the question just gets pushed back further, why are you sure it isn't conscious?
Because we know how GPTs work. They just return a high-probability "most likely next token" in response to what has gone before (the prompt and previous context). For that to be thought we would have to redefine the word.
"Just returning the next token" is an overly reductive description that misses what is powerful about these models. To accurately predict the next token requires the discovery of a model that has a correspondence with the system being modeled to varying degrees of fidelity. In the limit of infinite samples, the model must perfectly capture the structure and information dynamics of the system. Any deficiency in the model will necessarily reveal itself in a difference in output at some point. Of course LLMs aren't trained in the limit of infinite samples, but the question is how much of the system (a human mind producing language) has it modeled? If sentience or understanding is a necessary feature of a mind producing the kinds of texts we produce, it is an open question whether LLMs have or will recover the right sort of structure for these properties.
You're trying to pull yourself up by your own bootstraps here.
If sentience is required to produce the kinds of texts that LLMs produce then sentience is a feature of LLMs of course. But if sentience is required for my computer to produce the "Tada!" noise when it switches on then my computer is sentient also. (It's probably not sentient.) Nothing you have said establishes in any sense that LLMs are sentient, and the Chinese Room experiment shows that we can't just rely on our intuition.
I'm not sure what to make of comments like this. The discussion around LLMs and their capacities has just been universally horrendous.
>But if sentience is required for my computer to produce the "Tada!" noise when it switches on then my computer is sentient also.
But there is a large and obvious difference between your PCs boots up sound process and the output of LLMs. For one, its far more plausible that sentience is required to produce the kinds of texts that LLMs produce than to produce a boot up sound.
>Nothing you have said establishes in any sense that LLMs are sentient
I wasn't trying to, I was simply pointing out the error in concluding a lack of sentience from the arguments given. If you want to have a discussion about how/why LLMs might be sentient, I'm happy to have it. But most people don't actually want to have that discussion.
36 comments
[ 3.1 ms ] story [ 77.4 ms ] thread[0] “She Never Looks Back”: Inside Elizabeth Holmes’s Chilling Final Months at Theranos - https://www.vanityfair.com/news/2019/02/inside-elizabeth-hol...
>>>"Accustomed to the undomesticated life, Balto frequently urinated and defecated at will throughout Theranos headquarters," Bilton writes. "While Holmes held board meetings with people like Henry Kissinger, Balto could be found in the corner of the room relieving himself while a frenzied assistant was left to clean up the mess."
Suffice to say, I'm not sure this mans opinion is worth the hard drive it's stored on. He was taken in by a charlatan who wasn't even trying very hard.
We have all seen the limitations of ChatGPT at this point.
Also, there's the whole Kissinger is a fucking terrible person who should probably be considered a war criminal thing to top it off. His opinion is gobshite.
[1] The Trial of Henry Kissinger - https://en.wikipedia.org/wiki/The_Trial_of_Henry_Kissinger
ChatGPT is today. What's next week, do you think? Next year? But even if it is all bullshit, which I'm far from convinced is the case, how much does that matter if everyone who matters acts on the belief that it's not?
[2] https://www.youtube.com/watch?v=OQNV3KGQiS0
But Kissinger can never get the sh*t off HIS glasses.
That said I think this technology will only be exploited to make big tech only bigger.
I think there's something genuinely new in the world, and I think it is very foolish to assume too much about how that's going to play out. Humans will be human, because we always are. Beyond that, the only way anyone can know is by finding out.
The only relevant "crypto" is Bitcoin, when you're actually holding the keys and not delegated to an exchange, and it's primary purpose is not speculation but just transfering money without a trusted third party. I've been using it to move money for over 10 years, and many people I know do too.
"SquadGPT, kill everything in this building."
A computer that tries earnestly to save a human life - I'd never heard of that before today, but now I have [1], and now you have too. If that doesn't suffice to excite your imagination at least a little bit, I'm bewildered as to what possibly could.
[1] https://twitter.com/emollick/status/1629653225472131076
https://en.wikipedia.org/wiki/Bofors/Nexter_Bonus
This has been done since the year 2000, with research demonstrating the ability in the 90s.
-------------
Are you kidding me? We've been putting robots inside of our guns and bullets for decades. Ever since the Nazis flew V1 Flying Bombs automatically during WW2 to automatically hit London, everyone's been putting robots inside of bullets.
USA did Proximity Rounds (Radar "AIs" that automatically blew up our bullets before hitting a target, allowing for a maximum "shotgun" effect to spread our anti-air pellets and maximize damage against Japanese Fighters in WW2).
This isn't even called "AI", its just how weapons work. Guns shoot 3km+ consistently, do you know why? Because we have "fire control computers", which under laymen's terms is an AI gun that calculates distance, movement, and other attributes to land our shots consistently.
It's Kissinger's wet dream.
At the liberal arts college Wellesley College in 1971, 80% of the courses were aimed at teaching students to regurgitate the dominant theories in the arts and social sciences, and then to defend these views with university-level paper writing. Not much different than Feynman’s encounter with Physics in Brazil. Indeed, isn’t that what the scripts written for the TV meat puppets are, at a ninth grade level?
Then the indeed, LLM, by regurgitating stuff that it has “read,” is indeed “intelligent,” and a lot of people better get good at using it, lest they be part of the percentage that is not good, and is therefore redundant.
No new information, zero understanding. If that’s you, you’re f*cked, the machines are ready. It’s certainly a revolution.
This is just tabloid-level article wrapped in respectability because it's a famous past advisor and not some random tv show starlet...
With no competence in the field his opinion is as worthy as 99.9% of the population. Useless submission.
I addressed that point elsewhere. [1] I'll expand on that here by saying that I don't know or care what Kissinger thinks, but as an eternal student of the present moment I do pay attention to what he says, because so do a lot of other people and because he hasn't spoken only for himself in public since well before I was born if ever. One doesn't therefore need to believe what he says, of course, but petulantly refusing even to attend or consider the fact it's him saying it is like disregarding Cicero's place in Roman history because you think he personally was an asshole. (He was an asshole! But an asshole can still need taking seriously.)
The opinion section of a modern paper is written for the rubes. The Journal's is no different - which is why Kissinger is a sometime contributor there, in general. Even so, finding Kissinger there alongside gray eminences drawn from both the academic and the industrial tendencies of computing is novel. The byline is a tour-de-force performance of authoritative knowledge, partly to reassure any rubes who do remember Theranos, mostly on behalf of arguing that AI is a new thing in the world and will produce a new humanity altogether - or, stripping the text of esoteric baggage, that AI is the mother and grandmother of all growth sectors, and can't help but produce enormous returns for anyone who pours funding into it early.
People will act on that, is the point that I'm making, and I wouldn't assume it will be limited only to funding or only to the private sector. More people will be trying to make an AI future happen tomorrow than were trying to do that last week.
[1] https://news.ycombinator.com/item?id=34942962
[1] https://plato.stanford.edu/entries/chinese-room/
Why do you think it "definitely" doesn't think? If this is about lacking consciousness, then you should make that explicit. But then the question just gets pushed back further, why are you sure it isn't conscious?
The problem with the Chinese room at its core is the problem of attribution. We want to attribute properties like consciousness or understanding to "things" we are familiar with, and the only sufficient thing in the room is the man. But this intuition misdirects us to bad conclusions. The question to ask is what is responding when questions are being asked of the room. The responses are being generated by the algorithm reified into a causally efficacious process. Essentially, the reified algorithm is a set of properties without objecthood. It's so hard for people to see this as a live option because our cognitive makeup is such that we reason based on discrete entities. Considering extant properties without entities to carry them is just an alien notion.
We can go further and conceptualize a virtual objecthood for the process to help us reason about the entity we interface with in the Chinese room. We can then ask what properties does it have, how much does it understand, and so on. But we can't simply say it doesn't have any of these properties because it's merely a mechanical/computational process.
Because we know how GPTs work. They just return a high-probability "most likely next token" in response to what has gone before (the prompt and previous context). For that to be thought we would have to redefine the word.
If sentience is required to produce the kinds of texts that LLMs produce then sentience is a feature of LLMs of course. But if sentience is required for my computer to produce the "Tada!" noise when it switches on then my computer is sentient also. (It's probably not sentient.) Nothing you have said establishes in any sense that LLMs are sentient, and the Chinese Room experiment shows that we can't just rely on our intuition.
>But if sentience is required for my computer to produce the "Tada!" noise when it switches on then my computer is sentient also.
But there is a large and obvious difference between your PCs boots up sound process and the output of LLMs. For one, its far more plausible that sentience is required to produce the kinds of texts that LLMs produce than to produce a boot up sound.
>Nothing you have said establishes in any sense that LLMs are sentient
I wasn't trying to, I was simply pointing out the error in concluding a lack of sentience from the arguments given. If you want to have a discussion about how/why LLMs might be sentient, I'm happy to have it. But most people don't actually want to have that discussion.