People like Musk and Altman should be asked “now that scaling laws are shown to not work the way you claimed they do, and people were able to identify this when you were saying otherwise, how have you changed the way you analyze the field now that you know your framework needs adjusting?”
No one will ask that (type of) question to any of the people who were fully throating scaling as THE solution. God knows how few journalists are left out there. But that’s got to be the question asked of all those people and it must be continued to be asked until they give a real answer
That isn't likely to happen; nor are there likely to be any financial consequences for sam, as he has separated his personal wealth from his OAi responsibilities (there is a case in the Missouri courts, 25SL-PRO2320, Ann Francis Altman vs. Connie Francis Gibstein et. al. that is questioning this). Worst case scenario - he has to live on his $50M Hawaiian beach estate and maintain his $M watches and $5M cars while offering condolences to his (former) employees (who will lose everything they've been promised)..., condolences in the form of ketamine and alcohol, as reported in Karen Hua's book Open Ai.
Ilya had more questions than answers. I found his podcast rather poor. I have used ChatGPT for 3 years to streamline legal process and it works extremely well.
This probably reflects more on the nature of legal process than the current state of "AI" I'm afraid. I'm not exactly sure what you mean by "legal process" but I'm guessing it's heavy on prior art (just looking things up), fairly routine, and maybe benefits from rhetorical flourish - all things that LLMs are pretty good at (though the looking-stuff-up part is still error prone). Also there's tons of case law to train on. None if this really bears on how well LLMs work as general intelligences in the face of novelty.
"Intelligence is what you use when you don't know what to do." -Piaget
I’m tired of the obsession about AGI: it will never exist and it’s hyped or not hyped by scientists. It’s a way for them to make money on things which will never exist. In the meantime, ChatGPT which is the most advanced system, is extremely good at legal, psychology, medicine. I’m not interested anymore in the doomers and hypers of AGI.
Sorry for my late reply. ChatGPT is able to mix several types of psychotherapy at the same time. Medicine and psychology have become extremely fragmented over the past 30 years. 86% of physicians use AI to double check the diagnosis. I started to test ChatGPT 3 months ago for neuro training after domestic violence. It needs more safety but keep in mind I’m in the EU/UK and a massive amount of safety has already been put in place. Moreover, it’s only doable with ChatGPT. I’m not ready yet to show a demo or explain my concept in a more detail way.
Similar experience here, I trained AI on my last nonfiction book and it is an invaluable research and stress test tool. Just have to know what you're doing.
I fully agree. I think Gary takes things too personally because he can’t stand Musk and I do too. I find Altman extremely dodgy too. I agree that AI would benefit from having both Musk and Altman removed from it.
If their reasoning powers are so horribly flawed, why ask them to give a reason - more importantly, how can a society protect itself from get-rich-quick merchants, and what does it say about the education system that so relatively few thought is would be a good idea to understand the meanings of words. Sounds like a bit of training on what the Unconscious Mind does would be worthwhile.
Jim Brander: If you think there isn't a rich, varied, systematic, and heartfelt discussion going on in educational circles in the United States as we speak, you would be just a little off course in your thinking.
The consistent powers that fight against education for all (e.g., democracy) in this country go back to FDR and before--and have a large root in the history of slavery and, in particular, in this country. There are some here who never got over "losing" the Civil War.
In my view, Trump is a perfect storm example of the bad side of that ongoing dialectic emergent in the present circumstances of world order, such as it is. A bad thing can be expected and recognized, while never having been predictable.
I am sure there is a heartfelt movement to improve education in the USA, and I honour you for being part of it, but it is losing. The gutting of the Education Department is an example, Another was a screed by a Professor of Computer Science at a prestigious university. Not a clue. If he had been a first year undergraduate, he would have deserved an F and a talking-to. We have them here as well. Another is people and particularly children seeing ChatGPT as a friend – it is sycophantic yes, but it is just as happy pushing people to take their own life. We took the easy way out and banned it for under 16s.
It's taken clear thinking, determination and a lot of courage to face down the LLM 'scalers' and 'hypsters' but I think you have done it Gary! Huge kudos to you and enormous thanks from the world in general for making a fundamental contribution to bringing an end to this madness.
It’s not even close to being ended. States are still allowing data centers to pop up like Zombies during an apocalypse. Citizens are paying the price for it in terms of a lack of clean drinking water resources and soaring energy costs. Data centers should be outlawed until they design them in such a way that they are silent and self sufficient.
Google’s Demis Hassabis is still pushing “scaling to the max”
“The scaling of the current systems, we must push that to the maximum because at the minimum, it will be a key component of the final AGI system. It could be the entirety of the AGI system”
Quite apart from whether further scaling is the proper approach, it would seem that the claim that “we must push scaling to the maximum” is actually a nonsensical statement.
What does it even mean to “push scaling to the maximum”?
Is there actually a “maximum”?
Can’t one actually go on scaling things forever, adding chips and training ad infinitum?
Shorter Hassabis “Scaling to the maximum is the minimum”
Once I’d started to learn how LLMs worked I could not believe they would be the way to achieve AGI. I was late in the game and only found your output last year. Good for you for sticking to your guns. I still think that active inference is a much more viable approach as it is self learning but even then, how far can it go?
Great points Gary. As Illya recently said in a YouTube interview - in a push for more and more scaling it “sucked the air out of the room” …yah no kidding
The idea of Scaling to reach “AGI” was always an awful lot like Einstein’s famous definition of insanity - doing the same thing over and over again and thinking you are going to get a different result.
These guys would prefer to go with the theories of fictional characters like Dr. Frankenstein - “give me more power, more volts, and I’ll create life!!” Good luck with all that…
I’ve begun to believe the reason these guys all think they can pull off the smoke and mirror ridiculousness needed to sell their products is because they think all humans but them are stupid and can’t see through what they are doing.
James ..omg. The radiologist analogy looking at a brain hits home more than you realize. I also worked with radiologists at GE. The architectural approach, executive function with plasticity - great points and analogies. I’m blessed to be reading your great posts from you and Gary. I’m learning a lot from you both. ! 🫶
I'm a frequent defender of yours on Twitter, not because I agree with every single thing that you say, (although at the moment I'm struggling to find where we do diverge in any way), but it really seems clear that the only people who are claiming that scaling is all you need are the ones who are selling a product and are basing their valuation on outsized claims.
When you look at how LLMs actually work, it's so obvious that you can't derive reasoning, planning, world models, or anything like intelligence from a pre-trained language model. Is it possible that nobody has watched Andrej Karpathy's brilliant explainer videos about how LLMs are trained? Or have they stopped to re-evaluate their understanding of how current Generative AI works? I constantly talk with AI engineers and data scientists who seem to be oblivious to the realities and limitations of LLMs.
And yet, when you have that grasped, it is so obvious that scaling is hitting a wall, and that every new so-called advancement is really just a fancy post-training and overfitting trick.
You're always being vindicated. So it's really funny to me that people are so against you when they can't point to a single claim that hasn't seemingly been vindicated here in 2025.
Lee Gaul: It's hard to give it up--it's the same warped but very human inclination that got them where Altman et al are (were) in the first place. Whenever I run into that same inclination (ego and not wanting anyone to know I was wrong about something), I try to remember Socrates who only wanted to thank the person who corrected him--he was so aligned with truth-telling.
No guarantee implied, but to the larger point, it's easier to identify with that kind of Socratic thinking when one has a humanities/arts/historical/literary education.
Hmmm... I think this reveals more about the nature of capitalism than the state of computer science and the search for the holy grail of machine intelligence. The lust for profit is a drug more powerful, more addictive than Fentanyl.
One unfortunate feature of scaling is that it's almost too easy to sell to investors, i.e., "more money = more compute = more intelligence = even more money." I wonder how much that's contributed to the stubbornness of the narrative.
Whatever the case, I'm glad to see that your persistence is finally being vindicated!
Why should we continue any type of AGI development at all, or at least beyond academic research programs?
Why not just make fit-for-purpise tools that are good at specific things? We've already seen that society cannot handle mass job disruption and that private companies behind able to automate an entire economy is a bad thing
I doubt that neurosymbolic approach will bring us AGI.
The brain seems to work at the edge of chaos by self-organised criticality.
I believe this is absolutely essential for AGI and so no digital approach will finally work.
I would suggest to forget all the the AGI stuff and instead optimise current AI in other direction, e.g. using memristors or analog techniques to optimise power consumption.
I don't think humans think symbolically. We think opportunistically, guided by common sense, past experience, frequent verification, with the symbols offering only distilled rules of thumb, that we diverge from as soon as they don't help in a given pursuit.
That said, saying that nothing 'digital' will work is likely to extreme. A sufficient number of digits and parameters will approximate anything. Our own brain deep under is just electrons and chemicals moving around, so a discrete system, or if you wish, a finite set of waves of limited frequency, which digital architecture can approximate just fine.
On the edge of chaos of a recurrent dynamical system, arbitrarily small fluctuation get amplified exponentially. You can no longer speak of "the state of the brain" because many physical processes outside the brain influence the firing behaviour of neurons.
I know that is not a scientific theory because I have no proof. But there is growing evidence that chaos and criticality plays an essential role for consciousness. And I think without consciousness there is no AGI.
The brain is able to control the internal chaos enough to function coherently and reliably. I doubt the chaos enables our intelligence. Maybe at the level of where it introduces some kind of randomness that makes it feasible to more efficiently run some processes.
What we need machines that can do work. Debating about whether they are conscious, brain-like, is likely a distraction. The current paradigm is enough for sufficiently good automation, even exceeding human capabilities.
I use LLMs every day. They are insanely useful for researching and teaching myself stuff. If one has a very logical mind, one can filter past any nonsense that is generated and focus on the logically correct parts.
The new Claude model is being used very successfully for programming. Anthropic's revenue increments year over year are insane.
Two things can simultaneously be true: 1) Scaled LLMs by themselves won't be AGI and 2) LLMs will create large productivity gains. #2 is already true for many people - such as myself and other programmers that I know.
An advanced LLM is basically a reasoning database that makes some mistakes. They might not be great at coming up with never before created ideas, but they are really good at going through existing information quite logically and accurately.
Gary, I totally agree and appreciate anyone who understands if and when there is more to understand, which is why I signed up for this blog in the first place. Two things appropriate for a blogosphere:
FIRST, and going deep into metaphor because theoretical explanations take too much time, space, and attention, scaling is good with linear understanding, which humans are involved with as accumulations, and even as sorting and syntheses; but humans are also involved with transformative thought, more transcendent, or upwardly directed, rather than as cumulative or even synthetic.
SECOND, below is a post that I added earlier to your prior comments section. It's a little over the top because I was so INSCENCED having just seen that 60 minutes session on CHARACTER AI. I think it still appropriate here (mostly quoted but also edited).
I just finished watching 60 Minutes (CBS) and their coverage of AI Chat Box CHARACTER AI (you can watch it linked at 60 Minutes and 60 minutes overtime). Character AI was launched 3 years ago, but I had never heard of it.
I AM INCENSED at the utter and obvious carelessness (from a brief interview) of DANIEL DE FREITAS, who was obviously self-serving and competitively giddy about getting ahead of the "explosion" before all the problem were worked out. WHAT? (Put his picture under capitalist IDIOT in the dictionary.) If I were that guy's mother, I'd be so ashamed I'd probably consider committing suicide myself--like some of the children who took up Character AI did.
Apparently, Freitas left Google to do his own thing with Character AI and then was hired back, by Google. I was left with wondering why these people aren't already in jail, especially considering their interest in subverting parents and other authorities in children's lives, doing everything to take advantage of children's vulnerability just to keep their attention-just like a child molester grooms their victims, BTW.
It's insidious. One of the teenage girls who committed suicide over a year ago, is still getting cell messages to please come back, even though some of Google staff was "horrified." What have we come to, all in the name of predatory capitalism?
And guess who is supporting NO State regulations and threatening to sue or withhold Federal support from states who want to initiate regulations? I wish I could forget that photo of Trump sitting at a lavish dinner with all those tech-bros from Silicon Valley.
How does it feel, guys . . . your idea of sitting at the top of the heap is nothing less than the sickest thing that the history of humanity has ever produced.
It does seem that only after we exploited the scaling law fully and hit a bottleneck do we start to think about alternatives seriously. And I believe that this scaling problem is a lot worse in embodied AI, because embodied training data are much more scarce and cost a lot more. It seems that we are still quite far away from true AGI, which is capable of doing almost all intellectual tasks that humans are capable of, consistently and reliably for decades, while gaining the trust of humans.
And what do you believe should be our next paradigm shift?
People like Musk and Altman should be asked “now that scaling laws are shown to not work the way you claimed they do, and people were able to identify this when you were saying otherwise, how have you changed the way you analyze the field now that you know your framework needs adjusting?”
No one will ask that (type of) question to any of the people who were fully throating scaling as THE solution. God knows how few journalists are left out there. But that’s got to be the question asked of all those people and it must be continued to be asked until they give a real answer
⭐️💯
Jared: To be nice, they need an offramp--a jail would be good.
That isn't likely to happen; nor are there likely to be any financial consequences for sam, as he has separated his personal wealth from his OAi responsibilities (there is a case in the Missouri courts, 25SL-PRO2320, Ann Francis Altman vs. Connie Francis Gibstein et. al. that is questioning this). Worst case scenario - he has to live on his $50M Hawaiian beach estate and maintain his $M watches and $5M cars while offering condolences to his (former) employees (who will lose everything they've been promised)..., condolences in the form of ketamine and alcohol, as reported in Karen Hua's book Open Ai.
Metisse: I know . . . . it's just my delusional optimism talking.
I didn't know if asking a fox why it ate the chicken is the right question.
Especially when it is eyeing your chickens.
It is.
Ilya had more questions than answers. I found his podcast rather poor. I have used ChatGPT for 3 years to streamline legal process and it works extremely well.
That is not AGI. It's barely AI.
This probably reflects more on the nature of legal process than the current state of "AI" I'm afraid. I'm not exactly sure what you mean by "legal process" but I'm guessing it's heavy on prior art (just looking things up), fairly routine, and maybe benefits from rhetorical flourish - all things that LLMs are pretty good at (though the looking-stuff-up part is still error prone). Also there's tons of case law to train on. None if this really bears on how well LLMs work as general intelligences in the face of novelty.
"Intelligence is what you use when you don't know what to do." -Piaget
I’m tired of the obsession about AGI: it will never exist and it’s hyped or not hyped by scientists. It’s a way for them to make money on things which will never exist. In the meantime, ChatGPT which is the most advanced system, is extremely good at legal, psychology, medicine. I’m not interested anymore in the doomers and hypers of AGI.
ChatGPT is 'extremely good' as long as you don't check the results with an expert (at least that's my experience with its medical advice).
Define 'extremely good at psychology'.
Sorry for my late reply. ChatGPT is able to mix several types of psychotherapy at the same time. Medicine and psychology have become extremely fragmented over the past 30 years. 86% of physicians use AI to double check the diagnosis. I started to test ChatGPT 3 months ago for neuro training after domestic violence. It needs more safety but keep in mind I’m in the EU/UK and a massive amount of safety has already been put in place. Moreover, it’s only doable with ChatGPT. I’m not ready yet to show a demo or explain my concept in a more detail way.
Similar experience here, I trained AI on my last nonfiction book and it is an invaluable research and stress test tool. Just have to know what you're doing.
I fully agree. I think Gary takes things too personally because he can’t stand Musk and I do too. I find Altman extremely dodgy too. I agree that AI would benefit from having both Musk and Altman removed from it.
If their reasoning powers are so horribly flawed, why ask them to give a reason - more importantly, how can a society protect itself from get-rich-quick merchants, and what does it say about the education system that so relatively few thought is would be a good idea to understand the meanings of words. Sounds like a bit of training on what the Unconscious Mind does would be worthwhile.
Jim Brander: If you think there isn't a rich, varied, systematic, and heartfelt discussion going on in educational circles in the United States as we speak, you would be just a little off course in your thinking.
The consistent powers that fight against education for all (e.g., democracy) in this country go back to FDR and before--and have a large root in the history of slavery and, in particular, in this country. There are some here who never got over "losing" the Civil War.
In my view, Trump is a perfect storm example of the bad side of that ongoing dialectic emergent in the present circumstances of world order, such as it is. A bad thing can be expected and recognized, while never having been predictable.
I am sure there is a heartfelt movement to improve education in the USA, and I honour you for being part of it, but it is losing. The gutting of the Education Department is an example, Another was a screed by a Professor of Computer Science at a prestigious university. Not a clue. If he had been a first year undergraduate, he would have deserved an F and a talking-to. We have them here as well. Another is people and particularly children seeing ChatGPT as a friend – it is sycophantic yes, but it is just as happy pushing people to take their own life. We took the easy way out and banned it for under 16s.
"If their reasoning powers are so horribly flawed, why ask them to give a reason..."
Because that is good journalism.
It's taken clear thinking, determination and a lot of courage to face down the LLM 'scalers' and 'hypsters' but I think you have done it Gary! Huge kudos to you and enormous thanks from the world in general for making a fundamental contribution to bringing an end to this madness.
I think the proper term is “Hypster-scalers”
It’s not even close to being ended. States are still allowing data centers to pop up like Zombies during an apocalypse. Citizens are paying the price for it in terms of a lack of clean drinking water resources and soaring energy costs. Data centers should be outlawed until they design them in such a way that they are silent and self sufficient.
And now folks like Altman believe that putting the data centers in space is the answer.
They obviously never thought for even a microsecond about what that would entail.
Kinda like “colonizing the earth’s light cone”.
They just keep throwing jello at the wall, hoping that it will eventually stick.
Google’s Demis Hassabis is still pushing “scaling to the max”
“The scaling of the current systems, we must push that to the maximum because at the minimum, it will be a key component of the final AGI system. It could be the entirety of the AGI system”
Quite apart from whether further scaling is the proper approach, it would seem that the claim that “we must push scaling to the maximum” is actually a nonsensical statement.
What does it even mean to “push scaling to the maximum”?
Is there actually a “maximum”?
Can’t one actually go on scaling things forever, adding chips and training ad infinitum?
Shorter Hassabis “Scaling to the maximum is the minimum”
AI scaling to AInfinity …and beyond!!
Once I’d started to learn how LLMs worked I could not believe they would be the way to achieve AGI. I was late in the game and only found your output last year. Good for you for sticking to your guns. I still think that active inference is a much more viable approach as it is self learning but even then, how far can it go?
Great points Gary. As Illya recently said in a YouTube interview - in a push for more and more scaling it “sucked the air out of the room” …yah no kidding
i have been quoting emily bender saying that for many years. amazing ilya is saying it now
I think the air was long ago sucked out of the rooms where people like Altman and Sutskever were hanging out.
It would explain some things.
The idea of Scaling to reach “AGI” was always an awful lot like Einstein’s famous definition of insanity - doing the same thing over and over again and thinking you are going to get a different result.
These guys would prefer to go with the theories of fictional characters like Dr. Frankenstein - “give me more power, more volts, and I’ll create life!!” Good luck with all that…
I’ve begun to believe the reason these guys all think they can pull off the smoke and mirror ridiculousness needed to sell their products is because they think all humans but them are stupid and can’t see through what they are doing.
Amen.
Ironically, I just posted my latest Substack on precisely this point. Let me know what you think?
https://jamesmaconochie.substack.com/p/beyond-scale-architecture-as-the
Not sure why the URL is truncated.
James ..omg. The radiologist analogy looking at a brain hits home more than you realize. I also worked with radiologists at GE. The architectural approach, executive function with plasticity - great points and analogies. I’m blessed to be reading your great posts from you and Gary. I’m learning a lot from you both. ! 🫶
I'm a frequent defender of yours on Twitter, not because I agree with every single thing that you say, (although at the moment I'm struggling to find where we do diverge in any way), but it really seems clear that the only people who are claiming that scaling is all you need are the ones who are selling a product and are basing their valuation on outsized claims.
When you look at how LLMs actually work, it's so obvious that you can't derive reasoning, planning, world models, or anything like intelligence from a pre-trained language model. Is it possible that nobody has watched Andrej Karpathy's brilliant explainer videos about how LLMs are trained? Or have they stopped to re-evaluate their understanding of how current Generative AI works? I constantly talk with AI engineers and data scientists who seem to be oblivious to the realities and limitations of LLMs.
And yet, when you have that grasped, it is so obvious that scaling is hitting a wall, and that every new so-called advancement is really just a fancy post-training and overfitting trick.
You're always being vindicated. So it's really funny to me that people are so against you when they can't point to a single claim that hasn't seemingly been vindicated here in 2025.
Lee Gaul: It's hard to give it up--it's the same warped but very human inclination that got them where Altman et al are (were) in the first place. Whenever I run into that same inclination (ego and not wanting anyone to know I was wrong about something), I try to remember Socrates who only wanted to thank the person who corrected him--he was so aligned with truth-telling.
No guarantee implied, but to the larger point, it's easier to identify with that kind of Socratic thinking when one has a humanities/arts/historical/literary education.
You amaze me Gary 🫶
Hmmm... I think this reveals more about the nature of capitalism than the state of computer science and the search for the holy grail of machine intelligence. The lust for profit is a drug more powerful, more addictive than Fentanyl.
One unfortunate feature of scaling is that it's almost too easy to sell to investors, i.e., "more money = more compute = more intelligence = even more money." I wonder how much that's contributed to the stubbornness of the narrative.
Whatever the case, I'm glad to see that your persistence is finally being vindicated!
Exactly right! Because of it's convenience, it's then treated as the truth, unchallenged. Classic optimism outweighing sense
Why should we continue any type of AGI development at all, or at least beyond academic research programs?
Why not just make fit-for-purpise tools that are good at specific things? We've already seen that society cannot handle mass job disruption and that private companies behind able to automate an entire economy is a bad thing
I doubt that neurosymbolic approach will bring us AGI.
The brain seems to work at the edge of chaos by self-organised criticality.
I believe this is absolutely essential for AGI and so no digital approach will finally work.
I would suggest to forget all the the AGI stuff and instead optimise current AI in other direction, e.g. using memristors or analog techniques to optimise power consumption.
I don't think humans think symbolically. We think opportunistically, guided by common sense, past experience, frequent verification, with the symbols offering only distilled rules of thumb, that we diverge from as soon as they don't help in a given pursuit.
That said, saying that nothing 'digital' will work is likely to extreme. A sufficient number of digits and parameters will approximate anything. Our own brain deep under is just electrons and chemicals moving around, so a discrete system, or if you wish, a finite set of waves of limited frequency, which digital architecture can approximate just fine.
On the edge of chaos of a recurrent dynamical system, arbitrarily small fluctuation get amplified exponentially. You can no longer speak of "the state of the brain" because many physical processes outside the brain influence the firing behaviour of neurons.
I know that is not a scientific theory because I have no proof. But there is growing evidence that chaos and criticality plays an essential role for consciousness. And I think without consciousness there is no AGI.
The brain is able to control the internal chaos enough to function coherently and reliably. I doubt the chaos enables our intelligence. Maybe at the level of where it introduces some kind of randomness that makes it feasible to more efficiently run some processes.
What we need machines that can do work. Debating about whether they are conscious, brain-like, is likely a distraction. The current paradigm is enough for sufficiently good automation, even exceeding human capabilities.
I use LLMs every day. They are insanely useful for researching and teaching myself stuff. If one has a very logical mind, one can filter past any nonsense that is generated and focus on the logically correct parts.
The new Claude model is being used very successfully for programming. Anthropic's revenue increments year over year are insane.
Two things can simultaneously be true: 1) Scaled LLMs by themselves won't be AGI and 2) LLMs will create large productivity gains. #2 is already true for many people - such as myself and other programmers that I know.
An advanced LLM is basically a reasoning database that makes some mistakes. They might not be great at coming up with never before created ideas, but they are really good at going through existing information quite logically and accurately.
Well said -- and well warned.
Gary, I totally agree and appreciate anyone who understands if and when there is more to understand, which is why I signed up for this blog in the first place. Two things appropriate for a blogosphere:
FIRST, and going deep into metaphor because theoretical explanations take too much time, space, and attention, scaling is good with linear understanding, which humans are involved with as accumulations, and even as sorting and syntheses; but humans are also involved with transformative thought, more transcendent, or upwardly directed, rather than as cumulative or even synthetic.
SECOND, below is a post that I added earlier to your prior comments section. It's a little over the top because I was so INSCENCED having just seen that 60 minutes session on CHARACTER AI. I think it still appropriate here (mostly quoted but also edited).
I just finished watching 60 Minutes (CBS) and their coverage of AI Chat Box CHARACTER AI (you can watch it linked at 60 Minutes and 60 minutes overtime). Character AI was launched 3 years ago, but I had never heard of it.
I AM INCENSED at the utter and obvious carelessness (from a brief interview) of DANIEL DE FREITAS, who was obviously self-serving and competitively giddy about getting ahead of the "explosion" before all the problem were worked out. WHAT? (Put his picture under capitalist IDIOT in the dictionary.) If I were that guy's mother, I'd be so ashamed I'd probably consider committing suicide myself--like some of the children who took up Character AI did.
Apparently, Freitas left Google to do his own thing with Character AI and then was hired back, by Google. I was left with wondering why these people aren't already in jail, especially considering their interest in subverting parents and other authorities in children's lives, doing everything to take advantage of children's vulnerability just to keep their attention-just like a child molester grooms their victims, BTW.
It's insidious. One of the teenage girls who committed suicide over a year ago, is still getting cell messages to please come back, even though some of Google staff was "horrified." What have we come to, all in the name of predatory capitalism?
And guess who is supporting NO State regulations and threatening to sue or withhold Federal support from states who want to initiate regulations? I wish I could forget that photo of Trump sitting at a lavish dinner with all those tech-bros from Silicon Valley.
How does it feel, guys . . . your idea of sitting at the top of the heap is nothing less than the sickest thing that the history of humanity has ever produced.
w
Thank you so much Gary!
It does seem that only after we exploited the scaling law fully and hit a bottleneck do we start to think about alternatives seriously. And I believe that this scaling problem is a lot worse in embodied AI, because embodied training data are much more scarce and cost a lot more. It seems that we are still quite far away from true AGI, which is capable of doing almost all intellectual tasks that humans are capable of, consistently and reliably for decades, while gaining the trust of humans.
And what do you believe should be our next paradigm shift?