Not learning from new input may be a feature. Back in 2016 Microsoft launched one that did, and after one day of talking on Twitter it sounded like 4chan.[1] If all input is believed equally, there's a problem.
Today's locked-down pre-trained models at least have some consistency.
Incredible to accomplish that in a day - it took the rest of the world another decade to make Twitter sound like 4chan, but thanks to Elon we got there in the end.
Has anyone tried implementing something like System M's meta-control switching in practice? Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.
> Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.
If you like biomimetic approaches to computer science, there's evidence that we want something besides neural networks. Whether we call such secondary systems emotions, hormones, or whatnot doesn't really matter much if the dynamics are useful. It seems at least possible that studying alignment-related topics is going to get us closer than any perspective that's purely focused on learning. Coincidentally quanta is on some related topics today: https://www.quantamagazine.org/once-thought-to-support-neuro...
The question is does this eventually lead us back to genetic programming and can we adequately avoid the problems of over-fitting to specific hardware that tended to crop up in the past?
Or possibly “in addition to”, yeah. I think this is where it needs to go. We can’t keep training HUGE neural networks every 3 months and throw out all the work we did and the billions of dollars in gear and training just to use another model a few months.
That loops is unsustainable. Active learning needs to be discovered / created.
"he proposed framework integrates learning from observation (System A) and learning from active behavior (System B) while flexibly switching between these learning modes as a function of internally generated meta-control signals (System M). We discuss how this could be built by taking inspiration on how organisms adapt to real-world, dynamic environments across evolutionary and developmental timescales. "
If this was done well in a way that was productive for corporate work, I suspect the AI would engage in Machievelian maneuvering and deception that would make typical sociopathic CEOs look like Mister Rogers in comparison. And I'm not sure our legal and social structures have the capacity to absorb that without very very bad things happening.
Not just CEOs, Legal and social structures will also be run by AI. Chimps with 3 inch brains cant handle the level of complexity global systems are currently producing.
> If this was done well in a way that was productive for corporate work, I suspect the AI would engage in Machievelian maneuvering and deception that would make typical sociopathic CEOs look like Mister Rogers in comparison.
Algorithms do not possess ethics nor morality[0] and therefore cannot engage in Machiavellianism[1]. At best, algorithms can simulate same as pioneered by ELIZA[2], from which the ELIZA effect[3] could be argued as being one of the best known forms of anthropomorphism.
>As Weizenbaum later wrote, "I had not realized ... that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people."...
That pretty much explain the AI Hysteria that we observe today.
ELIZA couldn't write working code from an English-language prompt though.
I think the "AI Hysteria" comes more from current LLMs being actually good at replacing a lot of activity that coders are used to doing regularly. I wonder what Weizenbaum would think of Claude or ChatGPT.
>ELIZA couldn't write working code from an English-language prompt though.
Yea, that is kind of the point. Even such a system could trick people into delusional thinking.
> actually good at replacing a lot of activity that coders are used to...
I think even that is unrealistic. But that is not what I was thinking. I was thinking when people say that current LLMs will go on improving and reach some kind of real human like intelligence. And ELIZA effect provides a prefect explanation for this.
It is very curious that this effect is the perfect thing for scamming investors who are typically bought into such claims, but under ELIZA effect with this, they will do 10x or 100x investment....
Agents playing the iterated prisoner's dilemma learn to cooperate. It's usually not a dominant strategy to be entirely sociopathic when other players are involved.
You don't get that many iterations in the real world though, and if one of your first iterations is particularly bad you don't get any more iterations.
They still fail in the real world, where a single failure can be highly consequential. AI coding is lucky it has early failure modes, pretty low consequence. But I don't see how that looks for an autonomous management agent with arbitrary metrics as goals.
Anyone doing AI coding can tell you once an agent gets on the wrong path, it can get very confused and is usually irrecoverable. What does that look like in other contexts? Is restarting the process from scratch even possible in other types of work, or is that unique to only some kinds of work?
The paper's critique of the 'data wall' and language-centrism is spot on. We’ve been treating AI training like an assembly line where the machine is passive, and then we wonder why it fails in non-stationary environments. It’s the ultimate 'padded room' architecture: the model is isolated from reality and relies on human-curated data to even function.
The proposed System M (Meta-control) is a nice theoretical fix, but the implementation is where the wheels usually come off. Integrating observation (A) and action (B) sounds great until the agent starts hallucinating its own feedback loops. Unless we can move away from this 'outsourced learning' where humans have to fix every domain mismatch, we're just building increasingly expensive parrots. I’m skeptical if 'bilevel optimization' is enough to bridge that gap or if we’re just adding another layer of complexity to a fundamentally limited transformer architecture.
In this podcast episode[0] he does talk about this kind of model and how it "learns about physics" through experience instead of just ingesting theorical material.
Not learning from new input may be a feature. Back in 2016 Microsoft launched one that did, and after one day of talking on Twitter it sounded like 4chan.[1] If all input is believed equally, there's a problem.
Today's locked-down pre-trained models at least have some consistency.
[1] https://www.bbc.com/news/technology-35890188
Incredible to accomplish that in a day - it took the rest of the world another decade to make Twitter sound like 4chan, but thanks to Elon we got there in the end.
Has anyone tried implementing something like System M's meta-control switching in practice? Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.
> Curious how you'd handle the reward signal for deciding when to switch between observation and active exploration without it collapsing into one mode.
If you like biomimetic approaches to computer science, there's evidence that we want something besides neural networks. Whether we call such secondary systems emotions, hormones, or whatnot doesn't really matter much if the dynamics are useful. It seems at least possible that studying alignment-related topics is going to get us closer than any perspective that's purely focused on learning. Coincidentally quanta is on some related topics today: https://www.quantamagazine.org/once-thought-to-support-neuro...
The question is does this eventually lead us back to genetic programming and can we adequately avoid the problems of over-fitting to specific hardware that tended to crop up in the past?
Or possibly “in addition to”, yeah. I think this is where it needs to go. We can’t keep training HUGE neural networks every 3 months and throw out all the work we did and the billions of dollars in gear and training just to use another model a few months.
That loops is unsustainable. Active learning needs to be discovered / created.
"don't learn" might be a good feature from a business point of view
Imagine if AI learns all your source code and apply them to your competitor /facepalm
by Emmanuel Dupoux, Yann LeCun, Jitendra Malik
"he proposed framework integrates learning from observation (System A) and learning from active behavior (System B) while flexibly switching between these learning modes as a function of internally generated meta-control signals (System M). We discuss how this could be built by taking inspiration on how organisms adapt to real-world, dynamic environments across evolutionary and developmental timescales. "
https://github.com/plastic-labs/honcho has the idea of one sided observations for RAG.
If this was done well in a way that was productive for corporate work, I suspect the AI would engage in Machievelian maneuvering and deception that would make typical sociopathic CEOs look like Mister Rogers in comparison. And I'm not sure our legal and social structures have the capacity to absorb that without very very bad things happening.
Not just CEOs, Legal and social structures will also be run by AI. Chimps with 3 inch brains cant handle the level of complexity global systems are currently producing.
> If this was done well in a way that was productive for corporate work, I suspect the AI would engage in Machievelian maneuvering and deception that would make typical sociopathic CEOs look like Mister Rogers in comparison.
Algorithms do not possess ethics nor morality[0] and therefore cannot engage in Machiavellianism[1]. At best, algorithms can simulate same as pioneered by ELIZA[2], from which the ELIZA effect[3] could be argued as being one of the best known forms of anthropomorphism.
0 - https://www.psychologytoday.com/us/basics/ethics-and-moralit...
1 - https://en.wikipedia.org/wiki/Machiavellianism_(psychology)
2 - https://en.wikipedia.org/wiki/ELIZA
3 - https://en.wikipedia.org/wiki/ELIZA_effect
https://en.wikipedia.org/wiki/ELIZA_effect
>As Weizenbaum later wrote, "I had not realized ... that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people."...
That pretty much explain the AI Hysteria that we observe today.
ELIZA couldn't write working code from an English-language prompt though.
I think the "AI Hysteria" comes more from current LLMs being actually good at replacing a lot of activity that coders are used to doing regularly. I wonder what Weizenbaum would think of Claude or ChatGPT.
>ELIZA couldn't write working code from an English-language prompt though.
Yea, that is kind of the point. Even such a system could trick people into delusional thinking.
> actually good at replacing a lot of activity that coders are used to...
I think even that is unrealistic. But that is not what I was thinking. I was thinking when people say that current LLMs will go on improving and reach some kind of real human like intelligence. And ELIZA effect provides a prefect explanation for this.
It is very curious that this effect is the perfect thing for scamming investors who are typically bought into such claims, but under ELIZA effect with this, they will do 10x or 100x investment....
Agents playing the iterated prisoner's dilemma learn to cooperate. It's usually not a dominant strategy to be entirely sociopathic when other players are involved.
You don't get that many iterations in the real world though, and if one of your first iterations is particularly bad you don't get any more iterations.
But AI will train in the artificial world
They still fail in the real world, where a single failure can be highly consequential. AI coding is lucky it has early failure modes, pretty low consequence. But I don't see how that looks for an autonomous management agent with arbitrary metrics as goals.
Anyone doing AI coding can tell you once an agent gets on the wrong path, it can get very confused and is usually irrecoverable. What does that look like in other contexts? Is restarting the process from scratch even possible in other types of work, or is that unique to only some kinds of work?
The paper's critique of the 'data wall' and language-centrism is spot on. We’ve been treating AI training like an assembly line where the machine is passive, and then we wonder why it fails in non-stationary environments. It’s the ultimate 'padded room' architecture: the model is isolated from reality and relies on human-curated data to even function.
The proposed System M (Meta-control) is a nice theoretical fix, but the implementation is where the wheels usually come off. Integrating observation (A) and action (B) sounds great until the agent starts hallucinating its own feedback loops. Unless we can move away from this 'outsourced learning' where humans have to fix every domain mismatch, we're just building increasingly expensive parrots. I’m skeptical if 'bilevel optimization' is enough to bridge that gap or if we’re just adding another layer of complexity to a fundamentally limited transformer architecture.
LeCun has been talking about his JEPA models for awhile.
https://ai.meta.com/blog/yann-lecun-ai-model-i-jepa/
In this podcast episode[0] he does talk about this kind of model and how it "learns about physics" through experience instead of just ingesting theorical material.
It's quite eye opening.
0. https://youtu.be/qvNCVYkHKfg
We are rediscovering Cybernetics
Biological Computer Laboratory (1958-1976), https://web.archive.org/web/20190829234412/http://bcl.ece.il...
It's striking how cybernetics has gone from dated to timely.
Can I run it?