I know that what I described isn't what Gwern or anyone else who promotes nicotine is recommending. I just wanted to share an opposite experience from OP's as a warning and another reminder to be careful if you're dabbling with nicotine.

I was certainly reckless, so if you're smarter and have better self-control than me, then maybe you'll be fine. On the other hand, I used nicotine occasionally for years without becoming addicted and feeling like I wasn't at risk. Then I started ramping up slowly--I would use mild amounts for a few days or weeks, but I could always quit when I wanted to without much effort. It wasn't until recently that I became so addicted that I truly struggled to quit and it started to have major negative effects.

So my main points are: even if you don't become addicted at first, that doesn't mean you won't down the line. And if you have an addictive personality or other mental health issues like me, I would definitely stay away.

I feel the need to share my own negative experience with nicotine as a counterpoint. I want everyone reading this to tread extremely carefully and I personally don't think that the potential for a small cognitive boost is worth the risk of addiction.

I also have a somewhat addictive personality and have struggled with other addictions in the past. Like you, for a while I thought that for some reason, I wasn't susceptible to nicotine addiction. At first, I didn't use nicotine for cognitive or productivity purposes. I used it in social settings; if someone offered me a vape or cigarette I would have some, enjoy the rush, and never felt the need to continue using.

Eventually, I started buying disposable vapes every once in a while for fun. But I liked them so much that I would end up hitting them in rapid succession to the point that I felt lightheaded and nauseous, then I would throw them out and not buy another. Once again, I felt like I had avoided addiction.

Then, my friends started using Zyn (nicotine pouches). I would again use them socially if offered. I don't know how much these pouches differ from lozenges (I've never had a lozenge) but I think they are the absolute worst delivery mechanism. For me at least, the steady supply of nicotine is much worse than the quick hit from a cigarette or vape. Pouches make me way more anxious when I use them, yet I find them much harder to stop using.

Even still, it took me about a year of using them casually in social settings before I got addicted and started buying my own. And the reason that I got addicted is that I started using them for a productivity/cognitive boost. Specifically, I was having trouble sleeping for a while, and I realized that nicotine was incredibly helpful for staying focused and productive at work after a bad night's sleep.

This started a cycle where I would become addicted for several weeks, quit for several weeks, then find myself craving an extra boost and restarting the cycle. This continued for almost 3 years. Then 6 months ago, I started a new job, and I started using the pouches even heavier than before as I felt the need for even more productivity and focus to get up to speed.

This has led to my worst period of addiction yet, where I've basically been using all day every day for the past 6 months. This has ended up destroying my productivity and focus in the long run, but it's incredibly hard to stop because I become basically incapable of doing anything if I stop using. A few months ago, I quit for a week, and I had to do it when I was off work and I was able to just lie around doing nothing to get through the hellish withdrawals. Still, I started back up again the day I went back to work.

My mental health is awful right now; I'm anxious and agitated all day, I'm anhedonic, I'm sleeping poorly, I'm having trouble eating or exercising, I'm less productive at work than I would be otherwise, and I have no motivation to do anything outside of work. I have another week off soon and I'm going to try to quit again, but I'm not very hopeful that it will stick.

TL;DR, tread extremely carefully with nicotine!

This is an interesting model, and I know you acknowledged that progress could take years, but my impression is that this would be even more difficult than you're implying.  Here are the problems I see, and I apologize in advance if this doesn't all make sense as I am a non-technical newb.

• Wouldn’t it take insane amounts of compute to process all of this?  LLM + CoT already uses a lot of compute (see: o3 solving ARC puzzles for $1mil).  Combining this with processing images/screenshots/video/audio, plus using tokens for incorporating saved episodic memories into working memory, plus tokens for the decision-making (basal ganglia) module = a lot of tokens. Can this all fit into a context window and be processed with the amount of compute that will be available?  Even if one extremely expensive system could run this, could you have millions of agents running this system for long periods of time?
• How do you train this?  LLMs are superhuman at language processing due to training on billions of pieces of text. How do you train an agent similarly? We don’t have billions of examples of a system like this being used to achieve goals.  I don’t think we have any examples.  You could put together a system like this today, but it would be bad (see: Claude playing Pokemon).  How does it improve?  I think it would have to actually carry out tasks and RL on them.  In order for it to improve on long-horizon tasks, it would take long-horizon timeframes to get reinforcement signals.  You could run simulations, but will they come anywhere close to matching the complexity of the real world?  And then there’s the issue of scalable RL only working for tasks with a defined goal: how would it improve on open-ended problems?
• If an LLM is at the core of the system, do hallucinations from the LLM “poison the well” so to speak?  You can give it tools, but if the LLM at the core doesn’t know what’s true or false, how does it effectively use them?  I’ve seen examples like: an LLM got a math problem wrong, then the user asked it to solve using a python script.  The python script produced the correct answer, but then the LLM just repeated the wrong answer and added a fake floating point error that it said came from the python script.  So it seems like hallucinated errors from the LLM could derail the whole system.  I suppose that being trained and RL’d on solving tasks would eventually lead it to learning what works from what doesn’t, but I guess I’m just pointing out why adding additional modules and tools doesn’t automatically solve the issues of LLMs.