Optamark @Optamark - Twitter Profile

Optamark retweeted

about 2 months ago

This 30-minute speech by the Head of Anthropic "Coding Agents" researcher will teach you more about vibe coding than 100 paid courses. Bookmark it & give it 30 minutes today. This video will change the way you use AI forever,

88

10K

999

28K

2M

Optamark retweeted

Andrej Karpathy

@karpathy

3 months ago

I had the same thought so I've been playing with it in nanochat. E.g. here's 8 agents (4 claude, 4 codex), with 1 GPU each running nanochat experiments (trying to delete logit softcap without regression). The TLDR is that it doesn't work and it's a mess... but it's still very pretty to look at :) I tried a few setups: 8 independent solo researchers, 1 chief scientist giving work to 8 junior researchers, etc. Each research program is a git branch, each scientist forks it into a feature branch, git worktrees for isolation, simple files for comms, skip Docker/VMs for simplicity atm (I find that instructions are enough to prevent interference). Research org runs in tmux window grids of interactive sessions (like Teams) so that it's pretty to look at, see their individual work, and "take over" if needed, i.e. no -p. But ok the reason it doesn't work so far is that the agents' ideas are just pretty bad out of the box, even at highest intelligence. They don't think carefully though experiment design, they run a bit non-sensical variations, they don't create strong baselines and ablate things properly, they don't carefully control for runtime or flops. (just as an example, an agent yesterday "discovered" that increasing the hidden size of the network improves the validation loss, which is a totally spurious result given that a bigger network will have a lower validation loss in the infinite data regime, but then it also trains for a lot longer, it's not clear why I had to come in to point that out). They are very good at implementing any given well-scoped and described idea but they don't creatively generate them. But the goal is that you are now programming an organization (e.g. a "research org") and its individual agents, so the "source code" is the collection of prompts, skills, tools, etc. and processes that make it up. E.g. a daily standup in the morning is now part of the "org code". And optimizing nanochat pretraining is just one of the many tasks (almost like an eval). Then - given an arbitrary task, how quickly does your research org generate progress on it?

562

9K

802

7K

2M

Optamark retweeted

Miles Deutscher

@milesdeutscher

3 months ago

I got chills when reading this article. I've never been more bullish on AI. And I've never been more terrified of what that means. It's written from the POV of June 2028. But it's long, so I summarised it for you: • AI gets good. Companies lay off workers. Margins expand. Stocks rip. S&P hits 8,000. Everyone celebrates. • But fired workers stop spending. Companies weaken. They buy more AI to cut costs. More layoffs. Less spending. A negative feedback loop with no natural brake. • The top 10% of earners drive 50%+ of all consumer spending. They're the ones getting replaced. A $180K product manager ends up driving Uber for $45K. Multiply that across every major city. • Ghost GDP emerges - the economy is "growing" on paper, but the money never reaches real people. Productivity is booming. Wages are collapsing. • Then it hits housing. $13 trillion in mortgages, all underwritten on one assumption: you keep your job for 30 years. In 2008, the loans were bad on day one. In 2028, the loans were good. The world just changed after they were written. • S&P crashes 38% from highs. Unemployment hits 10.2%. Markets barely react anymore. • The punchline: you're not reading this in 2028. You're reading it in February 2026. Every domino they describe has already started falling. The canary is still alive. Barely.

214

3K

278

3K

1M

Optamark retweeted