Viewing profile — vessenes

well don't stop there, tell us more

Oh I love this a surprising amount. Also a good reminder of some of the dividends of vibe coding -- this would have been an extremely useful tool for at least two decades of web de…

This is a tough read. I'm not sure who it's aimed at, since the article mentions only 14% of Americans (surely one of the most climate denying countries) don't believe climate chan…

Agreed that confirming “This proposition has been faithfully translated to Lean” matters. As a side note, though, this is an area where LLMs used interactively can be helpful. Also…

Public warning: please don't trade on Ed's idiotic analyses. Or if you do, look at his track record so far, and apply the Kelley Criterion to your bet sizing.

When you can formalize it in Lean or some such, why would this be? I can understand the desire to separate out other forms of research from the human corpus. But theoretical math t…

To be clear, I'm not claiming Citadel created double digit drops in SK Hynix / Samsung. I am saying that as market vol hits, vol traders might choose to make it worse. And when wor…

...Unless trench warfare your son was living through was on your mind as you wrote

This is everywhere. For reference, former FTXer and OpenAIer raised $225m into a hedge fund structure, went long and short, and reportedly peaked at $40bn of value; leverage bit ha…

Yeah there’s a mountain of interesting interp work to be done, maybe lifetimes. Looking forward to seeing what you all put out next! In very early gpt-3 beta days, I did some work …

You could publish this in the US easily. 145 IQ peeps get tons of stuff wrong, btw, and in many domains my experience is the ‘wrongness’ can intensify as you move up into higher si…

This is interesting and important work, thank you! Question - has your interp group looked at any of Anthropic’s neuralese-to-words tech? I’d be curious to see thinking traces (as …

It is a very bad thing. Your friends do not have a solid understanding of geopolitics and are putting many, many people at risk worldwide, possibly for a generation or longer. The …

I didn’t say that. I said the US is ahead, and should consider its strategic adversaries and their behavior.

I don’t think it’s clear we’ll get commoditized AI out of this research loop. The only real datapoints we have are that Chinese open weights models seem able to stay a certain amou…

… while they address absolutely none of the above points, except for tacitly or overtly agreeing a panopticon surveillance state is needed / desirable to get their desired ‘slow do…

Bioweapons are good for .. war. Agentic AI is good for changing your economy, strategic advantage, labor explosion and making more effective Agentic AI. They are not the same categ…

This. Will. Never. Work. This research game is a zero-trust iterated game. Any defection carries the benefit of possibly being permanently valuable. There will always be defections…

Seriously. I asked for an elliptic curve math course to be generated last week and Fable instantly opted out and downgraded.

I do not believe that chosen plaintext attacks are obsolete. Any time you have an encrypting oracle setup in which the scheme itself does not require some sort of modification of t…

It’s worth thinking through threat assessment and security posture carefully at the design and production phase. First, I’ll say that it’s cool you’re working on security stuff! It…

Phi was smart (or had a smart training architecture, more precisely), and arguably pushed the conversation forward globally on the value of curated training data.

It's in the constitution, probably a lot. And in the RL training. You can tell Anthropic is really worried about it. I'm interested in just how dishonest the so-called 'helpful onl…

Yeah agreed. I have on my long-term project list a hardware 'oracle' that would have everything and a local good model as a librarian/assistant and be solar powered in a pinch.

Are you referring to the META lawsuit? I think the current landscape is not bad for the frontier labs -- it's settled law that it's legal to 'read' and ingest this data. Anthropic …