Adrien Ecoffet

Trying to make AGI go well. Researcher at @openai. Views my own.

Tweets deleted after 1y, retweets/replies after 90 days.

[Adrien Ecoffet @AdrienLE]

I want an artificial sweetener

[Acyn @Acyn]

BAIER: The president says SI. Are you into SI or AI?

BEZOS: I do like SI. Artificial is not a very flattering thing. Nobody wants an artificial sweetener

[Tom Davidson @TomDavidsonX]

@allTheYud If you thought this was what fast takeoff looked like, you also failed to convey that

Plz stop relitigating your old beef. It stinks of ego.

The ppl who followed those debates are trying to avoid ai takeover. It is ppl with far less context that oppose a slowdown

[Samuel Hammond 🦉 @hamandcheese]

Reminiscing about when people used to say dumb shit like "we already have superintelligence, they're called corporations." Yes, who could forget that fateful day in October when Dicks Sporting Goods dropped dozens of Fields Medal worthy math discoveries on the world all at once.

[roon @tszzl]

@zetalyrae @Quinurum the elderly, with zero labor power, are the most politically powerful group in the western world. their entitlement programs threaten to bankrupt great nations. it does not follow that without labor, you lose democratic power

[Adrien Ecoffet @AdrienLE]

Something very much like this underlies Knuth's intuition that P=NP, if I recall correctly

[ueaj @_ueaj]

> "sub quadratic"
> look inside
> n^1.9992

How is this even possible, I refuse to believe algorithms like this exist I don't care if there's a proof

[gfodor.id @gfodor]

So yeah that “they copied the proof from a guy’s chat logs” thing held up great

[Acer @AcerFur]

oh yeah btw guys we can do integer multiplication faster than n log n lol

I was definitely very surprised when this one came in lol

https://t.co/o7iBA131nY

[will depue @willdepue]

i asked GPT 6 Pro and Fable 5.1 to rank all discoveries in the last three years

🔵 for Human discovered
🔴 for AI discovered (before October 6th)
🟢 for AI discovered from OpenAI/math repo

81% of them have been released today. wtf

[Jeremiah Johnson 🌐 @JeremiahDJohns]

If I understand correctly:

-French govt spending is already 57% of GDP
-Despite this, students are protesting underfunded schools with poor conditions
-They're also in a huge debt crisis
-Yet both Melenchon and Le Pen are campaigning on expensive new policies

Is France cooked?

[Seán Ó hÉigeartaigh @S_OhEigeartaigh]

And adversarial position wrt OpenAI too. Seems clear even from the outside that it's pretty over the top, and has a real impact on the ability of the industry to converge around some of these important shared issues (caveat as always that a lot of the rank and file seem a lot more constructive, but leadership matters).

[Elliot Glazer @ElliotGlazer]

What, the headline doesn’t match what the article measures at all?? If OAI raised Sol’s API costs by 10x, this graph would present OAI’s “subscription value” as becoming 10x higher despite no consumer benefit lmao

[Tom Davidson @TomDavidsonX]

My read is that he was just contrasting OpenAI's open access approach with the approach that he is worried Anthropic will pursue: locking it down to prevent misuse.

Of course I agree with you that we shouldn't be blasé about the risks but it seems harsh how much criticism he is getting for this statement when I think that his intended point was bang on

[morgan — @morqon]

leaving out token efficiency kind of undermines the analysis, you’d think

[Joshua Achiam @jachiam0]

I'm excited and honored to share that I will be joining both the Institute for Progress and the Foundation for American Innovation as a non-resident senior fellow.

[roon @tszzl]

this chart should be the death knell for the 2010s tech left whose theology compelled them to argue that Uber was some sort of labor law arbitrage that created no productivity benefits. the truth is that it created a far larger market than taxis ever had. the industrialists win

[Jason Wolfe @w01fe]

I do not think we know how to answer this question scientifically. I don’t know what probability I put on the object level question but I think it’s certainly possible and i find this lack of humility and tolerance for other viewpoints to be quite distasteful.

The OpenAI Model Spec acknowledges that this is a matter of research and debate.

[Joshua Achiam @jachiam0]

David's critique is substantive and deserves careful consideration. My impression of David from our time overlapping at OpenAI is of a sober and thoughtful person who didn't approach the work from an ideological lens. This isn't someone who came in with pre-formed opinions about existential risk; he wasn't to my knowledge an EA. He was consistently productive, collaborative, and well-reasoned in his approach to safety.

The core of his critique - that professional safety engineering practices from other fields haven't taken enough root in AI safety - is substantial and correct. We're entering a phase shift where AI safety organizational practices that worked a year ago, for models of 2025 capability levels, are not sufficient to prevent serious incidents. This is a new and different world and every organization needs to uplevel accordingly.

I have much more confidence in my colleagues at OpenAI than David expresses in this article, and OpenAI does much more on this front than it gets credit for. But the actual bar for OpenAI as a whole isn't just "is it a clear field leader on alignment work, and is it successfully addressing issues as they arise?" - which is no small bar to start with - but rather, "can it earn the full trust and confidence of the public that it can safely pursue a path to superintelligence?" Because if it doesn't clear that bar, it will lose the license to operate. (Given that the public is actively debating whether to explicitly ban superintelligence and/or recursive self-improvement, I don't think this is an exaggeration.) Or, much worse, it could have a critical safety incident where the real harm is unacceptable.

That bar can only be hit by increasing the level of high-reliability safety engineering practices, embracing extreme and proactive candor around incident disclosure, and implementing third party verification that is unimpeachable from a conflicts perspective and persuasive to credible experts. (And I think it's even worth it to persuade the hostile ones.)

[Boaz Barak @boazbaraktcs]

I disagree with the framing in the title, but agree with many points @dgrobinson makes here. This is not OpenAI specific, but the entire AI industry evolved in few short years from research labs doing cool projects to creating a technology that fundamentally alters the world. The equivalent in aviation would be evolving from Wright brothers to flying billions of people. (And AGI would be like evolving from Wright brothers to commercial space travel.)

The aviation industry learned a lot of safety lessons the hard way, which was possible since the pace of transformation was sufficiently slow. In AI, we will not have that luxury.

[Adrienne LaFrance @AdrienneLaF]

New in The Atlantic:

@dgrobinson resigned this week. He was among the longest-tenured employees at OpenAI—and oversaw safety reports on 12 frontier launches.

He is very worried: “The time for trial and error is over.”

You can read his essay here:
https://t.co/DQlszrc0ru

[Dean W. Ball @deanwball]

To be quite frank, a lot (not all!) of these “rogue agent hacks” on government statistical websites are things that think-tank interns and research assistants have done for many years. I have known many a think tank paper that was enriched by an enterprising RA using eg urlquery to find CSVs and PDFs on government agency websites that were publicly accessible and non-sensitive but nonetheless not things the agency *intended* to be on their website. Is it “hacking” to find those things? Is it hacking in the even more innocuous examples where agents are literally just accessing tabular data available on a government statistical website, just a version of that data that is easier to access and analyze programmatically?

This constant trickle of examples (which tbc I have personally seen many models do, before I joined OpenAI; it is even something I have discussed with models before!) will probably have the effect of diminishing the importance of “an AI hack” on the eyes of the observing public by creating an oversupply of “AI hack” examples. OAI-HF was an example of AI hacking. Many of these more recent examples, very candidly, are straining the definition of the word “hack” and, I worry, cheapen the non-technical public’s understanding of that concept.