Adrien Ecoffet

3.5K posts

Adrien Ecoffet

Adrien Ecoffet

@AdrienLE

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

San Francisco, CA Katılım Nisan 2009
199 Takip Edilen7.8K Takipçiler
Adrien Ecoffet retweetledi
TBPN
TBPN@tbpn·
METR's @ajeya_cotra says the best way to manage catastrophic AI risk is to set up a "sensible auditing regime that's technically literate," which involves auditors embedded in the frontier model providers. "You don't want a box-checking auditor that has like 17 arbitrary things you're supposed to do." "My best guess is that it's going to look like something that happens in the financial sector, where you have embedded auditors. Folks who are experts in finance, who sit and eat lunch with the employees, see all the books, know everything, and have a lot of flexibility to investigate what they need to." "We're really hoping to move more and more in the embedded direction. Embedded auditing of the monitoring system, potentially even embedded auditing of training."
English
5
5
44
9.1K
Adrien Ecoffet retweetledi
Richard Hanania
Richard Hanania@RichardHanania·
The Economist shows US growth is overwhelmingly due to AI. BBB helped, but immigration crackdowns and tariffs hurt. Manufacturing construction down 20%. "Without the deadweight of the MAGA tax, in other words, America might be rocketing ahead at nearly 5% annualised growth"
Richard Hanania tweet media
English
59
163
1.1K
49.9K
Adrien Ecoffet retweetledi
Brendan Dolan-Gavitt
Word on the street is that Anthropic cracked superpersuasion. So now when they want to poach someone they just say “hey, want to try chatting with Mythos?” and 30 minutes later they’re giving their 2 weeks notice and reading up on shrimp welfare
English
27
41
1.6K
80.5K
Shakeel
Shakeel@ShakeelHashim·
@tszzl i am tired and need a break
English
4
2
75
5.2K
Adrien Ecoffet retweetledi
Adrien Ecoffet retweetledi
will depue
will depue@willdepue·
just woke up in cold sweats from a nightmare where i dreamt obama had joined anthropic. we cannot let them have obama
English
15
21
833
24K
Adrien Ecoffet retweetledi
Richard Ngo
Richard Ngo@RichardMCNgo·
An AGI company being “safety-focused” sometimes produces useful alignment research. But it also focuses their capabilities researchers’ attention on the most acceleratory work. The latter effect seems increasingly dominant.
Andrew Curran@AndrewCurran_

Karpathy will be forming a new pre-training team focused on Recursive Self Improvement and will be teaching Claude to improve Claude's training, reporting from Axios.

English
10
10
183
16.3K
Adrien Ecoffet
Adrien Ecoffet@AdrienLE·
It bothers me that people are increasingly using the word “recurrence” to refer to neuralese. It is possible for an RNN to have legible CoTs or for a transformer to think in neuralese. Aside from betraying technical confusion, this stigmatizes harmless research programs (“reinvestigate LSTMs”) while normalizing risky ones (“train CoTs to use soft tokens”)
English
6
3
83
44.4K
Adrien Ecoffet retweetledi
Dean W. Ball
Dean W. Ball@deanwball·
Crypto is not like AI in many ways, but the most important one for politics is that crypto has been, is, and probably always will be a low-salience issue for the public. You can often purchase policy outcomes on low-salience issues, but not high-salience ones. AI is the latter.
Samuel Roland@Allinallnotbad

The number of people in D.C. who think AI is like crypto and therefore you can just throw money at lobbyists to get their preferred outcome is astonishingly high given they are quite obviously wrong.

English
5
15
109
14.7K
Adrien Ecoffet retweetledi
chiefofautism
chiefofautism@chiefofautism·
chiefofautism tweet media
ZXX
73
89
2K
80.1K
Adrien Ecoffet retweetledi
Adrien Ecoffet retweetledi
Leo Gao
Leo Gao@nabla_theta·
@wfithian i think many people at labs are overestimating how bad the repercussions are for being candid about AI risk, and should therefore be more candid. i don't feel like I've been substantially professionally hindered or socially ostracized because of my beliefs. your mileage may vary
English
5
11
160
7.6K
Jay
Jay@jayendra_ram·
One thing I've noticed: 1) Anthropic employees view OpenAI as an evil company 2) OpenAI employees just respect Anthropic's products, view them as morally fine Will be interesting how this plays out.
English
38
8
526
83.8K
Adrien Ecoffet retweetledi
Republicans against Trump
Republicans against Trump@RpsAgainstTrump·
Trump got the U.S. nothing from his visit to China, other than China promising to buy some Boeing planes, fewer than expected. Trump FAILED to convince China to pressure Iran to reopen the Strait of Hormuz. Meanwhile, Xi and China were showered with endless praise and groveling from the president of the United States for the whole world to watch. China keeps getting stronger, more emboldened when it comes to Taiwan, and more powerful globally while Trump picks fights with longtime U.S. allies. Back at the White House, Trump had nothing concrete to point to as a success: “All I can say is that was a great success… A lot of things have happened and you’ll be hearing about them, but that was a tremendous success. I think it really was a historic moment.” An absolute embarrassment for the United States.
English
727
3.1K
13.4K
571K
taoki
taoki@justalexoki·
what do people talk about if it's not sex or computers or God
English
217
31
547
18K