she/her ⋅ LGBTESCREAL+ ⋅ reading committee @aft_technoprog ⋅ my xeets reflect the views of every person-moment I'm R-related to ⋅ 2ndary account: @MLaGrangienne
Joined October 2021
- Tweets66.4K
- Following2.5K
- Followers1.4K
- Likes78.8K
Pinned Tweet
Hi everyone! This was my first Twitter account, it got banned and I switched to @MLaGrangienne before I got the admins to unban it. I'm now making this my public-facing English account. @MLaGrangienne remain my personal and French account.
damn @ohabryka really renting out the Lighthaven recording studio to anyone these days
CENTCOM is now focused on producing an official podcast. Here's an exchange from the debut episode:
FLEET MASTER CHIEF COMPTON: You are the original fry guy. You like French fries. Most people don't know this, but I did notice when we stationed together years ago, my lunch would come and I always had French fries... You at least take one or two... I don't think you ever knew that I knew you took my French fries. So is that your favorite food? French fries?
ADMIRAL COOPER: We should let the world know. For years, I've been eating your French fries. The secret is now out. But yeah, I never met a French fry I didn't like.
Are we confident they're going to pro bono disclose *every* zero-day their expensive 'frontier cyber model' find?
We are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.
In one case, we studied the Linux kernel, which is a particularly load-bearing piece of software at OpenAI.
Before AI, security researchers might discover ~20 local privilege escalation (LPE) vulnerabilities in Linux in a typical year.
In one week's time, our upcoming frontier cyber model was able to find ~1000 LPE vulnerabilities using about $1M of tokens. In the process, it was able to discover at least 10 novel classes of Linux exploitation techniques.
6
241
Underdiscussed dynamic: AI company internal politics make no sense by normal company internal politics standards, like what tf do you mean your job is "lobbyist, but the only people you're lobbying are the other lobbyists in the same company"
Embedded agency issues imply the funniest possible answer to arguments about qualia: actually your body is inhabitated by as many functionally equivalent minds as there are possible bijections between qualia and relevant physical states (cc @jessi_cata)
It's certainly fortunate that the only country which has ever tried to embed a spy into OpenAI is one with an infamously incompetent intelligence service and immediately got caught (@_NathanCalvin tweet related)
Authorities say the wife of Alfozan, the OpenAI employee who allegedly bugged an OpenAI exec conference room, was to leave the U.S. on one flight, then a "member of the Saudi military booked and paid" for another one-way flight to Saudi Arabia for the wife and son.
27
547
Joe Carlsmith talks about this
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
1
5
256
NASA R&D on the logistics of self-sustained life environments and their maximum viable population, or, as I like to call it, "prosaic overpopulation on Mars research"
1
134
Would be interested in opposing arguments, but it's my impression currently that DeepMind being founded lengthened AI timelines, by sucking Google money into (in hindsight) pointless distractions like AlphaGo over the scaling and language modeling work Google Brain was doing.
In the world where Hinton (RBM pretrain), Ian Buck (CUDA), Hassabis (Deepmind acquisition starting gold rush), and Vaswani and Shazeer (transformers) were all never born, the deep learning revolution plays out very similarly on a delay of maybe 3-6 years??, I'd guess.
1
17
1,106
What was the post going around in cyborgist circles talking about how and why we should expect different RL algorithms to have different robopsychological consequences? I think it was by @FioraStarlight but I can't find it.
"one of the smartest, insightful takes I've heard coming out of The Curve" and it was already dunked on by @allTheYud in the Sequences in the 2000s
“We need a Manhattan Project for AI Safety” could be one of the smartest, insightful takes I’ve heard coming out of The Curve. More thinking like this, please.
The @americans4ri policy department is going to start thinking about this, like, today!
H/T to @adamconner
5
168
oh no
Margaret Hamilton, who led the 100-person team that wrote the onboard software for the Apollo missions, has died. nyti.ms/3U9IrTj
7
339
didn't knew the SNCF was doing that kind of train too
ML4 was trained on 3,800 NVIDIA Grace Blackwell GPUs in our European datacenters -- our cluster in Bruyères-le-Châtel built using our Series B fundraise.
We are investing heavily in infrastructure, with our Series C and D clusters coming online soon. This additional compute will let us train longer, pursue more ambitious post-training, and iterate faster.
We expect large and rapid improvements in the weeks and months ahead.
2/n
1
10
302