Jacques
@JacquesThibs
Followers
4K
Following
21K
Media
2K
Statuses
14K
AI risk reducer, alignment-accelerationist working to create a future where humanity can flourish even more. 🇨🇦 Building something new.
London, UK
Joined May 2008
From a Paul Christiano talk. Assume we're in a world where AI systems are broadly deployed and the world has become increasingly complex, where humans know less and less how things work. A viable strategy for AI takeover is to wait until there is certainty of success. If a 'bad.
11
4
69
Wow, holy shit this is far away from anything else I’ve seen. Insane. Google basically solved AI music:
136
594
3K
lmao. "We achieved significantly better results on the same evaluation by adding the sentence “Here is the most relevant sentence in the context:” to the start of Claude’s response. This was enough to raise Claude 2.1’s score from 27% to 98% on the original evaluation." The fix:
Claude 2.1’s 200K token context window is powerful, but requires careful prompting to use effectively. Learn how to get Claude to recall an individual sentence across long documents with high fidelity:.
20
127
1K
i’m on a date and she says “the world is better off without humans”, i am hit with a deep sadness.
143
11
994
@ryancbriggs That’s an interesting point. In the more extreme case, we might also be the only generation that knows how to code too.
22
5
968
@MarkovMagnifico He did it 10 times in a row and then eventually realized he couldn’t stop otherwise his wife would think he doesn’t love her anymore.
5
11
820
This was the first date in a while where I was actually nervous 😂
24
10
771
Huh: . “Murati and Brockman told Ilya that the company was in disarray and might collapse without him. They visited his home, bringing him cards/letters from employees urging him to return. Altman visited him and expressed regret that others at OpenAI hadn’t found a solution.”
11
25
708
@GiggukAZ The fact that they cover the Arlong arc makes me pretty excited about the pacing. Seems like they will tighten up the pacing enough to make it a better introduction to people who are new to One Piece!.
2
2
673
“Altman began approaching other board members, individually, about replacing her. When these members compared notes about the conversations, some felt that Altman had misrepresented them as supporting Toner's removal.”. "Things like that had been happening for years.".
Probably not worth the drama but still, this isn't a great look
19
59
655
Wow. This is an impressive amount of character from a former OpenAI employee who recently left the company “due to losing confidence that it would behave responsibly around the time of AGI.” He gave up a lot of money to retain his ability to criticize the company in the future.
26
54
564
Multiple comments by the same person in that thread; here are a few more:
5
20
451
@JoestarJokic The crazy thing about this is that Oda literally invented all the supernovas the week before they were introduced into the story:
5
10
467
@RichardHanania For the Trump cult:. 1 white lie from a democrat presenting as sincere. is worse than. 99 egregious lies from Trump, the guy who 'tells it like it is'.
56
7
451
@lulumeservey Because he's treating us as if we're stupid enough not to notice that the bots were all teleoperated and voiced by humans. Setting high surface-level expectations while doing typical behind-the-scenes startup bs. I still think these are decent feats, just don't try to trick us.
7
2
371
Maybe not well-known but:. Dario Amodei does have a PhD in (Bio)physics at Princeton, with a focus on electrophysiology of neural circuits. And post-doc at Stanford School of Medicine, where he did more biophysics (while contributing to open source proteomics software).
Guy who has never done biology.
8
17
382
She’s objectively correct, but I do find the psychology of people being pissed off by this kind of interesting. The empty space must be filled! I have to move forward to feel like the line is making progress!.
53
16
370
Sharing a long, but insightful comment @Gwern made in response to the question: "EA had a pretty weak hand throughout and played it as well as can be reasonably expected"?. Gwern: It was a pretty weak hand. There is this pervasive attitude that Sam Altman could have been.
10
41
338
Uhh @EpochAIResearch?. Could we get an explanation on what’s the full situation here? Cc @Jsevillamol
11
14
324
@andletmejustsay “i mean, i wouldn’t because i don’t seem him like that…but he could definitely date any other woman if he wanted”.
0
0
292
@DynamoSuperX Love that Kimiko was showing him books to communicate because he dropped her phone while sneaking in.
2
0
283
Tremendous amount of alpha in looking into the recent papers and background of new OpenAI, Anthropic and DeepMind employees before they got hired.
7
11
279
On “humanity’s last exam”
the company with agi as a little pet project is absolutely cooking your serious agi company.
5
17
284
If AI accelerationists fail to address the loss of jobs from AI, they may feel the wrath of the masses in a way they fail to comprehend as techno-optimists. The acceleration and lack of care for job loss may ironically backfire into becoming the most extreme deceleration.
60
29
258
@meltborne I just want Firecracker to unlock her actual explosive power by the end of the season, which forces Annie to get her shit together so we can get a full supe fight between these two.
1
1
241
This is what would happen if Anthropic actually dropped Golden Path Claude:
1
4
251
I’ve heard the same (~“most papers are bad; very behind actual SOTA”) from someone at a top lab @dwarkesh_sp . @tszzl has said similar things. GPT-4 finished training *1.5 years ago*. What do we expect world-class researchers at OAI have done since then, twiddled their thumbs?
12
15
243
@mijuohdele Definition of “everyone thinks they are the good guy, no matter how objectively bad they are.”.
0
1
234
An alarm clock, but instead of a ring sound to wake you up, it’s @robertwiblin saying, “what are you working on at the moment, and why do you think it’s important?”.
5
10
223
I wonder what will be the reaction from the mathematics community once AI solves the Riemann hypothesis.
24
6
211
At least we’ll all be paperclipped with a USA flag engraved on it.
9
12
220
Deepseek showing OpenAI they could have survived on a non-profit budget.
7
10
208
Karpathy finished building “JARVIS” at OpenAI and that’s what we’re getting tomorrow. His work was done and he left.
5
8
198
Here's an accompanying poll, but take it with a grain of salt (or less), given that anyone can click whichever option:
6
5
189
btw, I'm +50% sure that human-level text-to-speech is coming within the next year or two.
19
4
186
@mckaywrigley I sent this to someone, and she said, "I used to think this would be impressive, but now that I'm in a PhD with other PhD students, I'm not so impressed by this benchmark." lol.
2
0
192
When the Effective Altruism community has conversations about the best time to donate their money:.
3
19
187
We knew this was bound to happen
This is some of the darkest predictions of AI in warfare, coming true in 2024.
7
40
185
Really interesting to hear Sam's behaviour described as paper cuts in another setting (his previous company, Loopt). Does it sound strangely similar to "not being consistently candid"?
13
11
181
This ‘debate’ is essentially ethical theatre. Pretending as if this debate will impact the natural momentum and inevitable outcome of AI being used autonomously in war.
15
10
181
China told its companies, “solve alignment or we shut you down” and now we’re getting a bunch of ‘aligning language models’ papers from Chinese companies 😂.
Reasons to Reject? Aligning Language Models with Judgments. paper page: As humans, we consistently engage in interactions with our peers and receive feedback in the form of natural language. This language feedback allows us to reflect on our actions,
13
17
168
@benlandautaylor I mean, they all have the potential of losing millions in stocks so the incentives are overwhelmingly in favour of signing in that respect.
4
0
174
One step closer to grey goo. “the researchers specifically call out the fact that these machines can ‘self-replicate its entire 3D structure and functions.’”
7
14
167
Actually, you know what, it's probably Omni, Orion, and Operator. Orion also have 3 stars for "Orion's belt" so that's a clue to consider. Omni: includes more input/output types to 4o or something.Orion: Next big model after GPT-4.Operator: Computer use agent
101
46
120
People don’t even know what actually happened at OpenAI yet are attacking Toner and discrediting how smart, principled and experienced she is. Truly embarrassing behaviour.
10
4
161
This is one of the most frustrating threads I’ve read. Sometimes, I wonder if rationalists really believe in x-risks or just really like being in esoteric discussions that use flowery language that could never be distilled into human-speak for us mere peasants.
@ESYudkowsky I can’t speak to the details of those experiments but I at least read a much higher fraction of your paper outputs than your blog post outputs. Possibly I’m a minority here but I am certainly not the only one.
17
8
160
@vampy_gyaru @netflix Alabasta is also heavily influenced by India, not just the middle east. .
4
0
148
Worth noting that:
@JacquesThibs There's no way to confirm if this is a current employee or previous disgruntled employee. Blind doesn't force reconfirm emails.
1
3
147
In my opinion: . As soon as EAs started mentioning that we might be in a dangerous time due to nuclear war, biorisks, and AI risks, Pinker did a 180 on EA because, to him, it disagreed with the thesis of his book "The Better Angels of Our Nature: Why Violence Has Declined".
I was a fan of Effective Altruism (almost taught a course on it at Harvard) together w other rational efforts (evidence-based medicine, data-driven policing, randomista econ). But it became cultish. Happy to donate to save the most lives in Africa, but not to pay techies to fret.
9
4
150
@gurgavin In basketball we have a saying, “if you don’t want to get called on fouls, stop fouling!”. All Trump needs to do is not lie and babble about things that can be fact-checked.
5
4
140
This thread might convey an opinion I don't have; let me clarify: I feel horrible for OpenAI employees. You don't deserve this. I believe that most of you care deeply about the company and its mission. And I also don't blame people if they are *also* upset for financial reasons.
2
5
135
@KaiyoBreeze They want him to have 5 CTs, a matured domain expansion and god-tier RCT at least 100 chapters ago.
2
2
141
That new Eliezer podcast was pretty heartbreaking in some parts, not gonna lie. When he said, “I did all my crying late at night in 2015”, I was like damn….
13
4
135
This is exciting! From the author of, "Significantly Enhancing Adult Intelligence With Gene Editing May Be Possible":
6
11
131
@gbrl_dick instead read twitter engagement bait takes by tech nerds with cartoon profile pics who have not touched grass once in the last 3 years.
1
4
122
Meme concocted with my friends in @AiEleuther (discord channel) off-topic, the pinnacle of AI discussion. (image in OP by @BjarturTomas).
5
1
132
EA donations to global poverty/health over time (through givewell):
6
15
127
@Noahpinion You misread the tweet. We all make mistakes. Don’t double down; make the mature decision and delete your tweet. Even better would be to apologize for misrepresenting her.
2
0
125
@yacineMTB Rookie mistake, it’s actually 1 billion dollars because you could use that money to invest in yourself and sell your company for a billion.
3
1
125