Jai
@Laneless_
Followers
48
Following
5K
Media
44
Statuses
643
Never had a lane to stay in anyway.
Joined February 2023
I like epistemic uncertainty language. It's great and has no downsides, and that's just a fact.
“it’s not clear to me that” “it’s not obvious that” “non trivial” all wishy washy epistemic phrases ive come to hate in san francisco
1
3
13
Contrary to the Vice President, who apparently thinks that all men are racist idiots, I think that men are generally cool, capable, intelligent, and kind.
The soft bigotry of low expectations for boys is remarkable.
0
0
2
People should start pitching radical life extension as a solution to birth rate collapse. If annual birth rates per person are low, you can still maintain demographic stability/growth by just adding more years.
0
0
3
thank god
>@kaitlancollins on CNN: "For the first in over two years ... Hamas holds no living hostages in its captivity."
5
9
386
https://t.co/GvPbnDvZz6 is a beautiful glimpse into a parallel universe where the Internet worked the way it was supposed to, where people have their own unique presence that they control, where UX is optimized for deep engagement, and js works for the user instead of advertisers
gwern.net
Personal website of Gwern Branwen (writer, self-experimenter, and programmer): topics: psychology, statistics, technology, deep learning, anime. This index page is a categorized list of Gwern.net...
0
0
3
OpenAI to purchase $80 billion of Multiplication Boxes from Nvidia.
We should probably consider renaming GPUs at this point.
0
0
1
Everyone who isn't OpenAI or its investors: OpenAI should not be allowed to convert to a for-profit Encode: We agree OpenAI: hm it seems that Encode has taken the side specifically of Elon Musk and no one else. Very suspicious of them
There’s quite a lot more to the story than this. As everyone knows, we are actively defending against Elon in a lawsuit where he is trying to damage OpenAI for his own financial benefit. Encode, the organization for which @_NathanCalvin serves as the General Counsel, was one
13
41
734
When you see examples of bad behavior, like silencing critics and preventing internal and external dissent, consider: this is likely just the tip of the iceberg. It's notable that OpenAI successfully silenced dozens if not hundreds of ex-employees with financially punitive
4
12
128
People often have the mistaken impression that OpenAI spends most of their compute on inference. But actually less than half of its compute infra last year went into serving customers.
New data insight: How does OpenAI allocate its compute? OpenAI spent ~$7 billion on compute last year. Most of this went to R&D, meaning all research, experiments, and training. Only a minority of this R&D compute went to the final training runs of released models.
14
11
194
This work is exciting because it shows we might be able to steer how models generalize from our SFT demonstrations. What’d be even more exciting is showing we can steer how models generalize from their RL trajectories!
New paper! Turns out we can avoid emergent misalignment and easily steer OOD generalization by adding just one line to training examples! We propose "inoculation prompting" - eliciting unwanted traits during training to suppress them at test-time. 🧵
2
1
15
Things that were once considered inevitable by very large groups of people: - Rule by kings - Almost every family losing a child to disease or hunger - War as the main way to settle disputes between countries - Overpopulation leading to mass starvation - Widespread
9
46
366
There was a time when I was more open to saying "Oh obviously this makes sense, rampant sexism etc." and I still obviously believe that, but I've come to worry the risks of just universal low trust, when it's caused by memes and short form video, are just objectively pretty bad.
6
3
187
1. The GOP is capable of standing up to Trump 2. This should be encouraged
Worth noting: Jimmy Kimmel has now been back on the air for about two weeks. Brendan Carr said not long ago that if Kimmel stayed on the air, he would pull broadcast licenses of ABC affiliates. So... where’s Carr? He seems to have backed down after pressure from Ted Cruz and
0
0
2
Being this kind of buzzkill is my thing.
I’ve become the jerk who, when someone makes a surprising claim that implies two things are strongly correlated, pulls up the correlations in real time and checks. (Answer: Usually, no.) Here are a million free human correlations so you too can be jerk: https://t.co/kxoE3GbOBX
0
0
2
For years I've wanted to write a piece about self as an important neglected tractable cause. For any given person: you matter and are uniquely situated to improve well-being in the corner of the universe that is you.
@TetraspaceWest everyone's life is but usually they just go unsolved forever
0
0
2
The truest EA experience is trying to do the most good in a way that somehow gets everyone really pissed at you.
One of the more entertaining parts of running EA DC is being in spaces where I get accused of being an anti-AI luddite communist then going across the street to a space where I get accused of being a tech bro AI worshipper libertarian
1
0
5
CUDA parity when?
Excited to partner with AMD to use their chips to serve our users! This is all incremental to our work with NVIDIA (and we plan to increase our NVIDIA purchasing over time). The world needs much more compute...
0
0
2
Unironically yes, you need similar skills to do both well. You need strong communication skills, a grasp of their strengths and weaknesses, you need to give them the right tools and access, and you need well defined verifiable success criteria.
1
0
6
The default state of nature is poverty, and we've been clawing our way out of it for the last 200 years
Capitalism is unnatural. Capitalism created poverty. Poverty is manmade.
183
934
10K
Evidence Recent AI Gains are Mostly from Inference-Scaling 🧵 Here's a thread about my latest post on AI scaling... 1/14
5
13
87