Atlatic @_atlatic

I train AIs to rotate shapes. Joined June 2018

Tweets

471
Followers

32
Following

787
Likes

678

Atlatic @_atlatic

5 months ago

the amount of "labs" worship in the valley is gross

0 0 0 4 0

Liron Shapira @liron

2 years ago

Yoshua Bengio on @ylecun: “I don't understand his logic… I think it's dangerous not to look at the possibilities ahead.”

7 20 123 17K 22

Download Video

a lot of the magic of ChatGPT is nothing to do with AI, it’s just nice to consume high-quality internet content without ads or whacky custom formatting. the sad thing is, this isn’t a technology problem at all, it’s a business model, copyright & incentive problem. and I’m not…

55 99 1K 254K 174

Atlatic @_atlatic

2 years ago

I think writing out your thoughts very publicly, like in a book, is bad. Because it really hard-commits you to whatever you said in the book. You're now not allowed to change your mind on it, lest your legacy is ruined.

0 0 0 30 0

Atlatic @_atlatic

2 years ago

It's crazy to me that a indie video games made by a couple of dudes can make more money than movies made by 100s of very serious people

0 0 0 19 0

Atlatic @_atlatic

2 years ago

This explains his recently strong opinions on xrisk.

Andrew Ng @AndrewYNg

2 years ago

This explains his recently strong opinions on xrisk.

290 2K 7K 1.9M 4K

Download Video

0 0 0 35 0

Atlatic @_atlatic

2 years ago

academics once again prove that their thinking remains limited to publication cycles, in strict deductive chains backed by Evidence absence of evidence is taken as evidence of absence an academic fails to predict 10 years ahead even when shown the prediction from 10 years ago

0 0 0 10 0

Atlatic @_atlatic

2 years ago

Folks are still looking at current AI and making conclusions about AI in 10 years.

Halvar Flake @halvarflake

2 years ago

Folks are still looking at current AI and making conclusions about AI in 10 years.

76 368 2K 1.7M 857

0 0 0 52 0

Atlatic @_atlatic

2 years ago

watching old tv shows like star trek tng is a great way to remind yourself that men used to conduct themselves with much more honor and mutual respect, even when discussing matters with heavy disagreement picard and others actually changing mind live, feels surreal

0 0 0 13 0

Atlatic @_atlatic

2 years ago

Monarch Mixer seems promising!

0 0 0 15 0

Atlatic @_atlatic

2 years ago

random thought: in a transformer, most dot products are negative, thus don't contribute to softmax or relu/gelu/silu. For fast inference, can we not use nearest neighbor retrieval to only run dot products with the most promising rows?

0 0 1 48 0

Daniel Paleka @dpaleka

3 years ago

No one sees ChatGPT for the first time and thinks "just some n-gram correlations" or "no real knowledge inside". Those unintuitive beliefs trickle down from some experts, who should know better than to teach their controversial theories as established fact: 🧵 (1/12)

19 119 859 219K 467

Jeffrey Ladish @JeffLadish

3 years ago

Not going to lie, I feel pretty awful about the AI racing dynamics playing out right now. I think it's a really bad thing, and that AI companies should agree to slow down the growing rate of compute acquisition. I'm not saying they should stop doing research. On the other hand,…

43 64 517 116K 170

Jon Stokes @jonst0kes

3 years ago

These two goals are at odds with one another: 1. Make the AIs safer 2. Never let some large subset of public interact with them You gotta pick one.

3 5 22 2K 2

Scott Stevenson @scottastevenson

3 years ago

A path to high agency: 1. Person steps out of a highly structured environment to an unstructured environment (eg. Is left alone to write a PhD thesis, starts a company, etc) 2. Person has gut-churning anxiety for ~6 months as they realize they don’t know how to self-motivate 3…

8 19 338 69K 196

Jim Fan @DrJimFan

3 years ago

Transformers are all the rage today. But neither DALLE nor Stable Diffusion uses Transformer for image generation. Instead, they rely on a 7-year-old, Jurassic-era neural architecture. Why? 🤷🏾‍♀️⁉️ It’s finally time for Transformer and Diffusion to join forces! Quick 🧵👇:

23 232 1K 339K 648

Download Image

Yann LeCun @ylecun

3 years ago

A minimal dose of RL is inevitable. But the purpose of RL research should be to find ways to minimize its use because it's so sample inefficient. My vision is to use SSL-trained world models & intrinsic objectives (hopefully differentiable), and planning. openreview.net/forum?id=BZ5a1…

11 23 170 0 51

roon @tszzl

3 years ago

I bet what neuralink will find is that the FDA is fairly reasonable about trying new things and everyone was coping & blaming regulators for a fundamental lack of good ideas

42 18 528 0 21

Atlatic @_atlatic

3 years ago

Critiques of X-3 are very excited about the upcoming release of X-4, and want early access for the same.

0 0 0 0 0

Pete Florence @peteflorence

3 years ago

"Interactive Language: Talking to Robots in Real Time" interactive-language.github.io - Real-time, interactive, open-vocabulary, language+pixels -> actions - A new scale (~600,000 traj.) for language-conditioned behavior - Dataset, sim, models, code all to be released! (1/n)...