Matthew Atkinson @Matkinson1864

848industries.com Washington, DC Joined August 2009

Tweets

1K
Followers

209
Following

435
Likes

1K

Matthew Atkinson @Matkinson1864

3 days ago

Interesting idea on the data-center backlash: charge a little more per compute-hour and hand that extra straight to the local community. @AnjneyMidha says he'd happily pay it. youtube.com/watch?v=h5dlIP…

0 0 0 13K 0

View Details

Matthew Atkinson @Matkinson1864

2 weeks ago

I don’t think people aren’t taking the Mythos / Fable shutdown seriously enough. If it was truly about national security, then what happens when OpenAI releases a model that’s just as capable? Does that get shut down too? If was mostly driven by politics then that’s also bad. Either way, it's not good for AI progress. businessinsider.com/reaction-to-tr…

1 0 1 3K 0

View Details

Matthew Atkinson @Matkinson1864

2 weeks ago

@natolambert That was a good piece. You should consider joining the think-tank world. We need someone as knowledgeable and articulate as you, without a dog in the fight, helping shape policy.

0 0 0 45 0

View Details

Nathan Lambert @natolambert

2 weeks ago

Recent events are so heavy bc that this feels like a start of a new tumultuous era rather than a one & done policy calibration. It's clearer we need an open ecosystem, but powerful models are coming that could cause strong reactions (or bans) with no champion to defend them.

Interconnects @interconnectsai

2 weeks ago

Welcome to the AGI era of AI governance It's a one-way door and we weren't ready for it. interconnects.ai/p/welcome-to-t…

0 1 9 20K 3

11 12 137 18K 17

View Details

Matthew Atkinson @Matkinson1864

3 weeks ago

@natolambert Do you ever think of going the think tank route?

0 0 0 42 0

View Details

Matthew Atkinson @Matkinson1864

4 weeks ago

@jxmnop Jack is on fire today!

0 0 0 844 0

View Details

Matthew Atkinson @Matkinson1864

4 weeks ago

Marc Rowan from Apollo makes a great point that so many people miss. In domains with verifiable outcomes like coding "the rate of change is a vertical line…On the other hand, if you want an answer to what is the best Shakespeare essay, we're seeing improvement, but someone has to inform. There's no right answer. What is the best Shakespeare essay? And so, we're going to see change, but not that same rate of change.” I don't think we RL our way out of this.

a16z @a16z

a month ago

Financing the Global Industrial Renaissance with Apollo CEO Marc Rowan Marc Rowan is CEO of Apollo Global Management, one of the most important financial institutions and the largest provider of retirement income in the world. In this conversation, he joins a16z's David Haber to

13 42 314 372K 519

0 0 0 96 0

View Details

Jack Morris @jxmnop

4 weeks ago

it is wild that Cursor trained a model closer to the frontier than Google with 100x fewer people and (guessing) ~100x less compute i am surprised this was even possible. also praying for the Gemini comeback ofc

Next.js @nextjs

3 months ago

Cursor's Composer 2 just took second place on the Next.js evals leaderboard, beating both Opus and Gemini. See the full rankings ↓ vercel.fyi/next-composer2

35 71 1K 245K 165

39 20 771 97K 100

View Details

Matthew Atkinson @Matkinson1864

3 months ago

@natolambert @interconnectsai Love the cover art!

0 0 0 98 0

View Details

Matthew Atkinson @Matkinson1864

3 months ago

@NASA Why the Moon on this ship and not a SpaceX ship?

0 0 5 526 0

View Details

Jack Morris @jxmnop

3 months ago

Hate to break it to you, but the first LLM was created by Andrey Markov in 1913. he tallied up 20,000 letters from a famous novel and computed p(vowel | vowel) p(consonant | vowel) p(vowel | consonant) p(consonant | consonant) basically 'training' a bigram by hand

David Pfau @pfau

3 months ago

Oh god are we really doing this? Jeff Dean trained an n-gram model on the entire internet in 2007. Jelinek coined the term "language model" in the '70s. It's called "Claude" because Claude Shannon was estimating the entropy rate of the English language in 1951!

33 82 1K 474K 353

92 440 6K 372K 2K

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@entirelyuseles @jxmnop That's right. I don't think it's something you can hill climb on like coding because how many humans are there that can truly say what is good.

0 0 3 13 0

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@sama This is a strange way for an intelligent person to do market research. It's like he doesn't understand statistics and recsys.

0 0 2 16 0

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@thekitze @karpathy Highly recommend this talk BTW

0 0 2 48 0

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@TopOfTheCircle @6blueeyesx1 @TimJHanrahan Yes. You. Can. I. Have. Done. It.

1 0 2 45 0

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@6blueeyesx1 @TimJHanrahan 100% agree.

0 0 0 90 0

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@ckstevenson @TimJHanrahan A dump truck with a plow could easily push that.

0 0 7 156 0

View Details

Matthew Atkinson @Matkinson1864

5 months ago

@VaDOTNOVA you need to push the shoulders on US 29 and US 50 in Fairfax & Falls Church. I saw a wreck this morning because only 3/4 of the lane was cleared on 29 in FC. And don't say you need "specialized equipment," because it isn't all frozen solid.