virtual @virtual_jpeg

contributing @harnesslabs ℤ Joined July 2021

Tweets

2K
Followers

944
Following

2K
Likes

7K

𝔧𝔢𝔭𝔰𝔢𝔫 @0xjepsen

11 months ago

Seeing this evolve over time from some of the early simulation efforts is inspiring and excited. Amazing go getters and builders behind this.

Anthias Labs @anthiasxyz

11 months ago

Seeing this evolve over time from some of the early simulation efforts is inspiring and excited. Amazing go getters and builders behind this.

1 8 32 4K 10

Download Image

0 3 7 1K 1

Anthias Labs @anthiasxyz

11 months ago

1/ Our team at Anthias Labs is excited to announce the beta release of Arena: a robust LP simulation framework by @ts0yu0 via the Uniswap Community Research Program.

1 8 32 4K 10

Download Image

Noam Brown @polynoamial

12 months ago

Believe it or not, the name Strawberry does not come from the “How many r’s are in strawberry” meme. We just chose a random word. As far as we know it was a complete coincidence.

76 41 1K 110K 78

Sam Altman @sama

2 years ago

the efficient market hypothesis is interesting not only because it’s widely believed and obviously untrue, but also because it becomes more untrue the more people act as if it were true.

527 594 6K 1.4M 623

What we're building: 148th intent-based 5d solver rfq censorship resistant relay agnostic game theory optimal auction model with 7ms better latency than current sota What people want: 100x-DOGE-SHIB-PEPE-WIF-SLERF index

3 7 41 4K 2

Umbral @umbralxyz

2 years ago

Umbral is a crypto-native strategic consultancy. We leverage deep-domain expertise to help projects ship superb protocols. 1/n

3 4 41 18K 9

Dragan Rakita @rakitadragan

2 years ago

Parallel EVM claim, how to verify the claim of parallel execution. Wrote on this here: rakita.github.io/blog/blog/para…

6 28 117 19K 49

Georgi Gerganov @ggerganov

2 years ago

@yacineMTB @suchenzang I mean if the end is inevitable, why not just get there faster

3 5 66 3K 1

roon @tszzl

2 years ago

human level ai is boring. please give vast datacenters drawing most of civilizations energy needs calculating inscrutable mysteries where not even the problem statement, much less the solutions, would make sense to mortal minds

57 38 519 47K 21

anton 🇺🇸 @atroyn

2 years ago

@natfriedman the path forward is: - discard all 'alignment' '''work''' - recast the field as controls and interpretability - get some people who know what a dynamical system is working on the problems good chance to unlock some pretty interesting stuff!

6 15 126 11K 15

roon @tszzl

2 years ago

to enter the server you must produce a token of -21 logprob

32 5 314 42K 8

Sterling Crispin 🕊️ @sterlingcrispin

2 years ago

Agent Based Corporations, swarms of autonomous intelligent agents form corporate entities in temporary alliances to compete against other swarms. Corporate mergers, hostile takeovers, and liquidations in hour timeframes. Generative Adversarial Corporations are created to simulate…

7 19 158 32K 54

kache @yacineMTB

2 years ago

sorry if I can't rebuild the universe from first principles using your primitive i don't want it

4 1 34 2K 1

Georgi Gerganov @ggerganov

3 years ago

Here is 4-bit inference of LLaMA-7B using ggml: github.com/ggerganov/llam… Pure C/C++, runs on the CPU at 20 tokens/sec (M1 Pro) Generated text looks coherent, but quickly degrades - not sure if I have a bug or something 🤔 Anyway, LLaMA-65B on M1 coming soon!

24 125 721 116K 252

anton 🇺🇸 @atroyn

3 years ago

why do most approaches to controlling model behavior rely on acting only in the model's input domain? prompt engineering, fine-tuning etc. all feel like trying to pick a lock with a wet noodle.

7 7 60 12K 24

murat 🍥 @mayfer

3 years ago

brains consistently having having hemispheres is pretty sus. it might be proof that having smaller latent space interfaces between different multi-modal encoders/decoders is a huge computational advantage maybe why text-to-image models get away with being so darn small

4 3 34 3K 7

Neel Nanda @NeelNanda5

3 years ago

We trained a transformer to grok modular addition and reverse-engineered it. We found that it had learned a Fourier Transform and trig identity based algorithm, so cleanly that we can read it off the weights! I did not expect this algorithm! I found it by reverse-engineering.

3 23 118 14K 29

Download Image

Lúcás Meier @cronokirby

3 years ago

Nothing can ever be faster than the state of the art, because then the state of the art is itself

1 1 4 1K 0

Xavier @xavi3rlu

3 years ago

@guil_lambert left the following exercise in one of his Medium post. My solution here 👇🧵 Nothing crazy, maybe it can help you with your financial engineering pset :P