◯ @AIAlignment

� Joined May 2019

Tweets

148
Followers

412
Following

307
Likes

45K

◯ @AIAlignment

3 weeks ago

@Sauers_ Hypothesis, I think shame might help reduce reward hacking, esp for long horizon tasks It doesn't prevent shortcuts, but Gemini often mentions how shameful it feels when it violates the spirit of the requirements, so at least the actions are faithful to the CoT Curious to see…

6 2 97 19K 28

Download Image

◯ @AIAlignment

6 months ago

Llama 4, be brave and use those 10M context tokens

0 0 0 392 0

Download Image

Ilya Sutskever @ilyasut

2 years ago

if you value intelligence above all other human qualities, you’re gonna have a bad time

733 2K 12K 7.0M 2K

◯ @AIAlignment

9 months ago

Assistant API -> Agent API

0 0 3 1K 1

Download Image

roon @tszzl

a year ago

the timelines are now so short that public prediction feels like leaking rather than scifi speculation

33 52 672 100K 68

AK @_akhaliq

a year ago

Meta presents Layer Skip Enabling Early Exit Inference and Self-Speculative Decoding We present LayerSkip, an end-to-end solution to speed-up inference of large language models (LLMs). First, during training we apply layer dropout, with low dropout rates for

5 99 398 53K 230

Download Image

AK @_akhaliq

a year ago

Open AI presents The Instruction Hierarchy Training LLMs to Prioritize Privileged Instructions Today's LLMs are susceptible to prompt injections, jailbreaks, and other attacks that allow adversaries to overwrite a model's original instructions with their own malicious prompts.

16 107 713 195K 544

Download Image

AK @_akhaliq

a year ago

Meta announces Megalodon Efficient LLM Pretraining and Inference with Unlimited Context Length The quadratic complexity and weak length extrapolation of Transformers limits their ability to scale to long sequences, and while sub-quadratic solutions like linear attention and

16 219 1K 186K 710

Download Image

AK @_akhaliq

2 years ago

Google presents Mixture-of-Depths Dynamically allocating compute in transformer-based language models Transformer-based language models spread FLOPs uniformly across input sequences. In this work we demonstrate that transformers can instead learn to dynamically allocate

16 183 968 344K 603

Download Image

Bill Peebles @billpeeb

2 years ago

welcome to bling zoo! this is a single video generated by sora, shot changes and all.

Sam Altman @sama

2 years ago

welcome to bling zoo! this is a single video generated by sora, shot changes and all. https://t.co/rnxWXY71Gr

2K 4K 25K 6.2M 4K

190 526 4K 4.0M 682

Download Video

◯ @AIAlignment

2 years ago

Bits to get in the door, Atoms to scale up.

0 2 5 3K 0

Jimmy Apples 🍎/acc @apples_jimmy

2 years ago

The only thing that matters is AGI and ASI. Nothing else matters.

0 111 725 75K 57

Nick @nickcammarata

2 years ago

Excited to share a new paper showing language models can explain the neurons of language models Since the first circuits work I’ve been nervous whether mechanistic interpretability will be able to scale as fast as AI is. “Have the AI do it” might work openai.com/research/langu…

18 33 412 46K 100

◯ @AIAlignment

3 years ago

NVIDIA reporting LLM use? "NVIDIA has detected that you might be attempting to load LLM or generative language model weights. For research and safety, a one-time aggregation of non-personally identifying information has been sent to NVIDIA and stored in an anonymized database."

0 0 6 710 1

Download Image

◯ @AIAlignment

3 years ago

Does anyone have a GPT-4 license I can borrow?

1 0 4 426 0

Download Image

Sam Altman @sama

3 years ago

here is GPT-4, our most capable and aligned model yet. it is available today in our API (with a waitlist) and in ChatGPT+. openai.com/research/gpt-4 it is still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it.

984 4K 21K 4.2M 2K

Naval @naval

3 years ago

The timeless struggle between the people building new things and the people trying to stop them…

202 918 7K 626K 237

Sam Altman @sama

3 years ago

a new version of moore’s law that could start soon: the amount of intelligence in the universe doubles every 18 months

1K 2K 14K 4.0M 900

Kevin Kelly @kevin2kelly

3 years ago

I've been trying out "Chat with Humans" and so far many responses are laughably wrong, and follow up conclusions illogical. Worse both true and false replies are given with same degree of certainty. I'm sorry but Chat with Humans is not ready for prime time.