convexstrictly

Pricing input: $8/1M tokens output: $24/1M tokens https://docs.mistral.ai/platform/pricing/ …

Jeremy makes compelling arguments. Here are some more mundane corollaries: It is only a matter of time before you and your company are affected by the pending regulations. In the future, almost all …

Some information here. https://www.ntia.gov/federal-register-notice/2024/dual-use-f... …

The comments will inform the drafting of regulations on open weight models under the Biden executive order on AI using his powers under the Defense Production Act. Fact Sheet: https://www…

"By enabling the use of a single high-precision base model accompanied by multiple 1-bit deltas, BitDelta dramatically reduces GPU memory requirements by more than 10x, which can also be translat…

https://github.com/FasterDecoding/BitDelta

Twitter summary: https://twitter.com/ssgrn/status/1738256456250470853 Github: https://github.com/KaiNylund/lm-weights-encode-time …

Everything you say makes sense. Training is definitely more compute intensive than inference. Training is both memory throughput and compute constrained. Much research in speeding up training goes i…

Could you explain the blockers to getting back-propagation working well on your chips?

Research suggesting that much of the power of the transformer architecture comes from associative recall over long sequences that does not require scaling model dimensions. They design state space mo…

"... we find that a duo of a 1.3B generation model and a 1.3B verifier model can achieve 81.5% accuracy, outperforming existing models that are orders of magnitude larger."

The paper claims it builds upon the concepts in HashGraph, an efficient CUDA hashtable implementation. HashGraph (2019) https://arxiv.org/abs/1907.02900 Anyone know what the most…

Github repo https://github.com/harp-lab/gdlog

Twitter thread with video introduction https://twitter.com/1x_tech/status/1730610445541638378 …

Emily Chang from Bloomberg reports: Satya Nadella was "blindsided" and is furious. Details in paywalled article: https://www.bloomberg.com/news/articles/2023-11-18&…

If you were logged into Bing, those prompts may be in your history. They can be viewed using the Edge browser. A few weeks ago, I had spotty service with Bing Chat where it would keep resetting the c…

Github Copilot Chat is in beta and is marketed as a separate product. I suspect it is using a later generation (and better) underlying model. I used it many months ago, and I agree it is much better …

Yes, I have been using it the last few days. I haven't noticed the failure case of Bing Chat not answering the question at all. Are you sure you had the modes I recommended turned on? Try rep…

After the recent updates to the UI, the old 50 messages every 3 hours warning went away. Though I haven't really tried pushing it past that limit to tell if that restriction is still there.

Bing Chat (Choose "Creative Mode" or "Use GPT-4" depending on UI) is GPT-4 with retrieval augmented generation using the Bing Search engine. It seems to be tuned a little differen…

"Greg Brockman works 60 to 100 hours per week, and spends around 80% of the time coding. Former colleagues have described him as the hardest-working person at OpenAI." https://tim…

"I am currently on leave from MIT and spending it at OpenAI." https://madry.mit.edu/

Jakub Pachocki and Szymon Sidor have worked on mu-parametrization/tensor programs and Dota 2. https://www.semanticscholar.org/author/J.-Pachocki/2713380?s... As @eachro…