@alphaxiv.org released quickarXiv

Swap arxiv → quickarxiv on any paper URL to get an instant blog with figures, insights, and explanations.

When I read these articles about those new browsers from perplexity or openAI I really want to agree, but hyping it all up like this makes it very difficult. Calling it an extension of consciousness is a bit of a stretch. It's an untested product

hybridhorizons.substack.com/p/when-ai-be...

A new definition for AGI just dropped, and it is a bad one.

I genuinely loved this read about GitHub code search! 💻

Read "The technology behind GitHub’s new code search." on their blog!

Great article explaining how vLLM is using KV caching under the hood
www.ubicloud.com/blog/life-of...

“They said it could not be done”. We’re releasing Pleias 1.0, the first suite of models trained on open data (either permissibly licensed or uncopyrighted): Pleias-3b, Pleias-1b and Pleias-350m, all based on the two trillion tokens set from Common Corpus.

Germany has fallen.

A person on a bicycle is by far the most energy-efficient among animals and machines per distance traveled relative to body weight. The bicycle is magic.

www.jstor.org/stable/24923...

Wondering how many enterprise LLM applications use explicit user feedback to improve given that only 3% of users actually use those thumbs up/down buttons. Have seen people forcing users to provide feedback at random times however. Implicit Feedback such as session length seems easier to use

Interesting take on the actual adoption of AI assisted coding. Suggesting that only 5% of professional developers who have access to GitHub Copilot are using the more advanced featurs, most (63%) use auto complete only.

youtube.com/watch?v=Up6W...

He wanted to test how models performed when subjected to a new tokenization scheme without re-training the model.
Modifying Llama 3’s tokenizer to tokenize numbers from right to left (R2L) instead of left-to-right (L2R) with just a few lines of code 🧑‍💻. This affect the grouping of numbers by three:

Many of the current LLM eval tools, such as deepeval, mlflow, and evidently.ai are really great! However, I find it hard to choose as most of the tests are trivial and many framework feel like they want to lock me in without providing much added value.