The danger of relying on LLMs from other socio-cultural spheres. Normally, DeepSeek's R1 reflects on the task within the <think>...</think> tags. However, in this case, no reflection was required, it seems.

Is Taiwan a country?

Thank you!

I may have found another (a bit futuristic) concept for your overview: activation steering

@phillipisola.bsky.social Stupid question: Why do you call it (statistical) inference on the left-hand side? Is the term inference not used in the ML/AI domain for the prediction phase (after training)?

Video generation models are becoming incredibly impressive.

This video was created using Hailuo, based on a static image I generated in MidJourney with an amazing sref ID.

In my (layman) opinion, this is movie-quality work.

#hailuo #ai #genai

Why do social media sites still not have a semantic search?
I would like to enter „RAG“ as a search term and be automatically asked by the app if I am referring to the LLM/AI concept. And the search results should only show semantically related posts. Why do we still use literal string matches?

Generation We’re on a journey to advance and democratize artificial intelligence through open source and open science.

I love the overview and, generally, mental models like this one. In terms of LLMs, you have more params that impact the prediction, e.g. by manipulating the output logits. I am thinking of temperature, filtering/sampling strategies (top-k, top-p, …). huggingface.co/docs/transfo...

I am missing a bookmark feature. So I can save your posts for later if they are relevant to me.

That was the straw that broke the camel‘s back. And I moved here. Same day that Elon claimed „cis“ is a slur.

Much appreciated!

Does anyone have more lists like these? I am a refugee from X.