The lack of consent for datasets is a serious problem in ML. By addressing this, we could make a step towards healthier ML practices. I started building something around decentralised data sharing with consent and provenance verification. If you are interested, let's chat at #ECCV2026!
First slide: “AI sucks!” I will start to like the guy in the end 😂
I name this summer “Beer and AI” 🤖
One great use case of agentic LLMs, for which I haven’t seen much discourse about (although obviously I’m sure people do this), is cleaning large-scale pre-training data with calling tools, including open-source LLMs etc. It does an AMAZING 🤩 job, and saves tones of time!
I watch the discussion over Anthropic on Bluesky with 🍿🍿🍿
RT from @sashamtl.bsky.social's LinkedIn as I found it really important given the huge gap between AI developers and common public in here: "Everywhere I look, I see people pissed off at AI – voting in the midterms, organizing rallies and protests, even vandalizing buildings in my neighborhood." 1/1
I guess my takeaway from the maths OpenAI saga, by also taking into consideration how mathematicians respond to the situation (as everyone should), is: If it cannot reason, and therefore the humans still need to do all the verification/correction/deciphering, then tell it to shut the fuck up.
Does it sound a bit like Jev is probably in the direction of combining Transformers/LLMs with neurosymbolic reasoning or is it only me?
Suddenly everyone's algorithms started "self-improving"... Like, there are no human researchers working in the companies, or thousands of raters for RLHF. LIARS
