Ramon Astudillo
@ramon-astudillo
Principal Research Scientist at IBM Research AI in New York. Speech, Formal/Natural Language Processing. Currently LLM post-training, structured SDG and RL. Opinions my own and non stationary. ramon.astudillo.com
It seems the suspected 5T models from Anthropic and OpenAI are kinda close in cybersec skills?
It's funny how these things keep happening. OpenAI is B2C, has the largest consumer base ... but their LLM is best at harder problems (better suited for B2B), while having a poor personality (worse suited for B2C uses). Anthropic is B2B but their LLM has the nicest personality, ideal for B2C
Alternative: Pre-training at the largest scale is very hard and having a U-turn prone CEO makes it harder. From: x.com/i/status/204...
man, the "GPT-4o was my friend and you killed it" people are a different breed. Probably living in a different universe.
"playing only" seems more realistic than UBI as something that would compensate for the death of white collar jobs.
This whole thread is pretty amazing. Some profs. seem to want corporate jobs and they don't know it x.com/i/status/203...
Has anybody ported Boris's multi-post Claude setup x.com/i/status/200... here? (cc @bcherny.bsky.social )
Gemini 3 Deep Think having +1000 ELO against Claude Opus 4.6 is ... a bit too much?
I sound like an overfitted LLM with temperature one (I swear I did not look at this before 🤣)
One of the cool things of working in a lab in the middle of the forest is the animal sightings. Geese, just live in the lab, deer at the boundaries are normal to see, also turkeys. I have heard bears and last week an amazing bobcat (photo not mine)
👆also related to the "yes darling" hypothesis x.com/i/status/143...
Stack Overflow questions over time (via HN) data.stackexchange.com/stackoverflo...
There are clear difficulties in multi-modal architectures that became evident after this, but dreaming is still free. I think I'd like to upgrade the text-to-bitmap model of 22 to cross-attended bitmap/text LMs 👇 x.com/RamonAstudil...
So one possibility is that we enter the "playing only" stage of human development. That or "children of men, but with more robots" x.com/RamonAstudil...
It would be really funny if the neural translation ends up hallucinating content that meaningfully moves research forward chinarxiv.org
3y ago Amazon had a bad home assistant with the best market share, Google had the best voice tech with a so so LLM and OpenAI the best LLM tech with no product. Apple was beating everyone at product but Siri ... well ... 👇 x.com/RamonAstudil...
They are actually closest to a general purpose computer! See e.g. @karpathy.bsky.social 's Software 3.0 view. IBM defines this as "Generative Computer", I prefer "Neural Computer". The idea is basically what I thought would be the GPT4 paper's title (my title was less boring) x.com/RamonAstudil...
News from the LLM companies vs Neuro-Software companies front. LLM ones own the upstream and can optimize training to their products. NS companies have domain knowledge but more limitations on applying it. However ... they can optimize for runtime infrastructure, which they own!. Maybe this is key?