Does Baguettotron already have a space on @huggingface ? It would make for a greate model for showing how different HF tools work and can inference on CPU

Our 🧠brain-recording pipeline linked to LLMs is working exceptionally well I just received my best mental profile yet (family confirmed), based on a 1-hour technical product discussion while wearing an EEG The goal is to help LLMs understand their users and their preferences The next stage is reporting back after meeting what we: missed, didin't have time to consider misunderstand could we do better next time would like to know more (search & rethink) Also sorting events by importance for us, in our language and mental framing The report part is already looking interesting as when LLM understands user it all becomes easier and now LLM can do quaite a lot of proactive actions :D

Our benevolent Shoggoths in near future when they will be responsible for writing news and books

⚡️We are building BCI based proactive assistant⚡️ one that knows without writing what you like when and how We are starting Angel round this month and search for tiny $ from people engaged in AI comunity or whoich would like to partisipate or get the BCI thx @TheAhmadOsman 😊

We are building website for our startup (soon to be anounced) full of magic (but tested and used) BCI for LLM training and live generation odjustment - collection part works and semi live feedback improves users flow/happines Mechanistic interventions for LLMs robustness, creativity and controlability - creativity was the goal but we found that it can be achieved only with higher robustnes and dynamic adjustments - some live controlability achieved, more to come - entropix can be generalized to an absurd lvl RL and reasoning over open domains based on BCI feedback and more - we have quality info for each token, overal quality on multiple dimentions from humans - Mech interpretability gives us the same and more ( didin't anticipate that) - we help sister company that builds advanced judges swarms - will search for RL enthusiast as that's above current budget We found that algos (some patented) I build for past 15y for EEG analysis later generalized to other domains are god given for DL interpretability (and EEG+LLM integration) - What we found in LLMs is generalizable to new architectures - to new interpretability tools - but needs basic research (not experimentation) so a lot of time We are slowly starting to beta tests some smaller elements e.g more creative LLMs