@Hertz it’s 2026 - stop making me pick each location to show me it’s sold out. Just find me the nearest location that has a car available!!!
It shouldn’t be this hard.
Catastrophic forgetting in continual learning has an architectural solution.
I’m excited to release TFGN: a new transformer overlay that lets LLMs keep learning new domains (Prose → Python → Math → Biomedical → Chinese → JavaScript, 1B tokens per domain) with no replay buffer and no task IDs, while preserving prior knowledge at LLM scale.
Tested from ~398M to ~9B parameters.
Tightest result: BWT = −0.007 on LLaMA 3.1 8B Retrofit.
Full paper (arXiv): https://t.co/NVQABvRsAu
Readable explainer: https://t.co/VIXycvOV6i
Would love your thoughts, this is my first major ML paper.
We stopped working the fields so we needed physical gym
With AI, we will stop thinking critically for work - and so we will need brain gyms
Brain gyms will be hugeee
In case someone hasn’t spelt it out for Apple and Google - you need a Muse competitor ASAP!
In the mobile age - the phone came before the App Store - in the AI age, it will be reverse.
Win the personal AI platform, win the hardware that goes with it.
@HamelHusain It’s about the need to consistently manage the things you want - you say - “Muse, remind of this and note when new things like it happen” - with codex that’s more than a singular message once
First impressions on muse - this is a game changer!
My hair is tingly from this being free though - which of my body parts have I unknowingly sold to Meta for this now?
Allow me this one vague post for the month. Just did private beta test of the first ever continual learning model to be announced next week.
A tsunami is coming
I love cleaning really messy rooms. Pick one trash at a time, one corner at a time. No moment feels too impactful but you look back after an hour or two and you’re amazed.
Building a start up feels the same. One daily problem at a time.