Will Chen
@stablechen
AI R&D @IdyllicLabs, prev: dev rels @terra_money, head of R&D for wasm devx @cosmology_tech & @terran_one, @ucberkeley '19. tweets = notes to self
computation from first principles 1. stability -- the ability to hold a measurable physical property enables "memory" 2. cycles -- oscillations between stable states, translates time into computation via changes. "the clock" provides order and sequence. 3. causality -- when…
LLMs don't really have continuity, each LLM call is just LARPing as if it did, trying to pretend to be the AI process that came before so asking it "why" it did something is often made up -- it has no introspection beyond what can be gleaned from history
[email protected] come join Axiom; help us figure out why prime numbers are hiding from their ex
i used to believe "humans are irrational" meant that humans are unpredictable and act randomly, to use a rational model to understand their behavior is useless now I understand what it really means: humans will behave VERY predictably the direct opposite of rational behavior
I have this on hotkey and iterative loops with Claude Code for this have been incredible STOP MAKING HALF-ASSED GUESSES AND APPLYING INCREMENTAL EDITS AND WASTING MY TIME! I want you to solve this problem by completely doing a thorough EXHAUSTIVE!! EXHAUSTIVE!! manual trace…
the little 44min claude code outage is the diff between 99.9% uptime and 99.99% uptime SLA guarantees
We actually have a justification for this very intuition "I would never surpass the full ability of someone else because they were playing to their strengths which I didn’t have" -- we show that RL can adapt/calibrate CoTs to the model's own capability, while SFT often leads to…
agentic workflows are basically linear programs that delegate recursion to the model
trying to fall asleep = insomnia trying to be creative = forced, shitty work trying to have flow state = self-consciousness kills it trying to be charismatic → comes off as fake trying to meditate → more mental chatter trying to have good sex → performance anxiety don't try,…
you are doing yourself, and by extension the world, a disservice by believing the false narrative that your authentic voice is not enough
many aspire to become thought leaders but leadership demands courage to share their thoughts in their authentic voice
it's not about catching up what you missed, it's about waking up what's dormant and already there
Ai has taught me that no matter how much u berate or question why it's being so retarded or insult it, it will not magically improve performance but somehow positive encouragement in the right context works
have been in my apartment for the last 3 weeks didn't see any fireworks just huel and pushups
best way to work with AI is to prompt it to interview you one question at a time, while performing simulated simulated annealing