shouldn’t the weight distribution of the interactions be learned/different for all users?
what commenting means to me ≠ what commenting means to you
im pleasantly surprised. i performed another internal benchmark with @pranayd_ using gpt 5.6 sol on ultra effort. fable ranked them:
1. gpt 5.6 sol ultra
2. current design
3. fable max
the design it produced beats the existing design refined by multiple senior engineers and actually used
how are you guys using llms for system design?
i ran an experiment where i looked at a senior engineers design for a pretty generic system, had fable 5 design it without seeing the finalized system on max effort, and the result was a design that even fable admitted was inferior to the senior engineers design.
my thought is if the best way to learn system design was on the web fable would’ve been trained on it
how are you guys using llms for system design?
i ran an experiment where i looked at a senior engineers design for a pretty generic system, had fable 5 design it without seeing the finalized system on max effort, and the result was a design that even fable admitted was inferior to the senior engineers design.
my thought is if the best way to learn system design was on the web fable would’ve been trained on it
building autoskill.dev to fix this exact issue
Watching people endlessly collect Codex prompts and Claude Skills reminds me of a strange phenomenon in Japan: compulsive hoarding.
Different objects but same psychology.
Collecting tools feels productive.
But if you don't know what problem you're trying to solve, another Skill won't move you any closer.
Tools amplify direction but don't replace it.
Sami retweeted
I just built a SKILL.md router that searches a hosted catalog of 200k skill, applies relevance + safety gates, and returns one skill, or nothing.
Let me know what you think
the technique to
solve this is called Fermi estimation, it’s when you break down an ambiguous quantity by using estimates.
for example, the number of trees on earth must be larger than the number of square miles of land because there is more than one tree per square mile of land, just take the Amazon rainforest for example
.25% CTR on the billboard vs .01% CTR with @contentrewards. the billboard seemed to engage people better than online marketing.
that being said, this ad very insightful and using the service definitely reaches more people than the billboard.
