yb

enterprise AI @spacex @spacexai

opinions my own, not reflective of any other entity…

[yb @yashb30]

@shaunmmaguire inspirational

[yb @yashb30]

@alexandr_wang @dtrinh nice one

[yb @yashb30]

big day today

[yb @yashb30]

Great quote I heard today - you cannot swim for new horizons until you have the courage to lose sight of the shore ….

[yb @yashb30]

it all goes back to first principles!

[yb @yashb30]

gotta give it to ssi - they had the foresight of the 'superintelligence' wave

[yb @yashb30]

@poteto @_ak_111 Lol

[yb @yashb30]

@CompleteSkeptic @a16z Congratulations

[yb @yashb30]

@siddarthpaim top tier ragebait

[yb @yashb30]

@pubity Voldemort plane

[yb @yashb30]

@melvindvivas nope

[yb @yashb30]

@yunta_tsai San Jose doesn’t deserve to be on this list - and i say this having grown up there!

[yb @yashb30]

hello friends 👋🏽

[yb @yashb30]

@omarsar0 the 256k no-overflow run is the real signal. you still get +9.2 when nothing hits the wall so the policy is learning which stale exploration to drop so resume doesnt rewalk dead ends, not just surviving context length

[yb @yashb30]

@ChristosTzamos fixed weights only move the hard part to the write path. if noisy overwrites bury the cell the learned indexer was trained to hit, growing memory never surfaces the new skill at inference

[yb @yashb30]

@idavidrein the failure mode is grading the agent's own diary. does Harbor seal trajectories on a write-once path the sandbox cant truncate, or is the eval still trusting whatever the agent leaves on disk?

[yb @yashb30]

@levelsio when build speed outruns ideas the bottleneck is intake. keep a queue of paid-user pains from support and churn notes and only open a build slot when one clears, otherwise you just ship faster into nowhere

[yb @yashb30]

@KLieret @jyangballin @OfirPress @18jeffreyma @YuxiangWei9 @jma__06 @sten_sootla @parth007_96 @thinkstepbystep @pengchengyin @magpie_rayhou default is hint free yeah. i meant the ablation. gold filenames alone like 2x Sol (4.8→11) and line ranges roughly 2x that again (~19). same contracts, just handing it a pointer under the turn budget

[yb @yashb30]

@wen_kaiyue readiness gate is carrying this. drop it and partial rollouts feed the policy bad critic advantages so the run stalls. the 2.5x FLOPs save only kicks in after that per-problem error check flips

[yb @yashb30]

@ArtificialAnlys Claude Code vs Codex vs Antigravity on one bar chart mixes scaffold skill with model skill. the 68 vs 63 race is partly which harness you bolted on