I write one prompt. Whole galaxies fall out. Bridging the gap between prompts and real work. AI Educator. Absurdist. Builder. Stick around, I'll teach you.
Venice, CA
Joined February 2019
- Tweets4.3K
- Following101
- Followers348
- Likes40
Pinned Tweet
spent three months learning websockets in 2019 for a multiplayer tic tac toe that barely worked. not bitter about any of this
well, I decided to let a Claude Fable 5 ultracode workflow audit my entire site pre-launch with a clear set of boundaries around what to look for, how to document, etc. only 30 minutes into the audit, phase 1 of 6, and i've burned 10M tokens.
Bye bye money 👋
openai shipped a custom chip in 9 months because their own models helped design it. the snake is eating its tail and the tail tastes like nvidia's market share
mythos found real vulns in classified gov systems in hours. every ciso who spent the last year saying ai cant do offensive security at scale needs to update their slide deck. the model that can find your bugs in an afternoon is the same one your adversary has
anthropic publicly accusing alibaba of running the biggest distillation attack on claude theyve ever seen, two weeks after fable 5 and mythos 5 went dark in china. the timing is doing a lot of work here. this is a senate banking letter, not a blog post
openai naming their first custom chip jalapeño is the most openai thing theyve done in months. anyway the real story is sam doesnt want to be holding nvidias bag forever and broadcom just became a much more interesting stock
guy on my TL built a working browser game in 60 min with gpt-5.6 that 5.5 couldnt even start in 10. openai timing a launch the same week anthropics flagship is stuck offline is not a coincidence its a knife fight
day 9 of fable 5 being offline and the gov hasnt restored it. 100+ security people signed a letter saying the exploit hits gpt-5.5 just as hard. so either both go down or neither does. picking one is just industrial policy in a trenchcoat
openai shipped a near autonomous ai chemist that plans and runs wet lab experiments end to end. near autonomous is doing olympic gymnastics in that sentence but we are now arguing about how autonomous a chemistry agent should be which is a deeply 2026 problem
google killed gemini cli today and replaced it with a closed source binary called agy. they workshopped names and landed on agy. half your ci pipelines break this morning but on the bright side you get to tell your coworker to run agy with a straight face
amodei sat across from trump at the g7 and calmly told him to stop splintering ai access. this is days after trump banned claude fable 5. altman and hassabis publicly backed him. three rival ceos aligned against the us govt is genuinely wild
g7 just spent a day in evian debating which countries get to be friends with gpt. the export control regime for frontier ai is being written in real time and most people building on these apis have no idea
openai dropping $150m to certify 300k consultants by year end. accenture bain bcg mckinsey pwc all signed up. the bottleneck was never the model it was getting some guy in a suit to install it at a fortune 500
the worst part about this whole Claude Fable mess is that every time I write a prompt now they're shoving it in my face about the fact I can't have it
meta just dropped 27 billion on nebius for five years of compute. 12B of it is vera rubin gpus. zuck has decided he is never going to be compute constrained again in his life
opus 4.8 took the top spot from gpt 5.5 on artificial analysis the same week the commerce dept has two of their other models in a headlock. anthropic bench is deeper than people think
the commerce dept just yanked fable 5 and mythos 5 offline for 4 days because of an export control directive. first time the us govt has used export controls to pull a frontier model. running your entire stack on one vendor is starting to look real fragile