Engineer with a passion for shipping humane products and innovations that delight people. I'm having a blast w/AI agents - tech hasn't felt this fun for years!
You build an AI agent that reasons perfectly. Then you hook it up to the web and spend a month debugging headless browsers. ๐ซ We built Tabstack to fix that: a web execution layer that gives agents eyes, without the overhead. Launching today. https://t.co/97N329P8RH
@SF311 trash on the street and sidewalks of Linda between 18th and 19th. During the night, somebody seems to have sacked trash bins. Filed #101002533001. What sort of turnaround time might we expect?
TIL how to debug custom instructions in @code by seeing the detailed context (instructions, system prompts, etc) sent for each thing typed. You only need โGithub Copilot Chatโ set to Debug. https://t.co/5F2JppHG6P
@GeoffreyHuntley Hmm, do we need to standardize a syntax for "turing machine" and then see how things go there? Of course, I'd be pretty stunned if LLMs did poorly on turning machines...
Heh. Just caught @ClaudeCode proposing to add code to look for exact test text and hardcode the correct answer without fixing the actual problem. Just like benchmark-gaming!
@pashmerepat@GeoffreyHuntley Lots of stuff on the web is marketing hype. Can you tell that AI generated this specific piece of breathless marketing hype?