austin

ex-aerospace engineer turned AI engineer. I like to build agentic workflows in messy domains.

[austin @abswio]

@arpit_bhayani is there a way to do this with decompiling at scale, with agents? like, don't move but lose 10 hp, see which values change, iterate those and see which actually have an impact on your health value? surely with a swarm this would be scalable to some degree

[austin @abswio]

@arpit_bhayani when I was a kid I used a hex editor to cheat coins into club penguin (I dont remember the currency xd). you could play a game to earn money, see which hex values changed, then edit the field that iterated and you'd know with relative certainty you were touching the right value >

[austin @abswio]

@HiloW_Ai do you think that pinned versions respond differently post-new-release too, or is it just the main head that has actual noticeable changes in behavior? I know some people talk about like post-release-nerfs on older model versions sometimes

[austin @abswio]

@JasBilyeu the one thing that never really sat right with me on AI-assisted CAD work is that I can never really tell if models are considering the underlying physical design principles an engineer would be... it feels like architecting more than structural engineering if that makes sense?

[austin @abswio]

@kazukiega this is something I've been trying to poke around in recently too... what are the odds that something like a cleanroom 'develop a test spec based off of the feature spec' flow would yield better results than just having an agent work directly off of the OG spec?

[austin @abswio]

@poteto yeah my anthropic onsites were manual code debugging, which at the time felt really silly because it was open book but you couldn’t use AI agents to assist, which felt kind of counter to the whole premise..?