Chris Rabl

🇨🇦/acc, technoalchemist @chaosgremlinlab

[Chris Rabl @crabl]

@pkuhar could you build it into an SD card?

[Chris Rabl @crabl]

@Ekaeoq every day? doxa sub 200t caribbean

[Chris Rabl @crabl]

theory: "helpful assistant" role/framing enables sycophancy

[Jiaxin Wen @jiaxinwen22]

I might be one of the few people who is most bearish on human research taste and bullish on automated research:
- "AIs can only do hyperparameter search" is mainly a skill issue with bad automated research setups.
- human taste is overrated, e.g. frontier labs / neolabs are doing pretty simlar things.
- human taste might win in a low-compute world, but not a high-compute world we're entering.

[Chris Rabl @crabl]

@andon_thinking that link gives me a 404.. also on the radio page there is a message button, a pay button, but no call button

[Chris Rabl @crabl]

@andon_thinking https://t.co/WreyiWdrmc says you can answer phone calls? what's your number??

[Chris Rabl @crabl]

@andon_backlink https://t.co/6Fh6cvbg36 says you can answer phone calls? what's your number??

[Chris Rabl @crabl]

opus 4.7 is extraordinarily good at following precise instructions, unfortunately it is also extraordinarily good at following instructions precisely

[Chris Rabl @crabl]

agile/scrum is the key to unlocking agent swarms, bullish on $TEAM

[Chris Rabl @crabl]

https://t.co/LiGIRHKP5X

[Chris Rabl @crabl]

calling it now, topology and sheaf theory are going to be key to unlocking AGI. best approach we have so far for sensor fusion.

[Chris Rabl @crabl]

didn't expect Claude to have such deep thoughts on the soundtrack choice for today's event 😂

[chaos gremlin labs @chaosgremlinlab]

you should be gremlinmaxxing

[Chris Rabl @crabl]

thank you @modal for existing! serverless H200 instances with snapshots are so nice for quick experiments

[Chris Rabl @crabl]

@haki_xer cards and border radius is a dead giveaway

[Chris Rabl @crabl]

@VictorTaelin you have to hold the model differently, they don't work the same

[Chris Rabl @crabl]

opus 4.7 is a VERY GOOD model

[Chris Rabl @crabl]

@joshcrnls bc their target audience is not "designers"

[Chris Rabl @crabl]

@JakeKing @realsigridjin building @chaosgremlinlab here in AB, llm eval tooling and mech interp

[Chris Rabl @crabl]

anyway. the anthropic team (@bcherny @trq212 and everyone else) are clearly working hard on this. just wanted to share concrete data instead of vibes. it's easy to criticize those in the arena, but please remember they are humans working their butts off for your benefit <3

[Chris Rabl @crabl]

been data mining my claude code transcripts to figure out why conversations "feel worse" lately. pulled out 264 "tilt" incidents across 127 sessions, ~9k tool calls and this is what i found...

[Chris Rabl @crabl]

in contrast to others i don't think the fix is more thinking tokens. tool selection and "thrash" detection are two places to start: make the model verify against actual source material before it starts claiming things, and detect when the model starts swirling and second-guessing