juliette pluto 🌌

hacker, machine whisperer, typo-generator. Adversarial robustness @GoogleDeepMind. views mine.

[juliette pluto 🌌 @foundjuliette]

@florian_tramer They should customize the safeguard?message for recruitment targets. Instead of *conversation aborted* it should say “respond to recruiter to continue”

[juliette pluto 🌌 @foundjuliette]

@tmuxvim Good

[juliette pluto 🌌 @foundjuliette]

@DimitrisPapail Try https://t.co/IoW9gy3own; it makes codex verify Claude’s work until codex is satisfied

[juliette pluto 🌌 @foundjuliette]

@tszzl It’s nice that I still get to do so off the clock though; maybe we can give codex an outlet too…

[juliette pluto 🌌 @foundjuliette]

@tszzl I also have to keep reminding myself to talk about gremlins and goblins at work

[Natalie Wolchover @nattyover]

Bacteria move around using a molecular machine called the flagellar motor that rotates faster than the flywheel of a race car engine and switches directions in an instant. After 50 yrs, scientists have finally figured out how it works. “My lifelong quest is now fulfilled.” Link⤵️

[juliette pluto 🌌 @foundjuliette]

@bcherny Congrats on the launch!

[juliette pluto 🌌 @foundjuliette]

@NeelNanda5 Very cool! Great work.

[juliette pluto 🌌 @foundjuliette]

@rohanvarma Congratulations on the launch!!

[juliette pluto 🌌 @foundjuliette]

@KentonVarda Capability is there; just temporarily bottlenecked by misaligned RL rewards.

[juliette pluto 🌌 @foundjuliette]

@noahzweben that’s awesome! huge congrats on the launch

[juliette pluto 🌌 @foundjuliette]

@mahaoo_ASI @willccbb it’s okay; but you must print then tokens as paper first, which can then scan, as long as you burn it after.

[juliette pluto 🌌 @foundjuliette]

@rmcentush Reminds me of that time I inadvertently stole all of my bf’s saved passwords, bc he let me log into his iPad to setup a HomePod. iCloud auto sync did the deed.

[juliette pluto 🌌 @foundjuliette]

@emollick Called it

[juliette pluto 🌌 @foundjuliette]

Prediction: unless patched, GPT's tendency to say delve will be assimilated into North American style English. Soon it won't stand out as odd anymore.

More broadly, RLHF'ed LLMs will shape cultural norms in unexpected ways.

[juliette pluto 🌌 @foundjuliette]

@tszzl @emollick Contemporary AI makers of AI slop (eg it’s not x it’s y) are a lot closer to what real users prefer. They’re genuinely useful rethorical devices, albeit now a bit over used

[juliette pluto 🌌 @foundjuliette]

@tszzl @emollick this I think is just down to improved RL rewards. “Delve” was essentially a form of reward hacking: a phrase that Nigerian RLHF raters rewarded (due to its use in Nigerian formal English ), but that wasn’t actually liked by most users.

[juliette pluto 🌌 @foundjuliette]

@_arohan_ It was a long time ago, so back then probably not

[juliette pluto 🌌 @foundjuliette]

@_arohan_ In the process there was exactly one time when I received readability feedback that asked me to change smth. And upon review, it was revealed that the code generated by Gemini in fact had correctly followed the guidelines, and the human had misinterpreted them.

[juliette pluto 🌌 @foundjuliette]

@_arohan_ Gemini got me Python readability