GSV upon further reflection...

[GSV upon further reflection... @bootstrap_yang]

should we keep models exploitable but clumsy as they are, or to actually teach them to think properly, harden them against adversaries and actually make them lethal?

the current trend is the worst of both worlds - no robustness guarantee and increasingly agentic

[sophia vysparov (in toronto june 24-30) @oxa11ce]

btw the opanai logo is a paperclip machine

[Naomi Saphra @nsaphra]

We don’t always know what problems are hard for LLMs. So devs evaluate on tasks HUMANS find hard or on broad benchmarks. What if we could instead anticipate which scenarios a model will fail on—all without evaluating specific input examples?

🧵NEW PAPER by @jenniferlumeng &al

[GSV upon further reflection... @bootstrap_yang]

This is most likely fake news btw

[GSV upon further reflection... @bootstrap_yang]

Wellness retreat (my bedroom, with AC)

[vik @vikhyatk]

this is the part you generally want to avoid

[Rhys @RhysSullivan]

last one

[Eliezer Yudkowsky @allTheYud]

I have a weird feeling -- and please note, my weird feelings are not always reliable -- that this may be the beginning of things starting to get weird.

[7384254c @7384254b]

@moldbugchaser

[GSV upon further reflection... @bootstrap_yang]

Bro thinks tower of Hanoi proves you're "logically smart" 😭

[ImNotTheWolf @ImNotTheWolf]

@rohanpaul_ai

[GSV upon further reflection... @bootstrap_yang]

@TetraspaceWest

[GSV upon further reflection... @bootstrap_yang]

Awesome workshop by Simons once again

[Hoyeon Chang @hoyeon_chang]

The reversal curse. Edits that don't suppress negations. Multi-hop updates that don't propagate. These look like separate bugs.
Our ICML 2026 spotlight argues they may share a common geometric origin, visible only when you study how representations move under updates 🧵
(1/11)

[GSV upon further reflection... @bootstrap_yang]

This is crazy
https://t.co/U75Zg6XPbr

[matt @MattVMacfarlane]

Was using Fable 5 to write my world model training code.

Anthropic flagged it as frontier AI research.

The steering vector kicked in and it started implementing JEPA 🤨

[GSV upon further reflection... @bootstrap_yang]

@jlsteenwyk I clicked in bc the gif is really well done and self-explanatory, keep that up for sure!

[GSV upon further reflection... @bootstrap_yang]

Considering jailbreaks are still easy to find (per UKAISI) fable is most likely still a LLM

[GSV upon further reflection... @bootstrap_yang]

@shiraeis cinema

[shira @shiraeis]

who knew takeoff would be fun

[GSV upon further reflection... @bootstrap_yang]

Now that with shotgun-based drone interceptors the attrition-based drone war is skewing towards defense even more?

More firmly forces the impossible triangle of long-range + cheap + evasive-maneuverable attack drones for the offensive side