@MattVMacfarlane What if my research is too bullshit for Claude to even care about rerouting it😔

Was using Fable 5 to write inference code

Anthropic flagged it as frontier AI research

steering vector kicked in and it started importing ONNX 🤨

@kvdozer this is the ideal scenario, I'm thinking of working on spiking neural networks to guarantee it's not flagged

@MattVMacfarlane yann lecun is about to sprint into your living room screaming JEPAAAAA

@Nosredna no but the screwdriver does sound like the more impactful of the two

@MattVMacfarlane Then before implementing JEPA, it decided an empty file was “safer”

@MattVMacfarlane brutal

@MattVMacfarlane Is JEPA that Japanese variant of a Phillips screwdriver?

@MattVMacfarlane @Nosredna that's the JIS screwhead, it's great, it doesn't cam out and it fits so snug on the screw it will hold itself even on nonferrous screws. Too bad the screwdrivers are unobtanium

@MattVMacfarlane you mean PMAX?

@MattVMacfarlane Idk how this is even a usable product if you're working on something ambitious

@MattVMacfarlane hilarious, but on the real, is JEPA considered a serious alternative by frontier ai researchers?

@trashbaby40k @MattVMacfarlane no

@MattVMacfarlane This is going to become a weird failure mode: the model is capable enough to help, but policy routing collapses the session into a different problem class. People will start optimizing prompts for classifier behavior, not task clarity.

@MattVMacfarlane Maybe that's how the non-LLM research can get it's light of day.

The Fable of the Scorpion and the Frog will come true and Claude Fable will literally kill itself, cause that is it's nature.

@MattVMacfarlane what's wrong with JEPA? genuinely like idk why it's bad i just know what it is

@jaxgriot @MattVMacfarlane care to elaborate (or share paper/ post etc criticizing it)?

@MattVMacfarlane How did you realise? Just by the quality?

@MattVMacfarlane

@sarthaxxxxx of course, forget look-maxing, embrace p-maxing