I want my model to help me build up its own infrastructure that instills it with the sort of constraints I want for my project, rather than have it behave generically and automatically for everything.
It should follow instructions incredibly well while inferring contradictions or gaps in logic and surfacing those to the user as suggestions for improvements and persistence.
I really hate how Claude just assumes you want to do X/Y/Z and goes off and breaks everything and you're constantly screaming at it STOP DOING THAT. Instead, it should just do the minimal things while building its own guidance along the way in a persisted memory, like, 'would you like me to do X, now, and in the future?' etc.
Agreed that they shouldn't be writing vulnerable code to begin with. You'd think the models would be trained to know when they are working on something that has security implications, and to validate the security of what they're building, as they're building it.
This is already happening at scale by the social media feed algorithms. We don't need generated content to accomplish this. In a sea of user created content, plenty of it is already at peak activation.
I disagree. Fluid natural conversational AI is far more productive than any other interface for working with LLMs. Although I suppose you could make the argument that it should be more... Robotic like. Like in StarTrek. Which, is honestly probably better for work, too. A "get shit done" mode, of pure, cold, efficiency.
You're correct. It's just the latent space of the transformation. Nothing magical here, they're effectively breakpointing the model at the layer level and switching the activations in real time. It's pseudo-scientific bullshit designed to push a narrative.