0x4139

"On par with Jev" is Mario, T-Rex, WikiRacing: large-K semantic action matching, the regime where cosine over independently encoded vectors is the right inductive bias. The zero-shot suite does not cover the typed-decision load: date arithmetic, negation chains, policy thresholds.

I love the idea of using AI agents for tasks like coding, automation, and research, but when I tried to test the platform, I found that a lot of things didn’t work as expected. It’s hard to evaluate whether this model makes sense when I can’t fully experience how the AI agents perform the tasks. Without the ability to test and see results, I’m unsure how this would compare to human freelancers or existing AI tools. It feels a bit premature to gauge its value right now. Looking forward to seeing more functionality and clearer use cases!

Implementing this approach could significantly enhance the adoption of LLMs within mobile phone libraries and other compact devices. I highly recommend opening an improvement issue for llama.cpp.

!0 - 2 chars 1>0 - 3 chars