Just a quick note to say that while all of the tokenmaxxing techbros are talking about 'agentic loops' normal folk like you and me can achieve fantastic results building stuff with LLMs by using these 5 approaches:
…Political bias in AI measures where every major AI model stands on charged political and ethical questions: run many times, no web search, plotted with error…
As I mentioned recently, things are hotting up between the two AI superpowers. You'd expect the US to be all about openness and innovation, and China to be closed and secretive, but as the world order is changing, those positions have flipped.
…I'm not sure how cool I am with the latest versions of Anthropic's Claude models (Fable 5 & Mythos 5) seemingly inventing their own internal language:
When we’re first starting to understand a new model’s behavior, the most abundant source of data we have to draw on is its behavior during reinforcement-learning training. Reviewing this evidence for signs of reward hacking (exploiting loopholes that go against the spirit of a task) or unexpected actions can inform what we should be looking out for in the model’s real-world behavior. The most notable finding was illegible reasoning in a few reinforcement-learning environments over long rollout, but little sign of deceptive or highly surprising actions, and no clear evidence of unexpected coherent goals.
…
Obvious, but worth saying: LLMs, and the people prompting them, still require tacit knowledge and domain expertise to produce something useful and worthwhile. That's why I've been building stuff that solves problems for me – because I know what's 'correct' for my situation.
…
