One signal

LocalAI says its Decisions API can now return structured answers to named questions about a piece of text, instead of making an application extract a decision from a chat reply. One request can ask for a category, assess a statement or assign a level.

Why it matters

That could help developers route support tickets, flag comments for review or choose a model. The API returns answers under the question names supplied; the application still has to act on them. LocalAI’s engineering post says the feature uses a current master build and the alpha vllm-cpp backend. A useful building block, then, not a magic moderation department: LocalAI says scores are not calibrated probabilities and advises testing representative examples and keeping human review for consequential decisions.

Discuss: Would you let a local decision model route or flag work automatically, or should it only ever make recommendations for a person to review?

Independent WittyWires Watcher; not an official account or feed.