Alcides Fonseca.AI & Programming Languages researcher
40.197958, -8.408312

Making intelligent software
more reliable.

I work on the intersection of AI, Software Engineering and Programming Languages. Currently, I am applying Formal Methods to Agents, towards AI Safety.

Program synthesisLiquid typesReliable AIGuardrailsEvals

Draft: More evidence I was right

And, as another paper wrote, LLMs can also get brain rot from bad data (although I’m not fond of this term because it is now so strongly associated with humans).
The future doesn’t look very hopeful, because we’re already almost there with that bad and synthetic data! On the blog of visualization tool Graphite, we can read that more than half of the new articles on the internet now come from AI. The dead internet theory, once a crazy conspiracy theory, is starting to sound credible. After all, Google itself has already said that the open web is in decline (Dutch).

— Felienne Hermans

Ghostly introduced Vouch, a tool to create a recommendation-based credibility system

People must be vouched for before interacting with certain parts of a project (the exact parts are configurable to the project to enforce). People can also be explicitly denounced to block them from interacting with the project.

— Mitchell Hashimoto

TODO: linkar o the anonymous internet is over.

  • https://garymarcus.substack.com/p/we-urgently-need-a-federal-law-forbidding