
Justin Lê
5.4K posts

Justin Lê
@mstk
jle / mstksg / Lê Anh Khoa: Computational Science PhD, Physics BS, Haskeller & amateur musician; he / him






had problems with an LLM agent stopping before the objective was done so I made a BS "completion-checker" binary that always just exited with an error code immediately and told it to iterate until the checker passed lol


I HATE CONNECTIONS THIS IS NOT A RED HERRING ITS A CATEGORY RAAAAAA


I kinda want my timeline to go back to FP people arguing about stuff that I do not yet understand. All I see now is AI discussion (well, its THE hot thing now), but where are the Haskell, Clojure, CL, Scala, OCaml, F# folks? I barely see them FP posting anymore.



1. 🚨 Finally I get to share my new paper: "A Disproof of LLM Consciousness." I show that *no* falsifiable and non-trivial theories of consciousness could ever work for LLMs. Intriguingly, turns out that cracking continual learning might change this web3.arxiv.org/pdf/2512.12802

everyone is building an agent framework we already have an agent framework: it’s called English I’ve been building something too it’s not a product or service, it requires no dependencies, it ships as a skill it’s barely even anything at all, it’s just language but it works



I can watch endlessly as Gemini wanders through a temporal anomaly between the date of its training data and real time


The more you know.



Oh a new Sci-fi series I wonder what it’s about? “A group of quirky rebels works to free the universe from the hands of an evil Empire…”










