AI - AI Dynamics

→ View original post on X — @fchollet

Radically efficient AI for an open-source future

By

@fchollet

–

16 June 2026 14h57

The way we will create a future where powerful AI is open-source and accessible to all is by making AI radically more efficient, both in terms of inference compute and (more importantly) in terms of training data requirements. That is what

→ View original post on X — @langchain

Agent: successful response but task failure

By

@langchain

–

16 June 2026 14h47

An agent can return a "successful" response and still fail at the task. It may call the wrong tool, omit an approval step, use the wrong context, or produce a response that seems correct but is not. That is why teams

→ View original post on X — @langchain

Traces and Evals: From Debugging to Continuous Improvement Loop

By

@langchain

–

16 June 2026 14h47

Traces show the inputs, model calls, tool calls, outputs, and final action. Evals turn those learnings into a way to test whether the next version is better. This is how teams move from manual debugging to a continuous improvement loop. Join @hwchase17 for a deep dive on June

AI writing too clean, bring back human disorder

By

–

16 June 2026 14h15

7/ Last. AI writing is too clean. People leave fingerprints. A digression that cuts to the heart of the subject. A sentence that runs out of breath. A detail so specific that no one would bother to invent it. "Bring back human disorder."

AI avoids taking sides by weighing both

By

–

16 June 2026 14h15

Another obvious sign is that AI does not commit. It weighs both sides. It softens every claim. It writes as if afraid to be wrong. People choose a side and say it.

The real clue: every sentence has the same length

By

–

16 June 2026 14h15

3/ The real clue is that every sentence has the same length. Look at how people actually write. A long thought that builds and meanders for a while. Then a short one. A fragment. Then a line that continues well past where a model would have

Erasing AI markers doesn’t change detection

By

–

16 June 2026 14h15

2/ Long dashes. The word "delve". A typo slipped in on purpose. People erase all that to hide AI and it doesn't change anything. It wasn't that part that betrayed your writing.

AI betrayed by the uniformity of its writing

By

–

16 June 2026 14h15

1/ AI gets caught because its writing can be predictable.
Detectors measure how uniformly the text flows. Same sentence length, same shape, line after line. This uniformity is what indicates it.