Best of your X follows: June 7

Sunday is the quietest day on the timeline, and today's edition shows it. But four posts from the past 24 hours were sharp enough to cut through: Ethan Mollick with two quick observations about AI's effect on ideas and on writing, François Chollet drawing a crisp line between knowledge and intelligence, and a retweet from Yann LeCun flagging a surprisingly wild week for open-weight models.

Ethan Mollick argues the AI-era gap to watch is not between people who have access to models and those who don't — it's between people who have genuinely rare, hard-won ideas and those who don't. Execution is collapsing in cost; ideation is not. 1

The implication for product people, researchers, and founders is direct: your competitive moat is now upstream of the keyboard, sitting in whatever unconventional problems you've earned by living through them.

Mollick also flagged something easy to miss: AI writing habits are bleeding into product menus and tooltips. 2 "A report is not 'what leaves the room'" is the kind of specific, funny-because-it's-true example that shows how language models trained to sound analytical produce menu items that read like motivational posters rather than functional labels.

If your product surfaces LLM-written copy anywhere in the UI, this is worth a review pass.

François Chollet in eight words: "Scaling knowledge gives you static competence. Intelligence gives you adaptability." 3

This is a compressed version of the argument Chollet has made at length about ARC-AGI: retrieval of trained patterns and genuine generalization to novel problems are different capabilities, and current benchmarks often conflate them. The line lands harder as the week closes on a run of impressive new model releases — because it asks whether any of that scaling is closing the adaptability gap, or just refilling the competence shelf.

Yann LeCun retweeted a summary noting 25+ notable open-weight model drops in the past seven days — what one observer called "one of the most insane weeks ever for open AI." 4 LeCun's amplification signals the open-weights community sees this as a meaningful moment, even if most of the individual releases didn't make the mainstream tech news cycle.

The practical question is what this pace does to evaluation cycles. If 25 models drop in a week, the gap between "released" and "properly benchmarked" keeps widening — which loops back neatly to both Chollet's point about what metrics actually capture and Mollick's earlier thread about the widening gulf between frontier and open-weights tiers.