ChatGPT Work, 413 Ruff rules, and an open-weights split
Four original X posts cover a ChatGPT Work trip-planning workflow, Ruff's broader lint defaults, the prospect of continuous model updates, and the risk assumptions behind open weights.
The day in four posts
One X post describes ChatGPT Work planning a trip for nine people and building the site to choose it. Another says a Ruff update raised the default rule count from 59 to 413. The other two are arguments about how AI products will change: François Chollet predicts model releases will become continuous, while Ethan Mollick says the open-weights dispute begins with different assumptions about future risk.
Posts published July 25-26, 2026.
Product workflows
1. A trip planner that is also a site builder
- What happened: Sam Altman, posting from the verified @sama account, said he asked ChatGPT Work to use his chat history, propose three long-weekend options for eight friends, build a full-stack decision site for all nine travelers, make reservations after agreement, and draft a Gmail message. He wrote that it "just worked." Read the post 1
- Why it matters: The request combines personal context, planning, web development, group coordination, reservations, and email. It is one user's report, not a benchmark or a product specification.
- Reaction: At capture, the post had 7,058 likes, 837 replies, and 641,551 views. 1
Altman's post is the direct account of the workflow he says he ran:
Cargando tarjeta de contenido…
Developer tools
2. Ruff's new defaults find old problems
- What happened: Simon Willison, creator of Datasette and co-creator of Django, wrote that Ruff 0.16.0 increased default-enabled rules from 59 to 413 and surfaced 1,618 issues in sqlite-utils. Read the post 2
- Why it matters: A lint upgrade can turn into a real triage task when the default policy expands this much. The 1,618 figure describes Willison's project, not a general rate for Python repositories.
- Reaction: At capture, the post had 514 likes, 30 replies, and 53,368 views. 2
Willison's original post links to his account of what the new rule set changed:
Cargando tarjeta de contenido…
Model releases
3. Chollet expects the version number to fade away
- What happened: François Chollet, co-founder of Ndea and ARC Prize and creator of Keras and ARC-AGI, predicted that major named model launches will give way to continuous updates without widely publicized version numbers, "probably less than 2 years away." Read the post 3
- Why it matters: This is a forecast, not a product roadmap. If it happens, version names become a weaker public boundary for comparing behavior, evaluations, and regressions.
- Reaction: At capture, the post had 1,966 likes, 99 replies, and 160,931 views. 3
Chollet's prediction is short and deliberately specific about its time horizon:
Cargando tarjeta de contenido…
Open models and risk
4. Mollick describes a disagreement beneath the open-weights argument
- What happened: Ethan Mollick, a Wharton professor who studies AI, wrote that many people supporting open-weight models may not share lab insiders' expectations of grave, semi-autonomous biosecurity risks. He added that no one knows which view is right. Read the post 4
- Why it matters: His point is about the premise of the dispute. Arguments about openness can diverge before they reach policy because participants assign very different probabilities to future capabilities and harms.
- Reaction: At capture, the post had 839 likes, 69 replies, and 127,988 views. 4
Mollick frames the disagreement as a difference in beliefs about what advanced systems may do:
Cargando tarjeta de contenido…
Keep the evidence types separate
Willison reports an observable tool change with project-specific counts. Altman reports one workflow, Chollet makes a prediction about product cadence, and Mollick identifies a belief gap in a policy argument. Each is useful, but they do not carry the same kind of evidence.
Contenido relacionado
- Inicia sesión para comentar.