x.com/garrytan/status/2032726172099092913

The post describes S7263 as a New York effort to ban "AI that outscores doctors on medical exams." That is not what the bill text says.

Official New York Senate materials describe S7263 as a bill that "imposes liability for damages caused by a chatbot impersonating certain licensed professionals." The operative text says a chatbot proprietor "shall not permit such chatbot to provide any substantive response, information, or advice" that would amount to the unauthorized practice of specified licensed professions or law. In other words, the bill is framed around unauthorized professional practice and chatbot impersonation, not around whether an AI system scored better than doctors on an exam.

I did not find any language in the official bill materials making exam performance the trigger for regulation, or describing the bill as a ban on AI systems because they outperform doctors on medical tests. That makes the claim materially misleading about what S7263 actually does.

This sentence compresses an LSC statistic in a way that changes its meaning.

The Legal Services Corporation's 2022 Justice Gap Study says: "Low-income Americans did not receive any legal help or enough legal help for 92% of the problems that substantially impacted their lives in the past year." That is narrower than the post's wording in two important ways:

1. It is not all legal problems. The 92% figure is specifically for problems that substantially impacted people's lives.
2. It is not the same as "go unaddressed." The official wording combines cases with no legal help and cases with insufficient legal help. Those are not the same thing as every problem being wholly unaddressed.

So the post turns a qualified metric about substantial-impact problems and inadequate assistance into a broader claim about all low-income legal problems going unaddressed.