Paul Graham: "Emails Written by AI Feel Like Being Lied To"

Paul Graham, founder of Y Combinator and an early investor in OpenAI, has made it clear that he ignores emails from founders that were evidently written by AI. The reason: "It feels like being lied to."

Graham's reaction is hardly an exception. Multiple studies back up the legitimacy of the intuitive unease people feel toward AI-generated text. While the spread of AI tools has made "efficient writing" easy, the side effect of eroding recipients' trust is becoming increasingly visible. The fact that one of the startup world's most important gatekeepers is rejecting AI-written emails carries important implications for thinking about where the line should sit on AI use in professional communication.

US Federal Courts Face a Flood of AI Lawsuits

A joint study by MIT and the University of Southern California has released startling data: since ChatGPT's widespread adoption, lawsuits filed in US federal courts without a lawyer (pro se litigation) have nearly doubled. According to the survey, one in five complaints contains AI-generated text.

AI was expected to democratize access to the courts, but in reality it is heading toward sending judges into a panic. To process the flood of AI-generated complaints, judges are being forced into drastic countermeasures. It is an ironic situation: a technology that was supposed to close the "justice gap" is now creating an "existential paperwork nightmare."

MIT Tech Review Shows What AI's Impact on Employment Really Looks Like

Amid ceaseless warnings that AI will wipe out white-collar jobs en masse, MIT Technology Review has published two significant analyses.

The first is a big-picture research piece. An analysis of US Bureau of Labor Statistics data shows that the unemployment rate in AI-exposed occupations is actually lower than in less-exposed ones. Nor is there any evidence of a large-scale occupational shift toward "safe" manual labor driven by AI fears. At this point, no data exists to support large-scale job destruction.

The second piece, however, sounds the alarm. According to a working paper (November 2025) from Stanford University's Digital Economy Lab, employment fell by a relative 16% in highly AI-exposed occupations among 22- to 25-year-olds. Anthropic's March 2026 report supports the same conclusion. No comparable decline is seen among more experienced workers — the problem is concentrated in early-career roles with high AI exposure. The findings suggest that AI is replacing "junior tasks" in jobs such as software developers, customer service representatives, and programmers.

If companies have begun replacing the work of employees in their first one to three years with AI, then the bottom rung of the career ladder is quietly being knocked out. Redesigning educational institutions and policy intervention in youth employment have become urgent priorities.

Microsoft and Uber Raise Questions About the Cost-Effectiveness of AI Investment

Microsoft and Uber are running into unexpected cost problems with their adoption of AI coding tools. According to reports, cases are starting to emerge where the cost of operating AI tools exceeds the cost of hiring people.

An Uber executive has explicitly stated that "justifying AI spending is getting harder." It is becoming clear that adopting AI purely as a cost-cutting tool may not retain its economic rationality as it scales.

You could say AI investment has entered the stage where its "unit economics" are being called into question. Reality is starting to show that cost estimates from real-world operations have not kept up with the initial expectations of productivity gains.

China Expands Travel Restrictions to Private-Sector AI Talent

China has expanded travel restrictions on top-tier AI talent from state-owned enterprises to private companies. According to Bloomberg, key researchers and engineers at private AI companies will now face strict constraints on overseas travel.

This could have a major impact on the global battle for AI talent. There have been cases of leading Chinese AI researchers moving to overseas research institutions and companies, but such moves will now be physically restricted. The international openness of AI development is being further eroded by geopolitical factors.

A "Map of Meaning" Found Between the Brain and LLMs

According to a study gaining attention on Hacker News, researchers using Sparse Autoencoders have found a semantic mapping between the human cerebral cortex and large language models (LLMs).

Whether LLMs are mere statistical pattern matching or possess a structure of "understanding" shared with human cognition has long been a subject of debate. This study suggests that a meaningful correspondence exists between the two. The significance of adding empirical data to the debate over the nature of AI understanding — a debate that has continued since the "Chinese Room" thought experiment — is considerable.