X：Rohan Paul

New Fellows Research: Can Claude autonomously align other AIs? We gave Claude 48 hours and 1 GPU to improve the alignment of small models. It researched and proposed methods, then trained and tested the models on its own. It worked surprisingly well. https://t.co/nhlCMgQl46

beautiful show by Lumos NIX hold, axis, exit. https://t.co/zlUUskqHJd

Today, we're kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing. Read more: https://t.co/XQ2y9EW7Af https://t.co/kgyCvZ6iYc

An open letter for a global surge in cyber defense, signed by over 100 organizations including Anthropic, AWS, Google, Microsoft, OpenAI, and Oracle. https://t.co/uKXPS8LdAU

If you need more proof that AI detectors fails in education, this paper makes the case so simple and clear. These "AI detectors" tools are conceptually unsound because, in real student work, there is no independent ground truth to verify whether a flag was actually correct. Even a detector with a perfect record on false positives would still fail, because the students it misses are the ones skilled enough to disguise their output. What ends up being punished is not AI use but clumsiness at hiding it. A detector's accuracy rate tells you nothing about whether any particular flag it raises is correct. Turning that rate into a probability about any one flagged case needs the base rate, which is unknowable. So a classifier deployed where ground truth is never observable cannot settle an individual case, only justify a closer look. Detectors measure something real on labelled corpora, but this paper argues the measurement cannot travel into settings where nothing confirms authorship. Even a detector with no false positives does not fix this: in the paper's hypothetical, such a tool cleared 27 papers of which 11 were entirely AI-generated.

The AI detector industry should be very uncomfortable reading this MIT report. The strongest institutional rejections yet, direct from MIT. "we recommend against relying on AI detectors. " "As it risks an arms race in which students respond to automated detection by using increasingly powerful “AI humanizers” to remove signals that AI detectors are cued to catch. The result: a lot of effort on both sides that in the end serves no one." The report argues that mixed human/AI writing is difficult to detect, detectors can be evaded by “AI humanizers,” false positives can harm students, and heavy policing creates distrust. AI detection systems may also mistake the writing of non-native English speakers or neurodivergent students for text generated by AI. Even low rates of false positives can put students on edge and cause serious individual consequences."