Prompts

Stanford researchers Owen Queen and James Zou retested 50 widely cited findings about LLMs on 20 models released between July 2024 and September 2026. They class 44% as expired on today's models, 34% as weakened, 16% as persisting and 6% as strengthened. Capability failures such…

Trakkr, an AI search tracking company, asked ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews 200 questions that call for a number, then followed 1,285 of the cited statistics back through up to five pages. It reports that 40% reached a study saying the same thing, 8%…

Researchers at NYU and UT Austin compared text from 14 models, including Claude, GPT-5.6, GLM 5 and Kimi K2.5, with human writing and with plain paraphrases of the same content. Compared with human writing, all nine frontier and near-frontier models used more lifeless things doin…

In a blog post, Amrut Joshi argues that when you ask an LLM how confident it is, the number it gives (say 0.85) is just more generated text, not a measurement. It was never checked against how often the model is actually right, so it can change if you ask again or reword the ques…

A paper posted to arXiv on 28 September by six researchers at Google Research and one each from MIT and Harvard tests whether language models leave out flaws that undercut an otherwise good account of their work, which the authors call insecure reporting. In one of eight test sce…