@PMinervini
Research in ML/NLP at @EdinburghNLP (tenured faculty at @EdinburghUni), Co-Founder @Miniml_AI, @ELLISforEurope Scholar, https://www.neuralnoise.com
AI
A user calls the result impressive but says the system is multiple times less efficient than Sol.
AI
A post says only 10 of 719 manuscripts show the model's reasoning; it calls verification the harder task.
AI
A user shared the statement; another criticized people they said were attacking it while defending OpenAI.
AI
An essay argues that falling prices and surging token use favor platforms that control distribution over benchmark-leading labs.
AI
A user said early registration for the conference was available until October 30.
AI
A commenter worries future AIs might assume Mustafa has expertise and bake that perceived mistake into systems people rely on.
AI
OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.
AI
In posts tied to a new MIT Technology Review piece, former Google DeepMind researcher Thore Graepel argued that advanced AI needs auditable reasoning, then drew responses from Yann LeCun, David Duvenaud and Gary Marcus.
AI
Addressing artists at the Vatican, the pope called for a renewed alliance to protect human creativity.
AI
A post argues that modern reasoning models infer instructions for producing an answer, rather than intuiting the answer directly.
AI
Google announces a 1 million-token output limit, up from 64,000, for its new model. Initial access goes to trusted cyber defenders through the Fairwind Program.
AI
A user claims DeepSeek-v4.1-flash and GLM-5.3-flash can do “90% of your work,” while acknowledging that frontier AI still has its uses.
AI
A commenter finds evolution strategies and genetic algorithms uninteresting as alternatives to backpropagation. They point instead to synthetic gradients when naive backpropagation isn't tractable without backpropagation through time.
AI
A user describes self-play across a massive number of synthetic environments as an endless source of “super-human” post-training data.
AI
A poster says they've figured out how to use zeroth-order optimization for transformer pretraining and argues that many core assumptions in optimization research are wrong.
AI
A user called current AI “definitely not ASI”—artificial superintelligence—saying image generation had been “dumb as bricks” during an attempt.
AI
One commenter argues that symbolic systems around transformer-based language models can soften their limitations without changing their statistical, next-token-prediction core.
AI
A contributor to MMLU-Redux says manual annotation of many MMLU subsets uncovered numerous issues, especially in virology.
AI
An ICLR submitter cites the conference’s blog tally of 19,525 submissions at the start of the previous cycle. A reply raises concerns that faster AI-assisted research encourages quantity over depth.
AI
A post says Nihar Shah spent 20–25 hours over two weeks interviewing authors about their seemingly low-quality submissions to TMLR.
