What I found interesting this time

by: Artur Dziedziczak

August 17, 2025

ā€œLLMs’ ā€˜Simulated Reasoning’ Abilities Are a ā€˜Brittle Mirage,’ Researchers Find,ā€ n.d. https://arstechnica.com/ai/2025/08/researchers-find-llms-are-bad-at-logical-inference-good-at-fluent-nonsense/

Another paper finds issues with LLMs’ reasoning capabilities. Apparently, patches made with ’reasoning’ models do not reason at all.
Recently, OpenAI released a GPT-5 paper that contains basic problems in its charts. I really think that people are applying LLMs in places where they should not be used.

ā€œTrapping Misbehaving Bots in an AI Labyrinth,ā€ n.d. https://blog.cloudflare.com/ai-labyrinth/?utm_source=clivethompson&utm_medium=email&utm_campaign=linkfest-37-wind-theft-an-html-bomb-and-the-rice/

Cloudflare tool which creates invisible for regular users links which trap AI scraping bots.