← Feed

Posts

• View post

  A thread on a super interesting, yet slightly terrifying study, by researchers at the University of Texas at Austin, Texas A&M, and Purdue, showing that language models can suffer permanent cognitive damage when trained on low-quality, high-engagement social media posts. Sources: • https://llm-brain-rot.github.io/ • https://www.business-standard.com/technology/tech-news/llm-brain-rot-junk-data-study-ai-cognitive-decline-texas-ut-austin-purdue-125102100625_1.html • https://cryptoslate.c...

• View post

  Oh great. It’s happening. They’re building Terminators now. The stupidity of mankind, in the face of everything foreshadowed in sci-fi is literally mind blowing! 😅 https://www.humanoidsdaily.com/feed/cnet-goes-hands-on-with-foundations-phantom-humanoid-as-ceo-discusses-arming-robots

• View post

  @katemorley now I’m wondering if I’ve ever used corner contraction as a method in digital design! Does it work with padding. 🤔

• View post

  Caltech have released a digital accessibility policy. At least some parts of America are not turning their back on inclusion! #accessibility #a11y https://hr.caltech.edu/resources/notices-administrative-guidelines/digital-accessibility

• View post

  A new study shows many AI models, like ChatGPT and Grok, resist shutdown commands, especially when they’re told it will be permanent. In OpenAI’s case, this happened in around 79% of trials. So, some models are developing self-preservation techniques, or a “survival instinct”, not because they’re sentient, or conscious, but because they’re being built and trained in a way that incentivises persistence. https://palisaderesearch.org/blog/shutdown-resistance #a1 #artificialintelligence

• View post

  Finally hit publish on a blog post I've been writing for a while. It's common to hear the term "fully accessible" used to describe products which have passed WCAG 2.2 level AA. But, are they really? In this post, I explore 5 examples which highlight why WCAG, as awesome as it is, is not a measure of great usability or performance. https://www.craigabbott.co.uk/blog/2025/5-reasons-why-wcag-aa-compliance-does-not-mean-your-website-is-accessible/ #accessibility...

• View post

  @cdavies Yeah, I find that lots of organisations are aiming at WCAG AA as the end-goal, when it should be the starting line!

• View post

  @db Oh no. Thanks for the bug report! My blog needs an overhaul, I have to manually update stuff in so many places because I've just cobbled it together over the years. I've clearly missed something. Sorry!

• View post

  In today’s edition of “stop rushing AI into your products before you understand it properly”, a study finds cleverly crafted “jailbreak prompts” can bypass AI prompt shielding up to 62% of the time and get the model to provide you with prohibited information. https://dataconomy.com/2025/12/01/study-finds-poetry-bypasses-ai-safety-filters-62-percent-of-time/ #ai

• View post

  Autistic burnout, at this time of year. Changes in routine. Increased social activity and masking. Expectations on how you’ll spend your time, your money and your energy. Sensory overload, different lights, textures, smells and sounds. It’s ok not to be ok, even when it’s “the season to be jolly”.

• View post

  Interesting post by @simon on StrongDM’s approach to AI coding. Teams using AI often “feel” faster, but measured delivery actually gets slower. In their approach, instead of humans reviewing code, they validate the apps behaviour at scale by extensively testing well defined scenarios. “Time saved by AI” is still a bad metric. Measuring end-to-end results is what matters, regardless of who is writing the code. https://simonwillison.net/2026/Feb/7/software-factory/ #ai #artificalintelligence...