Some interesting links that I Tweeted about in the last week (I also post these on Mastodon, Threads, Newsmast, and Bluesky):
- AI can program drone controllers. But can they program other AI? https://www.theregister.com/2025/08/07/generative_ai_codes_drone_control/
- Google shifts computing around data centres according to local power demand, in an effort to reduce the effects of the massive energy demands of AI: https://www.theregister.com/2025/08/04/google_ai_datacenter_grid/
- The trouble with politicians using AI is that it makes it easier for the AI companies to manipulate them: https://www.theguardian.com/technology/2025/aug/05/chat-gpt-swedish-pm-ulf-kristersson-under-fire-for-using-ai-in-role
- AI still cannot be trusted with medical decision making, the risk of hallucinations is too high: https://futurism.com/neoscope/google-healthcare-ai-makes-up-body-part
- "Therefore I'm not surprised that there appears an openness to accept AI when ultimately it's an industry with little obligation to be accountable for their actions." That dude did not hold back: https://www.theregister.com/2025/08/07/real_estate/
- So when this AI crawler finds a site it isn't supposed to scrape, it switches to a stealthier mode and scrapes it anyway: https://arstechnica.com/information-technology/2025/08/ai-site-perplexity-uses-stealth-tactics-to-flout-no-crawl-edicts-cloudflare-says/
- The fight over copyright in AI training data is getting more and more invasive with regards to users' data: https://arstechnica.com/tech-policy/2025/08/openai-offers-20-million-user-chats-in-chatgpt-lawsuit-nyt-wants-120-million/
- There is still some way to go in using AI to make medical notes: https://www.rnz.co.nz/news/national/569348/artificial-intelligence-saves-doctors-time-but-makes-mistakes-study
- Primary school teachers are struggling with AI: https://www.rnz.co.nz/news/national/569059/dire-need-for-ai-support-in-primary-intermediate-schools-survey-shows
- What human-level AI is going to bring: https://www.theguardian.com/technology/2025/aug/04/demis-hassabis-ai-future-10-times-bigger-than-industrial-revolution-and-10-times-faster
- Think watermarking is going to identify AI generated images? Nope: https://spectrum.ieee.org/ai-watermark-remover
- Demands in Australia that AI companies pay for the use of the copyrighted material used to train their models: https://www.theguardian.com/technology/2025/aug/06/arts-and-media-groups-demand-labor-take-a-stand-against-rampant-theft-of-australian-content-to-train-ai
- Microsoft has always had a problem moving from prototype to product, its AI projects are no exception: https://www.computerworld.com/article/4033846/microsofts-real-ai-challenge-moving-past-the-prototypes.html
- The ridiculous salaries being paid to top AI developers: https://arstechnica.com/ai/2025/08/at-250-million-top-ai-salaries-dwarf-those-of-the-manhattan-project-and-the-space-race/
- New AI models are jailbroken on the first day of release: https://dataconomy.com/2025/08/07/new-openai-models-are-jailbreaked-on-day-1/
- The dodgy ways AI companies make money from their users: https://www.computerworld.com/article/4036568/the-dark-side-of-ai-monetization.html
- OpenAI's work towards human-level AI: https://techcrunch.com/2025/08/03/inside-openais-quest-to-make-ai-do-anything-for-you/
- AI crawlers do not respect robots.txt. Does it need to be made legally enforceable? https://spectrum.ieee.org/web-crawling
- Using AI to backstop human errors seems like a sensible application: https://www.theregister.com/2025/08/06/mod_taps_aussie_ai_shop/
- People do not want Meta to use their data to train their AI: https://www.theregister.com/2025/08/07/meta_training_ai_on_social/
- Google denies AI summaries are costing sites traffic: https://techcrunch.com/2025/08/06/google-denies-ai-search-features-are-killing-website-traffic/ In other news, Jack the Ripper claims to be a public safety worker.
- Looks like models for fashion advertising are the next group to be put out of work by AI: https://techcrunch.com/2025/08/03/the-uproar-over-vogues-ai-generated-ad-isnt-just-about-fashion/
- You can run large language model AI locall. If you have the resources: https://www.theregister.com/2025/08/07/run_openai_gpt_oss_locally/
- If the future of software developers is only to verify the code generated by AI, how are the devs going to learn what to verify? https://www.theregister.com/2025/08/07/github_ceo_ai_coding/