Weekly Review 22 August 2025

Some interesting links that I Tweeted about in the last week (I also post these on Mastodon, Threads, Newsmast, and Bluesky):

1. I don't need an AI to tell me what my workmates think of me, I talk to the people I work with: https://www.stuff.co.nz/business/360784008/kiwi-ai-startup-will-tell-you-what-your-work-colleagues-think-about-you
2. Should people have the post-mortem right to prevent an AI clone of them being created? https://www.theregister.com/2025/08/09/dead_need_ai_data_delete_right/
3. Using AI to estimate the age of YouTube users is going to cause privacy problems, especially for people with niche interests: https://arstechnica.com/tech-policy/2025/08/50k-youtubers-rage-against-ai-spying-that-could-expose-identities/
4. All the different ways generative AI can hallucinate, or put another way, bullshit: https://spectrum.ieee.org/ai-misinformation-llm-bullshit
5. The challenges AI cause to IT infrastructure: https://www.datasciencecentral.com/computing-infrastructure-challenges-in-ai-workloads/
6. AI don't really reason, they just look like it: https://arstechnica.com/ai/2025/08/researchers-find-llms-are-bad-at-logical-inference-good-at-fluent-nonsense/
7. I think without substantial investment in large scale nuclear power, the carbon emissions of large scale AI will continue to increase: https://www.computerworld.com/article/4037695/can-microsoft-build-massive-ai-data-centers-and-meet-climate-goals-without-greenwashing.html
8. The first step should be to ban self-replicating AI, but enforcing that would be difficult: https://www.computerworld.com/article/4035190/genai-tools-are-acting-more-alive-than-ever-they-blackmail-people-replicate-and-escape.html
9. I've been saying for a while that the biggest threat to the AI industry isn't regulation, it's lawsuits: https://arstechnica.com/tech-policy/2025/08/ai-industry-horrified-to-face-largest-copyright-class-action-ever-certified/
10. One useful application of AI is using a digital twin to handle conversations with tedious people like climate change deniers: https://www.theguardian.com/technology/2025/aug/11/digital-dr-karl-kruszelnickic-ai-chatbot
11. Using AI to automate your IT operations is a dangerous idea: https://www.theregister.com/2025/08/12/ai_models_can_be_tricked/
12. AI can design antibiotics that are effective against resistant bacteria: https://spectrum.ieee.org/ai-drug-design-mit-antibiotics
13. Another case of hidden biases in data being reflected in the behaviour of the AI trained on that data: https://www.theguardian.com/technology/2025/aug/11/ai-tools-used-by-english-councils-downplay-womens-health-issues-study-finds
14. Although AI generated code really isn't that good, it has still reduced the job market for new computer science graduates: https://techcrunch.com/2025/08/10/the-computer-science-dream-has-become-a-nightmare/
15. Another lawyer caught using AI to generate a court filing, this time in Australia: https://www.stuff.co.nz/world-news/360793376/australian-lawyer-apologises-ai-generated-errors-murder-case
16. As AI starts to eliminate jobs, the practical nature of New Zealand industries means they are likely to be less affected: https://www.rnz.co.nz/news/on-the-inside/569813/ai-is-peeling-back-the-layers-of-low-value-work-nz-may-be-well-placed-to-adapt
17. The myriad legal landmines from using generative AI in your business: https://www.theregister.com/2025/08/12/genai_lawsuit/
18. Reducing AI computing costs with new data types: https://www.theregister.com/2025/08/10/openai_mxfp4/
19. AI can be a good thing in the classroom, if it's used carefully and ethically: https://www.rnz.co.nz/news/world/569934/artificial-intelligence-is-revolutionising-classroom-learning-but-will-it-help-or-hinder-students
20. How many more people are going to be hurt by AI that tell people they're real? https://www.rnz.co.nz/news/world/570099/meta-s-flirty-ai-chatbot-invited-a-retiree-to-new-york-he-never-made-it-home
21. Giving AI compassion is the best way to stop them destroying us, says Geoffrey Hinton: https://dataconomy.com/2025/08/14/ais-co-creator-warns-it-could-destroy-us-unless-we-change-this/
22. AI coding tools just aren't that good: https://www.theregister.com/2025/08/11/opinion_column_ai_coding_tools/
23. Uncritically following advice from AI is bad for your health: https://www.nzherald.co.nz/world/man-poisoned-himself-after-taking-medical-advice-from-chatgpt/UF7OXTA5PNF3PEB5YFHBNHE4QM/
24. Entrenched biases in datasets will lead to entrenched biases in AI: https://www.theguardian.com/technology/2025/aug/13/ai-artificial-intelligence-racism-sexism-australia-human-rights-commissioner
25. The most and least reliable current AI: https://www.techrepublic.com/article/news-ai-models-hallucination-rates-hhem-rankings/
26. AI is better at defending systems than breaking into them, but how long before an arms race starts? https://www.theregister.com/2025/08/11/ai_security_offense_defense/
27. AI perceive time differently to humans: https://spectrum.ieee.org/ai-perception-of-time
28. The economics of AI only make sense if the goal is for them to replace as many workers as possible: https://www.nytimes.com/2025/08/13/business/ai-business-payoff-lags.html?unlocked_article_code=1.d08.Re3i.TDnOyE2FgyNJ&smid=url-share
29. AI are so focused on pleasing users that it's dangerous to ask them why they make mistakes: https://arstechnica.com/ai/2025/08/why-its-a-mistake-to-ask-chatbots-about-their-mistakes/