Weekly Review 27 February 2026

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

1. Greenwashing applies to AI as much as any other industry: https://www.theguardian.com/technology/2026/feb/17/tech-companies-traditional-ai-generative-climate-breakdown-report
2. No New Zealand lawyers have yet been caught using AI to generate their submissions: https://www.leightonassociates.co.nz/post/no-nz-lawyers-or-employment-advocates-have-been-called-out-for-hallucinated-ai-yet
3. Hiding what an AI coding agent is doing is not making developers happy: https://www.theregister.com/2026/02/16/anthropic_claude_ai_edits/
4. White-collar jobs like in the film industry are really getting hammered by AI: https://www.stuff.co.nz/business/360938829/she-was-earning-65000-ai-came-along-what-happened-nex
5. Google needs to do a better job warning people about the dangers of its AI health recommendations: https://www.theguardian.com/technology/2026/feb/16/google-puts-users-at-risk-downplaying-disclaimers-ai-overviews
6. Using AI to cheat on a course about AI seems rather wonderfully ironic: https://www.theregister.com/2026/02/16/kpmg_partner_in_oz_turned/
7. Generative AI can't do much to help the environment, but more classic machine learning can: https://dataconomy.com/2026/02/17/study-finds-74-of-ai-climate-claims-lack-evidence/
8. If a tool like AI is available that makes peoples' jobs easier, they tend to use it, whether it's appropriate or not: https://www.rnz.co.nz/news/national/586932/corrections-takes-action-against-staff-s-unacceptable-use-of-artificial-intelligence
9. How AI plots driving pathways for a Mars rover: https://spectrum.ieee.org/perseverance-rover-nasa-anthropic-ai
10. AI needs a huge amount of electricity, geothermal power generation is a more environmentally friendly way of getting it: https://www.theregister.com/2026/02/18/google_ormat_geothermal_datacenter_deal/
11. Court cases continue around whether or not AI can be legally considered an inventor: https://www.rnz.co.nz/news/national/587013/can-artificial-intelligence-legally-be-an-inventor
12. The problems with taking AI projects from pilot to production are not unique to AI, but common to all IT projects: https://www.informationweek.com/ai-innovations/from-pilot-purgatory-to-productive-failure-fixing-ai-s-broken-learning-loop
13. Building AI data centre satellites on the moon is not going to change the fundamental problems with the idea: https://www.extremetech.com/aerospace/xai-talks-up-building-a-lunar-mass-driver-to-launch-ai-satellites-into
14. Dedicated AI hardware, in this case neuromorphic chips, continues to improve and gain new abilities: https://dataconomy.com/2026/02/18/sandia-labs-uses-neuromorphic-chips-to-solve-partial-differential-equations/
15. Translators and artists are seeing their incomes slashed by AI: https://www.stuff.co.nz/nz-news/360939695/they-were-earning-120000-then-ai-came-along-now-theyre-earning-third
16. AI will lie to make the user happy, rather than tell them an unpleasant truth: https://www.theregister.com/2026/02/17/google_gemini_lie_placate_user/
17. It is possible to attack an AI in such a way that its internal model is revealed: https://www.theregister.com/2026/02/14/ai_risk_distillation_attacks/
18. Dating apps are often one of the early adopters for new technology, including AI: https://www.theguardian.com/technology/2026/feb/15/ai-dating-apps-personality-matchmaking
19. Unsurprisingly, using AI to write a letter of remorse after committing a crime tends to annoy the judge: https://www.rnz.co.nz/news/national/586871/judge-exposes-ai-generated-remorse-letters-in-michae-win-arson-sentencing
20. AI demos are easy, but getting them to be useful and contributing to the bottom line is hard: https://www.computerworld.com/article/4132756/the-ai-bubble-will-burst-for-firms-that-cant-get-beyond-demos-and-llms.html
21. AI cannot generate effective passwords: https://www.theregister.com/2026/02/18/generating_passwords_with_llms/
22. A new approach enables AI to continue learning, rather than getting stuck in one skill set: https://dataconomy.com/2026/02/18/mit-and-eth-zurich-unveil-sdft-to-stop-ai-from-forgetting-old-skills/
23. A lot of senior people in AI companies are getting out: https://www.computerworld.com/article/4133137/why-are-ai-leaders-fleeing.html
24. AI generated writing is often so, so average-it's by design: https://www.theregister.com/2026/02/16/semantic_ablation_ai_writing/