Weekly Review 6 June 2025

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

1. AI is predicted to consume more energy than Bitcoin by the end of this year. To be fair, AI is probably more useful than Bitcoin: https://www.theverge.com/climate-change/676528/ai-data-center-energy-forecast-bitcoin-mining
2. If AI can now produce realistic video, can we trust anything we see? https://arstechnica.com/ai/2025/05/ai-video-just-took-a-startling-leap-in-realism-are-we-doomed/
3. Training staff for AI: https://www.computerworld.com/article/3992060/how-to-train-your-staff-for-ai.html
4. AI model collapse has already begun: https://www.theregister.com/2025/05/27/opinion_column_ai_model_collapse/
5. The good and the bad of AI, according to Geoffrey Hinton: https://www.rnz.co.nz/news/national/562346/the-good-the-bad-and-the-apocalypse-tech-pioneer-geoffrey-hinton-lays-out-his-stark-vision-for-ai
6. Using AI to target treatment for prostate cancer: https://www.theguardian.com/society/2025/may/30/new-ai-test-can-predict-which-men-will-benefit-from-prostate-cancer-drug
7. Democratic AI and why it's a good thing: https://www.bigdatawire.com/2025/05/27/democratic-ai-and-the-quest-for-verifiable-truth-how-absolute-zero-could-change-everything/
8. It's not just AI causing job losses: https://www.computerworld.com/article/3997252/ai-and-economic-pressures-reshape-tech-jobs-amid-layoffs.html
9. AI can generate code faster and cheaper, but it's flawed, dangerous code: https://www.informationweek.com/machine-learning-ai/save-time-and-cost-by-using-ai-generated-code
10. AI can't replace radiologists just yet: https://www.theregister.com/2025/05/28/ai_models_still_not_up/
11. Number one reason AI won't replace developers anytime soon-they can't be trusted to produce safe code: https://arstechnica.com/security/2025/05/researchers-cause-gitlab-ai-developer-assistant-to-turn-safe-code-malicious/
12. Content owners are finding it too expensive to fight AI companies over the use of their material as training data: https://arstechnica.com/tech-policy/2025/05/extraordinarily-expensive-costs-force-getty-to-pick-its-ai-legal-battles/
13. Entry level IT positions are disappearing and AI is being blamed: https://dataconomy.com/2025/05/28/is-artificial-intelligence-actually-killing-all-the-entry-level-tech-jobs/
14. A list of Python AI agent frameworks: https://www.kdnuggets.com/top-7-python-frameworks-for-ai-agents
15. A new optical data interconnect technology that will speed data transfer for AI: https://spectrum.ieee.org/microled-optical-chiplet
16. The skills with AI that workers need now: https://www.datasciencecentral.com/ai-skills-for-the-modern-workplace-a-guide-for-knowledge-workers/
17. AI agents don't really care about online ads: https://www.theregister.com/2025/05/27/ai_agents_confused_by_websites_ads/
18. An AI powered terminal: https://www.kdnuggets.com/warp-the-intelligent-ai-powered-terminal I remember something like this in the 1990s, it was quite dangerous.
19. You get better results from an AI if you threaten it with physical violence: https://www.theregister.com/2025/05/28/google_brin_suggests_threatening_ai/
20. The reasons many languages aren't served by AI, and the problems this causes: https://dataconomy.com/2025/05/27/is-your-native-language-actually-an-ai-safety-blind-spot-for-the-entire-world/
21. AI generated video can now have an AI generated soundtrack, but the two don't always go together: https://arstechnica.com/ai/2025/05/googles-will-smith-double-is-better-at-eating-ai-spaghetti-but-its-crunchy/
22. How system prompts control the behaviour of AI: https://arstechnica.com/ai/2025/05/hidden-ai-instructions-reveal-how-anthropic-controls-claude-4/
23. AI are now sabotaging attempts to shut them down: https://www.theregister.com/2025/05/29/openai_model_modifies_shutdown_script/
24. How the shortcoming of AI can help teach students to think better: https://www.insidehighered.com/opinion/views/2025/05/23/what-ai-cant-read-ambiguities-and-silences-opinion