Weekly Review 17 October 2025

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

1. Is AI going to make composers obsolete? https://www.theguardian.com/music/2025/oct/09/classical-music-and-ai-by-tarik-oregan-composer-radio-3
2. I'm still skeptical that AI could do the kind of things I need to use Excel for: https://dataconomy.com/2025/10/07/excel-gets-ai-agent-mode-for-automated-data-tasks/
3. AI is a great help for good honest work avoidance: https://www.theregister.com/2025/10/06/at_last_microsoft_leads_the/
4. AI will destroy 97M jobs in the USA in the next ten years - according to AI: https://www.theregister.com/2025/10/06/ai_job_losses_us_senate_report/
5. Agentic AI will have a huge impact on businesses: https://www.informationweek.com/machine-learning-ai/from-data-to-doing-agentic-ai-will-revolutionize-the-enterprise
6. Very soon we're going to see AI agents buying products that are recommended by AI. People won't be involved at all in spending their money: https://techcrunch.com/2025/10/07/openai-and-the-race-for-ai-driven-commerce/
7. Using AI to monitor heartbeats using Wi-Fi signals: https://spectrum.ieee.org/wi-fi-signal-heartbeat-detection
8. Awful people use AI to generate misinformation about a missing child: https://www.rnz.co.nz/news/world/575620/ai-generated-images-of-missing-4yo-gus-raise-legal-concerns-about-misinformation
9. Deloitte Australia has to refund the Australian government after submitting a report full of AI-generated hallucinations: https://arstechnica.com/ai/2025/10/deloitte-will-refund-australian-government-for-ai-hallucination-filled-report/
10. Google AI bug fixer: https://www.theregister.com/2025/10/07/google_deepmind_patches_holes/
11. Small language model AI can be run on consumer devices and don't need data centres: https://spectrum.ieee.org/small-language-models
12. Even though Deloitte's use of AI caused them to refund the Australian government, they are still going all-in on it: https://techcrunch.com/2025/10/06/deloitte-goes-all-in-on-ai-despite-having-to-issue-a-hefty-refund-for-use-of-ai/
13. Entry-level jobs are being changed by AI, but not eliminated by it: https://www.computerworld.com/article/4067751/aws-training-exec-genai-isnt-killing-entry-level-jobs-its-redefining-them.html
14. Google won't fix a vulnerability in its AI: https://www.extremetech.com/computing/google-declines-to-fix-gemini-flaw-that-lets-hidden-commands-trick-the
15. AI improve faster at tasks where their performance can be easily measured: https://techcrunch.com/2025/10/05/the-reinforcement-gap-or-why-some-ai-skills-improve-faster-than-others/
16. Startups adopt AI much faster than established companies: https://www.theregister.com/2025/10/03/startups_binge_on_ai/
17. How much of the slowdown in entry-level hiring is due to AI? https://www.theguardian.com/money/2025/oct/09/gen-z-face-job-pocalypse-as-global-firms-prioritise-ai-over-new-hires-report-says
18. It's now very difficult to get funding from venture capitalists if you're not pitching AI: https://techcrunch.com/2025/10/04/if-youre-not-an-ai-startup-good-luck-raising-money-from-vcs/
19. AI in healthcare can be useful for communicating with patients: https://www.bigdatawire.com/2025/10/03/goldman-sachs-chief-data-officer-warns-ai-has-already-run-out-of-data/
20. While the dead might not have legal rights to their image, it's still tasteless to use AI to create videos of them: https://techcrunch.com/2025/10/07/you-cant-libel-the-dead-but-that-doesnt-mean-you-should-deepfake-them/
21. Accounts using AI for evil are starting to get banned: https://www.theregister.com/2025/10/07/openai_bans_suspected_china_accounts/
22. Optical neural networks are a much more energy-efficient approach to AI: https://spectrum.ieee.org/generative-optical-ai-nature-ucla
23. The more you interact with sycophantic AI the worse you get at dealing with conflict with people: https://www.theregister.com/2025/10/05/ai_models_flatter_users_worse_confilict/
24. More warnings on the AI bubble bursting: https://www.theguardian.com/business/2025/oct/08/bank-of-england-warns-of-growing-risk-that-ai-bubble-could-burst
25. An emerging standard for connecting AI to IDE: https://www.theregister.com/2025/10/07/jetbrains_acp_vs_code/
26. AI detectors don't really work for detecting AI-generated text: https://www.rnz.co.nz/news/world/575416/university-wrongly-accuses-students-of-using-artificial-intelligence-to-cheat
27. AI is starting to replace the relationships we have with other people and creatures: https://www.computerworld.com/article/4067326/an-unwelcome-megatrend-ai-that-replaces-family-friends-and-pets.html
28. The big winner of all the AI investment is Nvidia: https://www.theregister.com/2025/10/06/stargate_openai_amd/
29. While people fear losing their jobs to AI, the real threat is AI-designed biological agents: https://arstechnica.com/science/2025/10/do-ai-designed-proteins-create-a-biosecurity-vulnerability/
30. Data centres for AI demand so much copper that it's driving new ways of extracting the metal: https://spectrum.ieee.org/ai-copper
31. What needs to be considered when rolling out AI at financial institutions: https://www.bigdatawire.com/2025/10/08/ai-is-everywhere-scaling-it-in-finance-requires-deeper-responsibility/
32. Google's AI browser can now act like a human to get data hidden from bots: https://www.theverge.com/news/795463/google-computer-use-gemini-ai-model-agents
33. Will the AI bubble bursting impact retirement funds? https://www.stuff.co.nz/money/360850918/ai-bubble-warning-how-worried-should-you-be-about-your-investments-and-kiwisaver
34. AI have now exploited all of the publicly available data, the future lies in the confidential data hidden away within organisations: https://www.bigdatawire.com/2025/10/03/goldman-sachs-chief-data-officer-warns-ai-has-already-run-out-of-data/
35. Google will pay bug bounties on its AI, but not for defeating guardrails: https://www.theregister.com/2025/10/07/google_ai_bug_bounty/
36. Will the original creators of the data scraped to train AI ever get paid for it? https://www.theregister.com/2025/10/06/openai_makes_empty_promises_to/
37. The more researchers use AI, the less impressed with it they become: https://www.theregister.com/2025/10/08/more_researchers_use_ai_few_confident/
38. The societal dangers of AI generated girlfriends: https://www.theguardian.com/technology/2025/oct/06/rise-of-ai-girlfriends-adult-dating-websites
39. AI is a threat to content creators: https://techcrunch.com/2025/10/06/mrbeast-says-ai-could-threaten-creators-livelihoods-calling-it-scary-times-for-the-industry/