Weekly Review 26 July 2024

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

1. The new British government plans to regulate some, but not all, AI development: https://www.computerworld.com/article/2520506/new-uk-government-downplays-ai-regulation-in-program-for-the-next-year.html
2. Heartening that ethical AI leads this list of AI trends: https://dataconomy.com/2024/07/16/ai-and-machine-learning-trends-to-follow/
3. Some ways AI is used to detect credit card fraud: https://www.datasciencecentral.com/how-machine-learning-detects-credit-card-fraud/
4. OpenAI has closed one of the holes that allows misuse of ChatGPT: https://www.theverge.com/2024/7/19/24201414/openai-chatgpt-gpt-4o-prompt-injection-instruction-hierarchy
5. A bust in AI might knock Nvidia's business, but I doubt it will destroy them: https://www.tomshardware.com/tech-industry/nvidias-business-model-may-fall-apart-if-corporations-cant-make-ai-pay-says-sk-group-boss
6. How you can become an AI engineer, and why you should: https://dataconomy.com/2024/07/17/how-to-become-an-ai-engineer/
7. Proton mail, renowned for their commitment to privacy, is deploying an AI assistant. It will be available as a local model, to further support privacy: https://techcrunch.com/2024/07/18/proton-launches-privacy-first-ai-writing-assistant-for-email-that-runs-on-device/
8. An AI tool that helps GPs detect cancer faster and earlier: https://www.theguardian.com/society/article/2024/jul/21/gps-use-ai-to-boost-cancer-detection-rates-in-england-by-8
9. A list of domestic devices that run AI: https://dataconomy.com/2024/07/16/best-ai-powered-home-devices-2024/
10. I feel that it would be quite ironic if the first effective guardrails around AI were developed to stop them saying something that the CCP doesn't like: https://dataconomy.com/2024/07/18/china-ai-governance-core-socialist-value/
11. Comparing generative, analytical, causal and autonomous AI: https://www.datasciencecentral.com/synergy-of-generative-analytical-causal-and-autonomous-ai/
12. I'm still skeptical. With most organisations, rather than using the productivity gains of AI to allow staff to have four-day weeks, they'd just cut 25% of staff: https://www.computerworld.com/article/2514527/is-ai-the-secret-sauce-for-the-four-day-workweek.html
13. More evidence that the big AI companies have been wantonly scraping data any place they can get it to train their models: https://techcrunch.com/2024/07/18/meta-suspends-generative-ai-tools-in-brazil/
14. Google says AI is not responsible for increasing carbon emissions: https://www.theregister.com/2024/07/18/google_emissions/
15. This was an interesting case-the claim wasn't that the data used to train the AI was violating copyright, but that the AI was reproducing code without the correct attributions, which violated the license: https://news.bloomberglaw.com/ip-law/openai-dropped-from-first-ever-ai-programming-copyright-lawsuit
16. I suspect the success of this data rights broker will depend on the outcome of court cases that will establish if using data to train AI is fair use or not: https://www.computerworld.com/article/2519026/platform-lets-creators-monetize-their-content-for-use-in-llm-training.html
17. I'm somewhat heartened that at least some acknowledge that the biases inherent in AI training data will cause issues with its use in law enforcement: https://www.informationweek.com/machine-learning-ai/law-enforcement-eyes-ai
18. Predictions of continued shortages in supply of the chips used for AI: https://www.theregister.com/2024/07/18/tsmc_ceo_predicts_ai_chip/
19. A specialised large language model to bring AI to spreadsheets: https://www.techrepublic.com/article/microsoft-spreadsheetllm/
20. Where large language model AI came from, and where they are heading: https://www.datasciencecentral.com/history-and-future-of-llms/
21. An overview of ChatGPT and the tools that available for working with it: https://www.techrepublic.com/article/chatgpt-cheat-sheet/
22. Test Time Training, a way of improving the efficiency of generative AI: https://techcrunch.com/2024/07/17/ttt-models-might-be-the-next-frontier-in-generative-ai/