Weekly Review 27 September 2024

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

1. This is the real danger with AI, it can be used to supercharge surveillance, and big tech companies can make a lot of money doing so: https://www.theregister.com/2024/09/16/oracle_ai_mass_surveillance_cloud/
2. In addition to needing a lot of energy, AI also need a lot of water for cooling: https://www.techrepublic.com/article/generative-ai-data-center-water-use/
3. Using AI to eliminate bias in hiring. Except the AI inherits the biases in the data used to train it: https://www.datasciencecentral.com/how-machine-learning-is-delivering-next-generation-talent-acquisition-processes/
4. Are the EU regulations really stifling development of AI? Or is this just big tech companies whinging? https://www.computerworld.com/article/3535551/eu-privacy-regulation-stifling-ai-innovation-claims-open-letter-masterminded-by-meta.html
5. OpenAI really doesn't want people testing their new AI too hard. What are they hiding? https://arstechnica.com/information-technology/2024/09/openai-threatens-bans-for-probing-new-ai-models-reasoning-process/
6. AI needs a lot of energy, and it's contributing to climate change. But AI can also optimise renewable energy generation:  https://www.informationweek.com/sustainability/the-ai-power-paradox
7. Microsoft is teaming with a power company to recommission a nuclear reactor, to generate the energy needed for AI computation: https://www.nextplatform.com/2024/09/20/ai-powerhouses-choose-the-nuclear-option/
8. AI is good at making predictions, which is useful in supply chain management: https://www.datasciencecentral.com/role-of-ai-predictive-analytics-in-supply-chain-management/
9. Ways in which AI can make a meaningful impact on a business: https://www.datasciencecentral.com/how-to-transform-your-business-digitally-with-ai/
10. AI needs a lot of energy, and currently that's being met with fossil fuels. But AI can also help boost renewable generation, so I don't think it will derail things: https://www.informationweek.com/sustainability/will-future-ai-demands-derail-sustainable-energy-initiatives-
11. Microsoft scraped data from LinkedIn to train their AI, and updated their terms of service to allow it after they did it: https://www.theregister.com/2024/09/19/linkedin_ai_data_access/
12. An expanded context window for a conversational AI. So while it won't forget what you're chatting about, it will still hallucinate: https://spectrum.ieee.org/ai-context-window
13. A social network app populated entirely by AI: https://arstechnica.com/information-technology/2024/09/dead-internet-theory-comes-to-life-with-new-ai-powered-social-media-app/
14. Can quantum computers provide the computing power needed for AI? https://www.datanami.com/2024/09/17/quantum-and-ai-navigating-the-resource-challenge/
15. California bans election-related deepfakes near election time. Will it be enforced? https://www.informationweek.com/it-sectors/california-s-new-deepfake-laws-await-test-of-enforcement
16. Five ways to screw up an AI project: https://www.informationweek.com/machine-learning-ai/how-to-avoid-common-ai-newbie-mistakes
17. The UN releases a plan for how AI can help humanity: https://www.computerworld.com/article/3532360/un-lays-out-plans-for-how-ai-can-best-serve-humanity.html
18. Too big, too small, size does matter after all. At least when AI are concerned: https://www.computerworld.com/article/3529501/smaller-genai-models-for-every-app-might-be-the-future.html
19. An open call for questions that could stump an AI: https://futurism.com/the-byte/humanitys-last-exam-ai-benchmarks
20. An AI that can detect misinformation in social media: https://spectrum.ieee.org/fight-misinformation
21. Google is going to start labelling AI generated images in its search results: https://www.techrepublic.com/article/google-c2pa-generative-ai-search/
22. A portable AI controller for robots: https://spectrum.ieee.org/machine-learning-and-robotics You want SkyNet? Because this is how you get SkyNet
23. No guardrail setup can stop all the harms an AI could cause: https://www.theregister.com/2024/09/17/ai_models_guardrail_feature/