Weekly Review 23 August 2024

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

1. Large-scale AI needs large-scale nuclear power, but there are regulatory issues with getting the electricity from the power plants to the data centres: https://spectrum.ieee.org/amazon-data-center-nuclear-power 
2. Is the collective noun of AI driven taxis a "Gaggle" or a "Honk"? https://futurism.com/the-byte/robotaxis-gather-honk-all-night 
3. As generative AI gets better and better, how long before these AI-generated stories become impossible to detect? https://www.huffpost.com/entry/cody-enterprise-reporter-resigns-after-using-artificial-intelligence_n_66bcc610e4b03da4fc01b2d3 
4. Orwell got it wrong, it's not Big Brother who's watching you, it's AI: https://www.leightonassociates.co.nz/post/ai-is-watching-you-all-shift-long 
5. Using AI to decode the signals from a brain implant, to allow people to communicate again: https://dataconomy.com/2024/08/15/ai-helps-als-patient-speak-again/ 
6. Blaming remote work for Google falling behind in AI is like blaming a soldier's boots for losing a battle - ultimately, responsibility lies with the leaders: https://www.theregister.com/2024/08/15/googles_exceo_steps_back_from/
7. AI is like any other product, if it doesn't function, or isn't useful, then it's pointless: https://futurism.com/the-byte/google-demo-gemini-ai-tech-fail 
8. This technique might slightly reduce AI hallucinations, but it doesn't solve the underlying problem, which is an AI has no concept of what the words it spits out actually represent: https://www.computerworld.com/article/3487262/researchers-tackle-ai-fact-checking-failures-with-new-llm-training-technique.html 
9. You can't just add more AI to try to correct an AI that is broken because of bad data: https://www.computerworld.com/article/3487242/agentic-rag-ai-more-marketing-hype-than-tech-advance.html 
10. How to get started in a position training AI: https://dataconomy.com/2024/08/15/get-started-with-ai-training-jobs/ 
11. This newspaper doesn't even try to hide the fact its articles are generated by AI: https://futurism.com/entirely-ai-generated-news-site 
12. A new licensing deal means that actors will get a say in what their AI clones say and do: https://www.theregister.com/2024/08/15/actors_union_ai_voice_clone/ 
13. I wonder how many of these AI generated job applications are being filtered by AI? https://futurism.com/the-byte/recruiters-ai-generated-job-cvs 
14. How AI can automate the process of developing a brain connectome: https://www.extremetech.com/science/mit-scientists-detangle-the-brain-with-new-open-source-ai 
15. Replacing writers with AI without telling the editors - the long, slow decline towards AI generated mediocrity continues:  https://www.theregister.com/2024/08/15/robot_took_my_job/ 
16. More AI-based interference in elections, this time by Iran: https://www.theverge.com/2024/8/16/24221982/openai-iranian-chatgpt-accounts-banned-chatgpt-us-election 
17. Achieving general AI is not a matter to throwing more and more processing power at the problem, there needs to be a fundamental shift in the model used: https://futurism.com/the-byte/agi-supercomputer 
18. When an AI can change the parameters of its own execution, we need strong guardrails: https://arstechnica.com/information-technology/2024/08/research-ai-model-unexpectedly-modified-its-own-code-to-extend-runtime/ 
19. The AI bubble seems to be following the same trajectory as the Internet bubble of the late 90s. I expect what will happen is a lot of AI companies will go under, and the ones that provide useful services will survive: https://www.datanami.com/2024/08/14/is-the-genai-bubble-finally-popping/ 
20. AI technology will get to the point that it is useful in high school level teaching, but we're not there yet: https://futurism.com/high-school-starts-replacing-teachers-ai 
21. I don't think it's a good idea for companies to be driving changes to laws, especially laws that are supposed to protect the public from the problems that AI can cause: https://www.theregister.com/2024/08/16/california_ai_safety_bill/ 
22. This approach of stealing all the data they can get and training their AI on it has to stop, it's legally questionable and does not produce good AI: https://www.theverge.com/2024/8/14/24220658/google-eric-schmidt-stanford-talk-ai-startups-openai 
23. This is the kind of AI that is likely to survive the AI bubble bursting - it's fairly simple, straightforward, and useful: https://www.computerworld.com/article/3485725/ai-and-ar-can-supercharge-ambient-computing.html