Discussion about this post

User's avatar
Simon Kearney's avatar

One thing I forgot to add this week was the release of the second AI safety report which seemed to miss my attention last week when it was released. The report, written by 100 AI safety experts, has a bet each way on almost everything. Much as Erik says in his item this week we just don't know where things are headed and it's impossible to keep up. AI capabilities could plateau or keep marching ahead Computer Weekly says in its coverage of the report. The most interesting finding was about how we humans and AI are a mutual appreciation society. AI models exhibit sycophancy, the report says, which I've certainly noticed. ChatGPT gives me the best feedback on me being me that I've ever had in my life. But it turns out we have 'automation' bias and we tend to be overly agreeable back to the AI. Of course we are, we all have one eye on the future!

No posts

Ready for more?