Ezra klein 80,000 hours
@AI Listened to Ezra Klein on the 80,000hrs podcast this morning. He makes the point that we can’t ignore incentive strcutures to promote ‘good’ AI. If money can be made doing some bullshit quant trading that could lead to a radically misaligned AI, or some advertising optimization we need to be aware of it. Especially, if something like AlphaFold made no money, how would we just expect people out of the goodness of their heart to push against the inertia to create AlphaFolds.
He also mentions how the conversations should be centered more on what we want AI to do for us, its another kind of ‘imaginging a bright future’ that we really don’t do.
I was struck by the notion that books you’ve never heard of influence policy. That ‘thinktanks’ may release a report thats only read by 5 people but it depends who those 5 people are. Ezra said he surprised by the amount of AI development going on but not making relationships with policy makers for AI safety.
Kevin Kelly
- To bits of advice that struck me was the idea of not being afraid to throw things away. Create a prototype of something fully OK with the notion that it’ll be thrown away. It allows you to focus more on the questions you’re trying to answer and not precious about the generation of solutions.
- He talks about get the cheapest tool and through use you’ll realise why you need a more expensive one.
Model thinking or the ‘thinking like an economist’ kind of merges with Stats rethinking. Any scientific endevedaour starts with a theory.
Research
- Adding structure allows easier reading of research papers as you can file the information your receiving according to the structure of the document you’re trying to write.
- Paper can be thought of always as a response to a problem.
- He highlights staying n that zone of discomfort in terms of ‘writing block’ try and keep hold of the problem, even if it wont budge. Similar thoughts to AM I feel.
Tough going today, I don’t feel I understand anything. I’m trying to relate uncertainty with regards to LCA’s. This involves defining the overarching sources of uncertainty in their qualitative form. Maybe it makes no sense to try and ‘get formal’ with it.
reading: Wolff and Duffy, ‘Development and Demonstration of an Uncertainty Management Methodology for Life Cycle Assessment in a Tiered-Hybrid Case Study of an Irish Apartment Development’. whats good is that it cuts right to the chase and has the overlap of model and lca.