overrated/counterproductive AI risks: - copyright - resource usage - sophistry around “inherent” limitations proven wrong a month later underrated (in order of immediacy): - surveillance acceleration - persuasion (incl delusion) acceleration - existential risk

Show Post

i will never not get over seeing my name under the “europa-uvs team” 🥹 like what do you mean i’m on an instrument team for this cool ass spacecraft that’s on its way to jupiter right now!!!

Show Post

To put this into context, the amount of water required to produce an 8oz steak is 3,217,000 ml. So you would need to make 189,000 queries to equal the water cost of a steak dinner. 4/

Show Post

But as the authors state in their paper, the amount that ChatGPT needs to consume is around 500ml of water *for 10-50 queries*. This means that each query is about 500/30=17ml. 3/

Show Post

Calculating the water consumption of AI is complicated, as you need to account for both training and testing, and because a single task might require multiple queries of different lengths, on different hardware, and so forth. 2/

Show Post