https://www.darpa.mil/news/201...

My thoughts exactly. Did we run out of robot names?

I think what gamers want is both quality and originality; there's no guarantee that non-AI content is either, but there is a guarantee that anything generated by an AI tool is derivative of it's training data (so not original), and it's very likely that corners were cut in the process of reviewing the output of the machine (so not high quality). AI generated content isn't automatically bad, but there's a very high likelihood it's boring, unoriginal, and of low quality.

The difference between Generative AI and other procedural methods of asset creation are *intent*. For instance; the planets in "No Man's Sky" are generated procedurally, and that process is deterministic and carefully designed by a group of very talented people. If I wrote a new game, and prompted some neural network "Make a bunch of planets in the style of 'No Man's Sky'. Photorealistic. Correct number of teeth.", and then shipped the output, there's no intention or originality. You haven't added anything to the world, just watered down what's there.

That's true, and that's *also* not something Anthropic wants to point out

Effective at getting the word out that Claude isn't very smart -- not a message i'm sure they'd want to send

AI doesn't care about anything, it is a series of matrix multiplies. Don't personify the clankers!

I think this statement comes from the tendency laypeople have of conflating computer languages and spoken ones. "Fluency" in python does not mean the same thing as it does in Russian.

If i had a coworker that screwed up anywhere close to as often as even the best LLM, they would be fired. But somehow the magical machine gets a pass.

The parachute could be substituted for anything you want to *not fail*. I would like it if nothing failed.

one of the reasons we automate things is so that they're consistent. LLMs don't have this property.

If you told me that the latest model of your parachute fails to open 38% less often, I still don't want to take it skydiving. The failure rate of LLMs is so high that any productivity gains you get generating some output are lost verifying that output.

So how many times does it do the wrong thing before you stop using the tool?

Then they've just gotten money to lose from another source, it's still a big problem

Amazon is, by and large, a streaming company, who wants you to pay them for content. By not allowing you to stream pirated content, they don't have to compete with it.

So yes, they very much care what you stream.

Why would engineers need to be good at marketing? Isn't that what marketing people are for?

I'm as anti-AI as the next guy, but in this case does it really matter?