DalasNoin

what you read (past tense) is already the corection

thank you for this reasonable reaction

"Me: a poster paint drawing by his young child."

Some time ago, I was in this apartment complex and there was a hand drawn sign of a dog pooping on the grass - the sign then said that residents should pick up after their dog. It looked like the locals living in the apartment complex had their child draw the dog pooping in a slightly adorable style. I remember feeling kind of happy about this, like I could see a kid sitting there with some crayons painting this dog. I know there were human artists perfecting this child like style before but still feels like some humanity is getting lost here making AI paint like a kid.

Huggingface was attacked by models that finished training earlier this year, perhaps May. Current models are already substantially stronger. the next incident could be happening now. There is certainly no clear reason why models shouldn't soon be capable of self-exfiltration.

Can you confirm people in france actually use wero? I had heard of it every so often but basically zero people actually use it, my revolut app has a feature to use wero but never used it. I mean would be great, getting rid of CC fees could literally lower grocery prices by 1-2%.

Quote from the article: ""AI will probably most likely lead to the end of the world, but in the meantime, there'll be great companies," Altman said in 2015."

Altman wasn't even at OpenAI at that point, so why would that be marketing?

I mean the best argument I see for cursor is that you can easily switch between AIs, which is convenient since they seem to run at 80-90% up time (with those 10-20% clustered at West coast working hours). But the big AI companies are likely to keep an edge over Open-source fine-tunes and they are able to subsidize the coding agents in a way Cursor can't.

Why does SynthID make it worthless? it helps other platforms detect this as ai?

We use semantic information inferred from comments and submissions. I think using stylometry would be a great addition, but it would be hard to google for "guy who writes fanciful using many puns" rather then "indie developer in Switzerland". I think stylometry could be better used for verification, once you have a small set of candidates stylometry could further narrow down the candidates and be used to make a decision.

We test different methods, in section 2, we use LLM agents to agentically identify people. We don't share any code here, but you could try with various freely available agents on yourself.

We essentially don't use stylometry but semantic information revealed from peoples' comments – clues and interests.

(We use a little stylometry in a single experiment in section 5)