Browse Comments — Clean (de-noised)

They control all of this, and this is all fear mongering to impose regulations on their competitors and secure their spot at the top.

Before or after the IPO?

This company is good in marketing with shitty product

Scott Schobert Totally agree! This is a major concern that is surfacing right now. The hype and leadership reaction to AI first has made the gap almost unmeasurable. A pause to reevaluate is necessary.

Literal Asimov cascade

Currently this will not be a big issue - the current infrastrucure is clumsy and easily spoted by other AI engines. However when AI jumps to Quantum computing then AI starts to process similarly to the human brain. 2 more years and the chnage will be similar to going from a flip phone to a Smart phone.

Ah shut up..bored of this shht already

Shipping 8x more code is just faster interpolation, not true recursive self-improvement. Current models excel at pattern matching but lack the zero-to-one reasoning needed to invent new architectures. The "infinite loop" myth ignores the fact that models training on their own synthetic data face inevitable model collapse. AI agents operate strictly within human-defined loss functions and cannot engineer their way past physical hardware limits. Framing scaling milestones as an existential threat is a great way to build a regulatory moat, but it ignores technical reality.

Razi Miliani that’s it

Tamara Tate

He is Italian

He warns us and yet still makes it....it's like he's saying yeah this is bad for human kind but I love the money 💰

Lol, who will build the archotecture that allows that? sounds like a rick and morty episode 🤣

It will start raising Gummo Babies.

Honestly the SynthID news is the one I'd underline twice, Demis. Watermarking is useless if it's only one lab doing it, getting OpenAI, Kakao and ElevenLabs to actually adopt the same standard is the harder political win. PS — would love to connect.

Unfortunately, he is betting both ways to cover his ass. What a pity. GenAI Works

We need more conversations like this

I understand that it is less that AI decides to build AI, but that human use it to build and rebuild AI.

The pace of task completion doubling every four months is the stat that sticks with me, hard to wrap my head around what that looks like even a couple years from now.

My throughput and output have become unrecognizably large compared to pre-coding assistants, like codex and claude code.