Yeah, yeah, can you shut the fuck up? I work in a successful LLM startup. More successful than OpenAI and Anthropic if you measure by profit and not ability to con people into investing hundreds of billions into developing the promised machine god. We train models for specific use cases and rent them out. I know better than anyone there are real uses. But "use Blender and make a slopshit model that any 3D artist would be fired for if they presented to their boss", ditto the slop videos, is not it. It's fucking sickening listening to the way people talk about these models.
The trendline is clear, but not in the way you think it is. Scaling vertically has basically stopped working already. Generalization plateaued more than a year ago. Since then the gains have been training models for iterating in a harness, training in objective reward-verified domains like math, training in specific tasks. All of which are useful, and which the field of ML has been demonstrating the unreasonable effectiveness of for at least 30 years when DeepBlue beat Kasparov. But code quality has stopped meaningfully improving; what actually improved is the ability to fix compile errors in a loop, giving you the worst code you've ever seen that does compile. If the trendline were significantly upwards, we would be seeing something approximating code that isn't absolute dogshit. But that's not what we're seeing at all.
Where? I can't find one company that did "AI layoffs" and their app/services got better. Give me just one good example please
I'm by no means citing that AI is perfect, or that there are no problems or challenges ahead, but the self-lobotomizing AI doomerism/rejection is turning otherwise intelligent people into a horde of anti-AI lemmings that are afraid to apply the critical thinking they believe is being lost by AI users.
These things are puppeteering blender to model things for gods sake. Are you all blind?