Something strange just happened in AI.
The man who invented RLHF (the exact technique that made ChatGPT feel human) just turned his back on his own creation.
Diogo Almeida helped build ChatGPT at OpenAI. He co-invented reinforcement learning from human feedback, arguably the single most important breakthrough behind this entire AI boom.
And yet, he wasn't happy.
His words: 'We have lightning in a bottle, and yet it is not useful.'
Read that again. The guy who caught the lightning is telling us the bottle is broken.
For four years, the entire industry optimized AI models to master human language. Sound smart, talk smooth, mimic us perfectly.
But Almeida's realization was brutal: talking like a human isn't the same as being useful to one.
So he built something different. A new kind of model, based on a completely different philosophy, and developers are already buzzing about it.
Not an upgrade. A rethink.
If the person who literally invented the 'secret sauce' says it wasn't enough... how far are we really from AI that's actually useful, versus AI that just sounds impressive?
🔗 Original Source & Reference: https://techcrunch.com/2026/09/18/a-new-kind-of-ai-model-from-a-chatgpt-inventor-is-thrilling-developers/
Published automatically via FeedMind AI Content Pipeline.
Top comments (0)