Andrew Ng co-founded Google Brain and taught half this industry its first AI course:
“Prompting is already dead.
Self-improving agents and loops are the only thing that matters now”
He shows why most people are still stuck on the surface while the real game moved to agents that finish the work themselves
Prompting → Agents → Loops → Self-Improving Systems
The first 20 minutes go past where most $700 AI courses stop
Most people are still writing prompts while the leverage moved two layers up
Same models, same tokens, and the only thing that changes is the system you run them in
Save this. Then find any lecture that interests you thanks to the article below ⭣
Really cool work from the @GeYan_21 here.
A robot fixing its own gripper, ending with a screw going into a hole with barely any room to spare, and it gets there most of the time when other models barely could. That alone is worth a look.
But the bigger idea is what stands out. Instead of picking between a fast action model and a slower one that thinks ahead, this is one model that can do both, and you choose which at the moment you need it. No retraining, same weights.