Yeah. I also think the current architecture of LLMS has stalled. Im not too upset about it tho. The world needs to digest the giant leaps we've made for a minute anway imo.
You can't "vibe code" an app like nostr:nprofile1qqs83nn04fezvsu89p8xg7axjwye2u67errat3dx2um725fs7qnrqlgzqtdq0
I used to have doubts, now I'm 100% certain.
I say this after having created #purplestack with that idea. (It still can do smaller apps right)
These agents/LLMs are just not good enough. With the latest models including GPT-5, progress on this front appears to have stalled.
Medium-term bearish on AI systems not involving regular human input.
Discussion
I'm fine with it stalling too. Time to adjust expectations.
What I really want them to introduce is the bullshit detector. These things constantly lie even though they have the answer in their context window.