@chamath Anthropic already released the proper use of expensive SOTA models via Advisor architecture. Just need to create the harnessing to use it with open models
@abhishekn the bottleneck isn’t really personalized drills or worksheets, it’s personalized instruction. Without a screen, you’ll need a human or the agent has a voice for verbal lessons but it won’t work for other modalities (no screen hard rule)
@ericzakariasson seems I can do this by selecting an element then adding it to the chat, then selecting another then add to chat, repeat. So maybe not needed, but if you save me some clicks that would be cool too
@ericzakariasson want to be able to select multiple non-contiguous elements in design mode. Can get by with selecting a big area and calling out each element in the prompt but this would give more control
@levelsio Wondering if instead of letting people download your compiled extension code, you distill your prompts into an idea file (credit karpathy) and people can prompt their agents to build it
Might be more secure way to share tools in the future as ai gets more and more capable
@daniel_mac8 it's socratic mode sprinkled with systems thinking frameworks like Rumsfeld matrix. You can probably tell it to act like a consultant from the Big 3 consulting firms
I’ve been building an interactive narrative game where you type what do next and the story generates forward from ur choices in realtime. Think interactive comic book, no predetermined paths.
You have to beat this guy 3 times to win. Reply below with what you would do next