#DeFi Notes To Self
1. Most of the time, if you believe in a token, it’s probably better to buy and hold it outright than to LP, which is an implicit short vol trade where IL can get you rekt. Feels like the alpha that exists is the various iterations of LPing stable coins.
@irl_danB !!! Ok that definitely did not click for me, I think i was approaching it too “rigidly” like a regular formal language 😂 time to get freaky with this thing
@irl_danB Need to noodle on that as it’s still not “natural” to me yet to pivot to using the ** ** to talk to the VM to get it to do things. Said differently, using sessions is sort of like syntactic sugar for ** use Task Tool for X **, but directly do the latter wouldn’t “compile”?
@irl_danB It’s an interesting thought exercise since openprose makes heavy use of subagents via the Task tool as a major primitive, but it’s not clear to me if subagents themselves can invoke the TaskList tool. If they can then I agree, openprose should be able to use it “out of the box”
@irl_danB Thanks Dan! One thing that I’m curious to get your take on is whether there are gains to be had implementing the new TaskLists orchestration features that were just released in CC along with OpenProse
@irl_danB Pattern 2 is exactly what I was looking for when I was pinging you…didn’t think it would ship that quickly 😂 cheers…using the input syntax means when importing a prose program, you can essentially make it interactive by calling it without supplying the args
@irl_danB@steipete@RobertHaisfield Appreciate the quick response! I was thinking you could only use the 4th wall syntax as part of evaluating things like loop conditionals, maybe I need to check the compiler docs again to see exactly where that ** ** syntax can be used!
@irl_danB@steipete@RobertHaisfield Sorry I do not have the link, I had just snapped a screenshot when I was browsing…maybe it was indeed the outer agent?
@irl_danB@steipete@RobertHaisfield So i did some digging and it looks like the AskUserQuestionTool cannot be used with subagents, which in my understanding is triggered by each “session”, so still not sure how we’d implement a sort of robust HITL feedback mechanism 🤔
@irl_danB@steipete@RobertHaisfield Is there a way to have the agent discretionarily “break down a problem into subtasks and deploy subagents to achieve those tasks”? For example when CC would use a bunch of subagents to explore a code base “discretionarily”…does prose allow that type of expression? (2/2)
@irl_danB@steipete@RobertHaisfield Hey Dan - one question: I like how CC can currently use its discretion to subdivide a task and fan out parallel subagents to achieve it. In Prose we sort of have to explicitly specify tasks a priori and then “fan out”. 1/2
@irl_danB@steipete@irl_danB Does “human in the loop” exist in OpenProse? I’m thinking like bubble something up to the user for confirmation (ideally using the AskUserTool). The captain’s armchair examples seem to suggest some sort of human “checkpoint” but not it’s still not clear to me