i realized something recently
bernie sanders and AOC have damaged the fabric of america immensely but we rarely talk about it
bernie/AOC have vilified successful businessmen & women. calling CEOs like elon musk, bezos, zuck evil. yes these people have shortcomings - but they are also truly self made, come from normal/humble backgrounds. with hard work they created trillions in economic value. isn’t that a fundamental part of what america is about? the land of opportunity?
in the 2000s when i was in elementary school, bill gates and steve jobs were viewed as role models. our teachers taught us about how much they achieved + given to society. these CEOs were actually respected and i believe this culture brought up a generation of ambitious hard working americans
now, kids are being taught that working hard and becoming a billionaire is evil. we’re essentially telling them to have victim mindset and stay poor
i know bernie/AOC & crew are never going to take responsibility. i can’t even tell if they realize what they’re doing
agree. tldr just use it for research or to synthesize differing povs into what the best argument is. this is helpful for achitectural decisions and getting rid of little quirks in the models (usually opus). https://t.co/AdXMjnxwck
> What do you use subagents for?
many things, but my favorite:
the good old fan-out-fan-in
and I think there is more to this than "you can parallelize token spraying" (which... is fun, but... careful)
rather, the more important fan-out pattern is one in which each branch (subagent) accumulates experience, and then the fan-in synthesizes this experience into condensed learnings
what do I mean by experience?
it's conventional wisdom by now that the models do better when they have back-pressure to spew their tokens against
the thinking goes: if you're using agents to one-shot something, chances are it may be wrong on the first try. but if you give them some back-pressure--say, tests that they can run against the real world and whose results they can observe--their outputs converge on something more accurate
and it's not just parallelizing unit tests... any experiential "theory meets reality" observation of the world rolls up into this category. the one that emerges most often in my own usage is parallel research and synthesis
so it's not only interesting to parallelize work to just generate more tokens, it's interesting to parallelize work because you can accumulate experience faster
the fan-out-fan-in is an efficient empirical learning pattern
imagine splitting yourself to parallelize your lived experience into a sort of multiverse reality all of which you remember after your shards re-converge with learnings in tow
@paulg every single reply bot i've seen either tries to deliver some sort of punch or reveal a "genius" hidden flaw that isn't actually consequential when given any sort of real thought.
@changloria0816 only issue i'd see with this is it's not really clear when implied or less exact instructions in the CLAUDE.md are being ignored vs explicit ones.
The events of the last 6 months in technology are arguable amongst the most important in human history
The tools now increasingly exist for recursive self improvement of models & agents
We are likely in very early lift off & exponential
Largely unnoticed outside of tech
i have seen enough proof now that using a coding agent is a deep skill
it's confusing because the people you see heavily using them produce horrible results
but that's because it's a skill! you can get better and the ceiling seems pretty high - this is very exciting to me
I just checked with my agent and it recommended Pinecone, Weaviate, Milvus, Qdrant, and Chroma as options. I assume if somebody's building this with an agent, they'd probably ask the agent more about architectural decisions, so if the agent doesn't recommend Helix, this probably hurts a lot too.
I didn't find it too difficult to use if I did it myself, but my agent really struggled with the helix query language and keeping track of all of the rules/syntax as well as the commands. It didn't understand how the backups worked, often left me with deleted data.mdb files, and forgot to check what was running in docker, which led to it being difficult to manage multiple helix instances running at once. It was lowkey a pain to work with, but I assume a good skill would fix this.