findjashua

Would it be fair to say that this is tailored for tool-selection subagents?

You're absolutely right!

nasdaq & russell, yes. but not s&p - that has a 6-month requirement for eligibility, which ensures stock is past the lock-up period.

almost all 401k plans offer funds based on s&p 500, not nasdaq/russell others. s&p has also halved their trading days requirement from 1 yr to 6 months, but that's still sufficient to be past the post-ipo lock-up period.

It's not even a fair comparison - the effort involved in learning and building stuff from the internet (in the pre-LLM era) is an order of magnitude higher than turning your OAI/Anthropic account into an absolute slop cannon.

I'm primarily backend dev, but want able to prompt my way into building tiny JS web apps for personal use.

Wife always wanted to write animated story books for kids, and was actually able to create a bunch with Gemini (our daughter enjoys these more than the store-bought ones, since they're tailord/personalized for her).

won't this essentially disable prompt caching, that you get from a standard append-only chat history?

failed the car wash test.

i think instead of postiioning as a general purpuse reasoning model, they'd have more success focusing on a specific use case (eg coding agent) and benchmark against the sota open models for the use case (eg qwen3-coder-next)

you need a reviewer agent for every step of the process - review the plan generated by the planner, the update made by the task worker subagent, and a final reviewer once all tasks are done.

this does eat up tokens _very_ quickly though :(

providers' ToS explicitly states whether or not any data provided is used for training purposes. the usual that i've seen is that while they retain the right to use the data on free tiers, it's almost never the case for paid tiers

20-yr CAGR seems to be consistently higher than SP500: https://testfol.io/?s=7boYMdNxqjh

i workshop a detailed outline w it first, and once i'm happy w the plan/outline, i let it run while i go do something else

NME at all - 5.1 codex has been the best by far.

i think the standard recommendation is to do range partitioning on the hash of the key, aka hash range partitioning (i know yugabyte supports this out of the box, i'd be surprised if others don't). this prevents the situation of all recent uuids ending up on the same shard.

RAG != EBR

why can't you create separate git worktrees, and open each worktree in a separate IDE window? then you get the same functionality, no?

the broker should be adjusting the cost basis in the 1099 to account for wash sale. are you not seeing that? what broker are you using?

I had checked out SPR, but found the workflow based around reordering history via interactive rebase unintuitive & clunky. That was the motivation behind building this on my own.

I have been using this for a few months now, and it has served me well! I haven't spent much (any) time marketing it, so haven't really had any feedback from other users yet. Feel free to check it out & lmk if you have any suggestions. It's also open source, so feed free to open issue/PR on Github

I miss Phabricator from my time at Meta so much, I made this to achieve a Phabricator-like stacked-commit experience via the git cli: https://pypi.org/project/stacksmith/

Shameless self-promotion: https://pypi.org/project/stacksmith/

wrong

Every single person I know who pays for an LLM is a developer who pays for Claude because of coding ability

for a lot of tasks that aren't as cut & dry, i often find myself having to provide it pseudo code, which it can then one-shot to working code.

don't get me wrong, it's still a massive upgrade from the pre-sonnet era, but i still don't think it can take a high-level requirement and convert it into a working project... yet

+1 for Sceptre. Mine is coming up on 7 years, and I've never had any maintenance issues w it.

hash based partitioning makes repartitioning very expensive. most distributed DB now use key-range based partitioning. Iirc, Dynamo which introduced this concept has also made the switch

LM Studio is the easiest way to do it

did a comparison on LM Studio - the answers are eerily similar, but Mixtral is way way faster. Codellama-70B is slow to the point of being unusable.

(M1 Max, 64 GB RAM)

do you find Mixtral to be better than the new 70B one that Meta released a couple days back as well?

is this targeted towards non-technical users? for technical users, wouldn't an IDE plugin be a more natural part of the workflow?

phabricator & scuba are far better than anything i've used anywhere else.

i mean it’s Andreesen. They’re like the Softbank of Silicon Valley