dmarwicke

6 ms·

1.
by dmarwicke 8mo ago
2FA publishing still doesn't work for me. just use legacy tokens at this point, gave up trying to figure out what's wrong
2.
by dmarwicke 8mo ago
npm's package.json and package-lock.json get out of sync constantly on my team. at least go only has one file to mess up
3.
by dmarwicke 9mo ago
this is just optimizing for token windows. flat code = less context. we did the same thing with java when memory was expensive, called it "lightweight frameworks"
4.
by dmarwicke 9mo ago
does it handle skewed distributions? faker's always been useless for this - like, your test data ends up with everyone having 5 orders when real data is all long tail
5.
by dmarwicke 9mo ago
how does this decide what's safe to delete? i've nuked docker caches before and broken builds in annoying ways
6.
by dmarwicke 9mo ago
curious what the token costs look like on a real codebase. opus ain't cheap and C++ headers get big fast
7.
by dmarwicke 9mo ago
we had to restrict ours to views only because it kept trying to run updates. still breaks sometimes when it hallucinates column names but at least it can't do anything destructive
8.
by dmarwicke 9mo ago
couldn't find anything about invalidation in the docs. how does that work? usually where these abstractions fall apart for me
9.
by dmarwicke 9mo ago
does this end up flagging legit packages that just have 'ai' or 'gpt' in the name? feels like half of pypi would trigger at this point
10.
by dmarwicke 9mo ago
curious about the startup latency in practice. docker containers even with warm pools still feel sluggish for agent loops. e2b does firecracker and it's noticeably snappier
11.
by dmarwicke 9mo ago
how well does the flatten() translation work in practice? every time i've used localstack or similar the queries work locally then break in subtle ways once deployed
12.
by dmarwicke 9mo ago
the sql-first thing is interesting. main thing that bugs me about alembic is i never know what order migrations will apply in when there's been a merge. how does jetbase handle branching?
13.
by dmarwicke 9mo ago
hit this with support ticket filtering. llm kept missing weird edge cases. wrote some janky regex instead, works fine
14.
by dmarwicke 9mo ago
does this work with multidimensional or just 1D? polars had the same limitation, ended up just sticking with numpy
15.
by dmarwicke 9mo ago
why postgres for this? feels heavy for a queue. tried something similar with redis and it was way simpler
16.
by dmarwicke 9mo ago
does this handle the thinking blocks that claude puts in? those break json.parse constantly when it streams reasoning before the actual response
17.
by dmarwicke 9mo ago
22gb for mostly text? tried loading the site, it's pretty slow. curious how the query performance is with this much data in sqlite
18.
by dmarwicke 9mo ago
does this actually fix metadata filtering during vector search? that's the thing that kills performance in pgvector. weaviate had the same problem, ended up using qdrant instead
19.
by dmarwicke 9mo ago
does this do continuous batching or just static? couldn't tell from the code
20.
by dmarwicke 9mo ago
how are people syncing these? i just have a cron job running overnight and it feels dumb. tried debezium but it was overkill
21.
by dmarwicke 9mo ago
does this do point-in-time recovery? tried like 3 postgres backup tools, they all claim to support it but then you actually need to restore and it's a mess
22.
by dmarwicke 9mo ago
memory voting sounds interesting but does it work? i tried having agents mark useful chunks once, they just marked everything as helpful. accuracy went to shit
23.
by dmarwicke 9mo ago
this has to be insanely expensive right? merge sort does n log n comparisons, so sorting 100 items is like 600 llm calls
24.
by dmarwicke 9mo ago
wait this seems expensive? why vectorize everything when postgres is sitting right there. is the point to search through logs or something
25.
by dmarwicke 9mo ago
had this happen with a retry loop. hit $80 on anthropic before i caught it. how does this handle retries? seems like an agent could just keep retrying and blow past the limit
26.
by dmarwicke 9mo ago
wait, zero-copy deserialization isn't rust-specific. you can mmap structs in C. done it before, works fine
27.
by dmarwicke 9mo ago
the lockfile thing is what sold me honestly. had a nightmare with torch pulling different cuda builds on my linux box vs mac. uv just... works? still half expecting it to break
28.
by dmarwicke 9mo ago
yeah i have this setup. pain point is syncing between them, still haven't figured out a good pattern. just running a cron overnight which feels dumb but works
29.
by dmarwicke 9mo ago
been bitten by this. some sync call buried in a dependency, blocked everything. wrote a janky decorator to catch it but this looks cleaner
30.
by dmarwicke 9mo ago
tried this with msgpack last year. accuracy tanked. models have seen a trillion json examples, like 12 of whatever format you invent

More ›