munro

5 ms·

1.
by munro 3mo ago
If you bounce off the wall at the start, and RNG gods look down on you, you're lined up to go through castle on first shot. Then knock it through castle, and lined up for 3rd final putt into hole. But I agree with those points. I fee
2.
by munro 4mo ago
Lately I just have Claude build most things in Rust, it's really amazing. I tried Go, but I found it wasn't as good--Rust really does to me feel like Python. That said, it still struggles with the same class of errors of buildin
3.
by munro 6mo ago
Awesome haha, they look so shocked when you send a message, and then I get embarrassed and spin again
4.
by munro 6mo ago
> Rule 5. Data dominates. If you've chosen the right data structures and organized things well, the algorithms will almost always be self-evident. Data structures, not algorithms, are central to programming. It's so true, when
5.
by munro 7mo ago
ChatGPT needs to catch up hard, for me their model is unusable and I cancelled my subscription 5-6 months ago, so to me this post is hot air. I need to see results, going back to try Codex 5.2, then 5.3, downloading their new desktop clien
6.
by munro 7mo ago
>=99% accuracy wtf?!? I was initially excited until i saw that, because it would reveal some sort of required local min capacity, and then further revelation that this was all vibe coded and no arXiv, makes me feel I should save my attn
7.
by munro 9mo ago
I was looking at using this on an LTO tape library, it seems the only resiliency is through replication, but this was my main concern with this project, what happens with HW goes bad
8.
by munro 9mo ago
seriously cool, i just did something similar with quantiles for even bucketizing on arb key data types (still needs ord tho)
9.
by munro 9mo ago
just use a calendar event, it's more robust, and gives you the same feeling of 'oh yea...'
10.
by munro 10mo ago
I had built an agent with LangGraph a 9 months ago--now seems React agents are in LangChain. Over all pretty happy with that, I just don't use any of the dumb stuff like embedding/search layer: just tools&state But I was actu
11.
by munro 10mo ago
I built an ML classifier for product categories way back, as I added more classes/product types, individual class PR metrics improved--I kept adding more and more until I ended up with ~2,000 classes. My intuition is at the start when
12.
by munro 10mo ago
Amazing, some people are so enamored with LLMs who use them for soft outcomes, and disagree with me when I say be careful they're not perfect -- this is such a great non technical way to explain the reality I'm seeing when using o
13.
by munro 11mo ago
I'm more looking at the problem more like code https://bbycroft.net/llm My immediate thought is when the model responds "Oh I'm thinking about X"... that X isn't from the input, it's from atten
14.
by munro 11mo ago
I love the irony of the site showing a MASSIVE banner with a huge green "Download Extension for Mac (Free)" button. This thing is 280px tall! I clicked it for shits and giggles and upon returning it showed a popup XD https:&#x2F
15.
by munro 11mo ago
I wish they dug into how they generated the vector, my first thought is: they're injecting the token in a convoluted way. {ur thinking about dogs} - {ur thinking about people} = dog model.attn.params += dog > [user] wh
16.
by munro 1y ago
But what if `ty` was also eventually merged into `uv` as well? 8-) That's probably the vision, given all from astral.sh, but `ty` isn't ready yet.
17.
by munro 1y ago
Link to paper here https://arxiv.org/pdf/2506.21734 Still reading, but the benchmarks for ARC-AGI-1, ARC-AGI-2, Sudoku-Extreme (9x9), and Maze-Hard (30x30) look impressive.
18.
by munro 1y ago
was a bit hard for me to find, but here's a video of the trackball being used: https://www.youtube.com/watch?v=RCXgPqlpZeM
19.
by munro 1y ago
That's sort of how I felt about things before, but the reality I believe is we wouldn't have uv if they 'just contributed to poetry'.
20.
by munro 1y ago
No model is perfect, but some are useful. Their "baseline LTV" looks at the sum of listings on their platform (plus other features), then tries to forecast the next 365 days — so this should indirectly capture people coming and g
21.
by munro 2y ago
Here's an interesting write up on various algorithms & different epsilon greedy % values. https://github.com/raffg/multi_armed_bandit It shows 10% exploration performs the best, very great simple algorithm. Al
22.
by munro 2y ago
Wow, this is cool-- first time I've seen generating well formed output has been made easy, the use of Pydantic is clever/easy... Which has lead me to the prior art: using same Pydantic API to generate structured output from local
23.
by munro 2y ago
Lots of amazing ideas here! One more that would be great is 'sandboxing' if a program can use network/filesystem/etc, then I would feel safe to run any random scrap if I knew I would get some sort of oauth-looking scree
24.
by munro 3y ago
> Haha used to be 1PB, stored completely on Google Drive (they had unlimited), but then they cut me off so I cut down to 300TB and switched to self hosting at a data warehouse. > Now me and like 100 people share a ceph storage that we
25.
by munro 3y ago
And then I always think, Netflix only has ~3,600 movies... My friend's Plex server has 4x that (in just movies). I'm also often underwhelmed by Netflix's engineering posts
26.
by munro 3y ago
> 500ms is an acceptable latency for a search that runs offline without network calls. Definitely disagree, having fast search makes a huge difference. Apple Photo's search is fast. Especially since CLIP embeddings aren't per
27.
by munro 3y ago
I just tried to use atuin... I have to login, but I don't want to :( I tried to run a search locally... but I have to run a PostgreSQL server??? Why isn't it just allowing embedded DB like mcfly? I know this isn't the pla
28.
by munro 3y ago
I just had an argument over IRC with a stranger on the internet last week. We're still out there. I thought the humor on this site hadn't aged well, but this one got me: < pronto> :( < GiftdKook> Turn that f
29.
by munro 3y ago
I have been using Polars on and off for the past few years, but now I've been using it 99% exclusively for the past 2–3 months. I would say it's ready for prime time! I don't get any @TODO seg faults anymore lol I don'
30.
by munro 3y ago
Wow!! Very cool! I tried to use a KVM to switch from my personal macbook, work macbook, and gaming computer.... the hardware was awful, I had just ended up manually switching. The gaming computer had its own keyboard&mouse, so it wa

More ›