scosman

11 ms·

211.
by scosman 11mo ago
Highly suggest trying Azure Trusted Signing on a CI system with windows boxes (I use Github). Windows signing was an expensive nightmare before, but is now relatively painless and down to $10/mo (which isn't cheap but is cheaper t
212.
by scosman 11mo ago
Notarization is completely optional when building any OSS software on a Mac, and not part of any default build process I know. A Mac can sign builds for running locally, a process which is fast, completely local, and does require building t
213.
by scosman 11mo ago
I suggest the SenseAir sensors. They don't seem as susceptible to fakes, and auto calibrate when exposed to fresh air. Supported by ESP home so build is simple. Any evidence the Ikea sensor are actual CO2 censors and not just cheap &qu
214.
by scosman 11mo ago
Word on reddit is they discontinued the blinds due to reliability problems. I have 3 and 1 failed so makes sense to me. They want to bring them back, but it's a complete redesign, not just normal planned product lifecycle iteration.
215.
by scosman 11mo ago
Copyrighted content is copyrighted content. It doesn’t matter if/how Apple published it, others don’t have a right to redistribute it without a license.
216.
by scosman 11mo ago
Not so much features as design. Svelte files look like HTML+TS files. You aren’t learning some abstraction to HTML, you are just using HTML. But it adds the modern bits you need: reactivity, loops, components, routing, etc. Nothing react do
217.
Mail your parents a tailscale node(tailscale.com)
3 points by scosman 11mo ago|1 comments
218.
by scosman 11mo ago
Go
219.
by scosman 1y ago
There are guarantees from several providers that they don’t train on, or even retain, a copy of your data. You are right they could be lying, but some are big enough that would be catastrophic to them from a liability point of view. Re:crea
220.
by scosman 1y ago
Nothing open is quite as good as Sonnet 4.5 and Codex 5. GLM 4.6, MiniMax M2, Deepseek v3.2, Kimi K2 and Qwen Coder 3 are close. But those are hundreds of billions of parameters, so running locally is very very expensive.
221.
by scosman 1y ago
Makes sense that you'd run locally then. But really no host you trust to not keep data? Big tech with no-log guarantees and contractual liability? Companies with no-log guarantees and clear inference business model to protect like Toge
222.
by scosman 1y ago
What are folks motivation for using local coding models? Is it privacy and there's no cloud host you trust? I love local models for some use cases. However for coding there is a big gap between the quality of models you can run at home
223.
by scosman 1y ago
Technically it’s deterministic. It just might not be correct :)
224.
by scosman 1y ago
You can zero out temperature and get determinism at inference time. Which is separate from training time where you need forms of randomness to learn. The point is for the quote "all LLMs that I know of rely on entropy and randomness to
225.
by scosman 1y ago
That's at training time, not inference time. And temp/top_p aren't used to escape local minima, methods like SDG batch sampling, Adam, dropout, LR decay, and other techniques do that.
226.
by scosman 1y ago
No one said it was a user setting. The person building the spreadsheet agent system would tune the hyper-parameters with a series of eval sets.
227.
by scosman 1y ago
> all LLMs that I know of rely on entropy and randomness to emulate human creativity Those are tuneable parameters. Turn down the temperature and top_p if you don't want the creativity. > Claude is not going to improve this. We c
228.
by scosman 1y ago
> GPT-5 is proving useful as a literature review assistant > No, it does not. > It is excellent when just finding something is enough.
229.
by scosman 1y ago
Saying it isn't useful is a bit of an overstatement. It can search, churn through 500k words in a few minutes, and come back with summaries, answers, and sources for each point. Should you blindly trust the summary? No. Should you veri
230.
by scosman 1y ago
Another potential one: new and not ready for primetime yet, but from the Lance team so it's promising: https://github.com/lancedb/lance-graph
231.
by scosman 1y ago
Correction: there is an embedded mode. https://surrealdb.com/docs/surrealdb/embedding
232.
by scosman 1y ago
Don't forget the environmental impact of a smaller box. The box will probably be less than half as thick, doubling shipping efficiency. These are air freight, so the CO2 impact is not negligible. I'll take the discount and use one
233.
by scosman 1y ago
No that's exactly how open source works. They are under zero obligation to wake up tomorrow and continue to work on it. Everyone is lucky they open sourced it, and anyone else is welcome to pick it up where they left off. Contributing
234.
by scosman 1y ago
amazing reply. Thanks!
235.
by scosman 1y ago
Anyone know how this compares to 'ty': a new typechecker from Astral (uv/ruff team), also written in Rust and super fast. I had been waiting on it to reach beta, but would love to move to something faster than pyright sooner
236.
by scosman 1y ago
They tried something hard and it didn’t work out. In the process they did some great work. Plus they gave it all away for free, allowing literally anyone in the world to continue the work. They are awesome. If the rest of us criticized peop
237.
by scosman 1y ago
“Cloud native” - not embedded
238.
by scosman 1y ago
Not embedded last I checked. Unless that changed.
239.
by scosman 1y ago
Oh too bad. Small fast embedded graph DBs are rare. Any good alternatives?
240.
by scosman 1y ago
Yeah exactly. Some people have 100k photo collections. The cost of scanning isn’t trivial. They should limit the number of times you turn it on, not off. Some PM probably overthought it and insisted you need to tell people about the limit b

More ›