dangoodmanUT

10 ms·

1.
by dangoodmanUT 5d ago
> Reasoning models sometimes put everything in the reasoning field and hand back content: null, finish_reason: "stop". 345 completion tokens, HTTP 200, nothing to show the user. This is actually expected behavior. No content an
2.
Show HN: Tinysandbox-JS-Runtime - JS in WASM in v8(github.com)
2 points by dangoodmanUT 17d ago|0 comments
3.
by dangoodmanUT 18d ago
This feels suspiciously AI-authored
4.
by dangoodmanUT 19d ago
> DbOptions::durable() > Appended to the WAL without a per-write sync So… it’s not durable? Durable doesn’t mean “survives a process restart”, it means “durably saved to persistent storage”. For example, this “durable” mode wouldn’t s
5.
by dangoodmanUT 19d ago
Iirc that’s how badger handles the WAL as well
6.
Show HN: Waltier – a generalized version of Cursor's S3 WAL(github.com)
4 points by dangoodmanUT 26d ago|0 comments
7.
by dangoodmanUT 1mo ago
I hate these "The best X thing Y has ever released". Unlike when Apple says "it's the best iphone we've ever made", LLMs are more or less interchangeable. So "OpenAI's best model" means nothing i
8.
by dangoodmanUT 1mo ago
It's funny to me that we still don't have incremental materialized views. All of the parts are there (export state, agg_state (forget fn name), finalize). I wonder if they're avoiding an explicit war with clickhouse or someth
9.
by dangoodmanUT 1mo ago
> How this was written: drafted with an AI assistant from my own notes, then edited by me. Doesn't feel that way with "The tradeoffs I am not going to hide" and "How it resolves, honestly" This site needs Bryan G
10.
by dangoodmanUT 1mo ago
> In April, Anthropic announced a new model called Mythos that was optimized for software vulnerability finding No, it was just good at it because it wasn't RL'd against it. I know this is a small detail, but it tosses journali
11.
by dangoodmanUT 1mo ago
> But you just can't trust them not to invisibly censor sensitive clinical/legal docs What's an example of this?
12.
by dangoodmanUT 1mo ago
> turn on the tv { "function_calls": [ { "name": "lock_door", "arguments": { "door": "tv" } } ], "confidence": 0.0158 } Very in
13.
by dangoodmanUT 1mo ago
I'm seeing opus 59.2, Qwen 58.4?
14.
by dangoodmanUT 1mo ago
If it was 200ms, then it had already charged the card...
15.
by dangoodmanUT 1mo ago
The major labs don’t allow assistant prefill, so you have to “summarize”
16.
by dangoodmanUT 2mo ago
> The models we reached for first, Claude Opus and Fable, refused a large part of that work: their safety guardrails treated reverse-engineering an exploit the same as launching one. Interesting single getting into their cyber program, a
17.
by dangoodmanUT 2mo ago
This is really cool, and really glad they shared it! It's clearly a lot of Claude though: > Neither is a URL fetch, so the allowlist never saw them. > No code ran; this vector is purely a file disclosure, and it is the one the ag
18.
by dangoodmanUT 2mo ago
This. The title might as well be “Choose a hammer rather than a wrench (when driving nails)”
19.
by dangoodmanUT 2mo ago
Because huggingface is not charging them? CFAA doesn't just mean the feds kick down your door, you actually have to get reported and sued over it.
20.
by dangoodmanUT 2mo ago
throws away spec doc for my own Are we all having this idea?
21.
by dangoodmanUT 2mo ago
25k tokens is a few turns
22.
by dangoodmanUT 2mo ago
J-space sounds oddly similar to...
23.
by dangoodmanUT 2mo ago
finally, this was needed. A big worry was always "why does workers sit in front of my cache? that's a waste of an invocation if i'm returning a cached result"
24.
Show HN: Tinysandbox – An ultra-minimal sandbox with QuickJS WASM isolates(github.com)
4 points by dangoodmanUT 2mo ago|0 comments
25.
by dangoodmanUT 3mo ago
i believe the author is confused this is the wrong direction (jk i read the readme)
26.
by dangoodmanUT 3mo ago
If you actually bench it locally (local S3, actually writes to disk for "staged" operations), ZeroFS performs horrifically. Ceph blows it out of the water. I don't have the exact numbers, but when I was building a toy CoW dis
27.
by dangoodmanUT 3mo ago
How are they going to be competitive with top models at 70B size?
28.
by dangoodmanUT 3mo ago
iroh is consistently one of the most delightful projects i've ever worked with. The people reflect that too. Congrats iroh team!
29.
by dangoodmanUT 3mo ago
These are actually the most durable businesses because they have network effects: restaurants alright signed up, sellers onboarded, etc. Easier for them to adopt AI than AI companies to rebuild the networks
30.
by dangoodmanUT 3mo ago
Not comparing to GPT Pro models is a bit strange, considering that's the natural comparison

More ›