behnamoh

54 ms·

691.
by behnamoh 1y ago
> rebranding the original o3 model as "o3-pro" interesting take, I wouldn't be surprised if they did that.
692.
by behnamoh 1y ago
Same here. One of my apps straight out stopped working because the gpt-4o outputs were noticeably worse than the gpt-4 that I built the app based on.
693.
by behnamoh 1y ago
> for near negligible drop in quality Hmm, that's evidently and anecdotally wrong: https://github.com/ggml-org/llama.cpp/discussions/4110
694.
by behnamoh 1y ago
good question, and I don't know of any, although it's a no brainer that someone should make it. a proxy to that may be the anecdotal evidence of users who report back in a month that model X has gotten dumber (started with gpt-4 a
695.
by behnamoh 1y ago
how do we know it's not a quantized version of o3? what's stopping these firms from announcing the full model to perform well on the benchmarks and then gradually quantizing it (first at Q8 so no one notices, then Q6, then Q4, ...
696.
by behnamoh 1y ago
OP is talking about the UI that let's you change the beam strength and focus
697.
by behnamoh 1y ago
> more distracting and gimmicky This. The animations on iOS are already a bit too much—now they've taken it to the next level.
698.
by behnamoh 1y ago
Elixir's concurrency model is fundamentally different than Go's; it's not just syntax difference.
699.
by behnamoh 1y ago
> concurrency by that logic Elixir is even better for agents. also the link at the bottom of the page is pretty much why I ditched Go: https://go.dev/blog/error-syntax The AI landscape moves so fast, and this conser
700.
Ask HN: How to remove all MCP servers from my Mac?
1 points by behnamoh 1y ago|1 comments
701.
by behnamoh 1y ago
I disagree. A lot of open source software literally come with "it's what it is" clause in the license, meaning that the maintainer has ZERO responsibility towards you, the user. No promises were made, you take it as is or lea
702.
by behnamoh 1y ago
How about Google Jules? also, of course OpenAI Codex would perform well because the tool is heavily tailored to this type of task, whereas Cursor is a more general-purpose (in the programming domain) tool/app.
703.
Serving Local LLMs with MLX(kconner.com)
1 points by behnamoh 1y ago|0 comments
704.
uv just keeps on getting better(loopwerk.io)
2 points by behnamoh 1y ago|0 comments
705.
Logos(tonsky.me)
1 points by behnamoh 1y ago|0 comments
706.
Why human experts may be useful for longer than you think(ninapanickssery.substack.com)
1 points by behnamoh 1y ago|0 comments
707.
by behnamoh 1y ago
Okay Apple, you got my attention. But I'm a strong proponent of "something is better than nothing" philosophy—even if OpenAI/Google/etc. are building reasoning models with the limitations that you describe, they are
708.
by behnamoh 1y ago
You're getting downvoted but what you said is true. The cost of self-hosting (and achieving +70 tok/sec consistently across the entire context window) has never been low enough to justify open source as a viable competitor to prop
709.
Claude Code can help check open-source code for security flaws(twitter.com)
2 points by behnamoh 1y ago|0 comments
710.
by behnamoh 1y ago
> Follow Sagyam's Blog's journey > By following, you'll have instant access to our new posts in your feed. > Continue with Google > More options As soon as I see this in a blog, I quit tab. Why do authors do this
711.
Mprocs written in Rust: Run multiple commands in parallel(github.com)
1 points by behnamoh 1y ago|0 comments
712.
by behnamoh 1y ago
Depending on your personality type (perfectionist or pragmatic), this is either the best of both worlds or the worst.
713.
by behnamoh 1y ago
While Tokasaurus’s Async-TP shows impressive throughput gains, it seems over-engineered for common use cases. The CPU overhead from async tensor parallelism only pays off at 6k+ token batches, and you need NVLink-connected GPUs to see real
714.
by behnamoh 1y ago
I mean, Samsung and Xiaomi at this point pretty much just copy Apple, sometimes shamelessly.
715.
by behnamoh 1y ago
yes there are, depending on the license.
716.
by behnamoh 1y ago
> at least in the 90's when we shared software it was for the reasons we outlined... With all due respect, that's not how most open source software is today [1]. A lot of CS students on the job market need Github stars or gre
717.
by behnamoh 1y ago
IDK man, I liked their API for a while due to its low latency but after a few requests the API crashes with 90% probability.
718.
by behnamoh 1y ago
Since this is a fork of continue.dev I wonder about the legal implications of monetizing the project.
719.
by behnamoh 1y ago
This is why I don't like open sourcing my projects anymore. Someone else with more resources just forks it and makes profits, thinking that merely acknowledging me pays my bills. For years, ollama didn't acknowledge llama.cpp and
720.
Ask HN: Test-Driven Physics?
1 points by behnamoh 1y ago|1 comments

More ›