As a replacement for ollama, LM Studio is probably the most appropriate. Though I haven't tried Unsloth Studio yet. You get more control using llama.cpp directly, but there's more of a learning curve.
vLLM has the best performance if you can fit your entire model into VRAM. llama.cpp is the usual go-to if you're partially loading into RAM. LM Studio is a sensible front-end to llama.cpp. ik_llama.cpp has more advanced CPU quantization strategies than llama.cpp. If you're running super large models mostly from RAM, ktransformers can sometimes be the highest performer, if it works for your model.
Meta literally ran a man in the middle attack to spy on it's user's encrypted network traffic when they used third party apps [0]. More recently (and relevant to this issue), the engaged in industrial scale piracy to get training data for their LLMs [1]. The idea that they have changed since Zuck was a college student creeping on his female classmates and now wouldn't commit actual crimes against their own users in order to get a bit more data is just demonstrably false.
"If you need information on anyone at Harvard, just ask"?
"I'm going to fuck them [Winklevoss twins] in the ear"?
Applies nowadays? Ok, when? When he copied Snapchat into every single product? When he retracted all his messages on Facebook? (https://news.ycombinator.com/item?id=16770818) When he explicitly killed Instagram (buying it) because 'they can hurt us'? Do you want more examples...?
> You really think a throwaway quote when Zuckerberg was a college student applies nowadays?
What does "nowadays" mean? What changed?
> You really think Meta would risk getting a massive (losing) lawsuit on their hands, in exchange for what?
And what is the risk? Food companies are at risk when they put addictive chemicals into food because they get controlled regularly. What is the equivalent here?
To be fair, they're not exactly wrong. Worth repeating with N=10,100,... until we get a robust conclusion, but as it is, there's not much to go on in this one
Exactly. We're not researchers and many of us will internalize the finding without the proper confidence weighting. I wish reporting and HN had a higher standard for studies linked on the site with higher N and ideally some independent replications.
If the results is really interesting and novel then why aren't others racing to replicate it? Because it is not. Yet we're reporting it here with N=1.
The lack of an option on Windows makes it harder to justify when alacritty + nvim achieves great speeds as well, with all the customizability and what not.
Can anyone chime in on whether using zed on wsl is viable, or loses all the speed benefits?
This is such an all-or-nothing thought. We can aim for "better", or even just "good enough", without aiming for absolute perfection or mediocrity, and things will still keep getting better.
His point is excessively perfectionistic at best, to the point of detriment, and at worst... well, it's pedantic.
To follow this analogy, isn't it possible then for someone to study precisely what the difference is, and become an expert, thus bringing us back into the level of expertise we were at at the start of the analogy?
Surely the experts didn't all learn from each other; who was the first expert? That expert surely learned in some other way, so the only thing lost at the start of the analogy is the time required for someone interested to (re)achieve mastery.
Sure, related to tuning because it's a pretty closed problem. The expertise in terms of tuning was developed over about ~800 years, but the math for modern tuning was known 500 years ago. It's conceivable that one could re-invent equal temperament and then quickly re-invent the modern tuner given everything we know about electronics and audio processing. However, that knowledge all builds on itself. If we decide that all audio processing is done with RNNs/ML instead of objective ("old") mathematics, then we're going to lose the ability to make a tuner, too, and eventually we'll need a new Fourier to come back up with the Fourier transform.
About the tomatoes in the other comment chain? Your guess is as good as mine whether we can recover that knowledge.
I heard that pianos have stretched harmonic series due to string tension/weight/something, so piano tuners actually have to tune upper notes higher and lower notes lower, while ensuring various harmonics interact well. There's quite a bit of art to it rather than pure numeric ratios (which may or may not be possible to encapsulate in ML).
