Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

twitter.com

Post reply on HN
Print view View original

181–190 of 194 posts

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#181

Post by numpad0 »

londons_explore wrote: post #144
numpad0 wrote: post #139

Earlier quoted context omitted.

> I create a bunch of kanban tickets and assign them to one or more AI personas[1], Yeah that. Why can't we just `find ./tasks/ | grep \.md$ | xargs llm`. Can't we just write up a government proposal style document, have LLM recursively down into sub-sub-projects and back up until the original proposal document can be translated into a completion report. Constantly correcting a humongous LLM with infinite context len…

In my experience, this sort of thing nearly works... But never quite works well enough and errors and misunderstandings build at every stage and the output is garbage. Maybe with bigger models it'll work well.

I had hoped that this recursive breakdown approach could remove the need for bigger and bigger monolithic LLM for ever bigger tasks, by allowing every tasks to be at same granularity, but... I guess I should just try building one myself.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#182

Post by physix »

stogot wrote: post #54
physix wrote: post #29

Earlier quoted context omitted.

That reminds me of a thought I had about the poachings. The poaching was probably more aimed at hamstringing Meta's competition. Because the disruption caused by them leaving in droves is probably more severe than the benefits of having them on board. Unless they are gods, of course.

I thought that too

In the meantime, I discovered that it might simply be a type of acquisition that circumvents regulatory oversight https://medium.com/@villispeaks/the-blitzhire-acquisition-e3... seen from https://news.ycombinator.com/item?id=44553257

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#183

Post by neoromantique »

qmmmur wrote: post #39
simonw wrote: post #17

Pelican on a bicycle result: https://simonwillison.net/2025/Jul/11/kimi-k2/

I'm glad we are looking to build nuclear reactors so we can do more of this...

I honestly don't see an issue with that.

Except that instead of this, we're spinning up old coal plants, because apparently nuclear bad.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#185

Post by cosmojg »

simonw wrote: post #17

Pelican on a bicycle result: https://simonwillison.net/2025/Jul/11/kimi-k2/

For what it's worth, I think Kimi's modified MIT license still meets the OSI definition of "open source." For example, the explicitly OSI-approved "Attribute Assurance License"[1] contains similar wording:

> each time the resulting executable program or a program dependent thereon is launched, a prominent display (e.g., splash screen or banner text) of the Author’s attribution information

[1] https://opensource.org/license/attribution-php

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#186

Post by simonw »

cosmojg wrote: post #185
simonw wrote: post #17

Pelican on a bicycle result: https://simonwillison.net/2025/Jul/11/kimi-k2/

For what it's worth, I think Kimi's modified MIT license still meets the OSI definition of "open source." For example, the explicitly OSI-approved "Attribute Assurance License"[1] contains similar wording: > each time the resulting executable program or a program dependent thereon is launched, a prominent display (e.g., splash screen or banner text) of the Author’s attribution information [1] https://opensource.org/l…

Huh, I hadn't seen that one before!

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#187

Post by selfhoster11 »

apitman wrote: post #174
selfhoster11 wrote: post #170

Earlier quoted context omitted.

The full thing, 671B. It loses some intelligence at 1.5 bit quantisation, but it's acceptable. I could actually go for around 3 bits if I max out my RAM, but I haven't done that yet.

I've seen people say the models get more erratic at higher (lower?) quantization levels. What's your experience been?

If you mean clearly, noticeably erratic or incoherent behaviour, then that hasn't been my experience for >=4-bit inference of 32B models, or in my R1 setup. I think the others might have been referring to this happening with smaller models (sub-24B), which suffer much more after being quantised below 4 or 5 bits.

My R1 most likely isn't as smart as the output coming from an int8 or FP16 API, but that's just a given. It still holds up pretty well for what I did try.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#188

Post by selfhoster11 »

apitman wrote: post #173
jimjimwii wrote: post #167

Earlier quoted context omitted.

Had you posted this comment in the early 90s about linux instead of local models, it would have made about the same amount of sense but aged just as poorly as this comment will. I'll remain here happily using 2.something tokens / second model.

But local aka desktop Linux is still an awful experience for most people. I use Arch btw

I'd rather use Arch over a genuine VT100 than touch Windows 11, so the analogy remains valid - at least you have a choice at all, even if you are in a niche of a niche.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#189

Post by Lennie »

GenerWork wrote: post #52
sergiotapia wrote: post #49

Earlier quoted context omitted.

me too - we must energymaxx. i want a nuclear reactor in my backyard powering everything. I want ac units in every room and my open door garage while i workout.

You're saying this in jest, but I would LOVE to have a nuclear reactor in my backyard that produced enough power to where I could have a minisplit for every room in my house, including the garage so I could work out in there.

I'm gonna ask something stupid, maybe: what is keeping you from having a minisplit in each room ? You don't have to run them the whole day. Just where you are going to be for a couple of hours.

My guess is: the cost of the minisplits, pretty certain if you had them and turned them all on, you could still draw that much power from the grid.

And probably you are underestimating the cost of nuclear anyway.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#190

Post by kragen »

hnfong wrote: post #168
mindcrime wrote: post #150

Earlier quoted context omitted.

I personally have no (or at least little) problem with attribution. As you say, quite a few licenses have some degree of attribution required. There's even a whole dedicated (and OSI approved) license who's raison d'être is about attribution: https://en.wikipedia.org/wiki/Common_Public_Attribution_Lice... What I'm saying, if I'm saying anything at all, is that it might have been better to pick one of these existing l…

You speak as if "license proliferation" is actually a problem. But is it really? Sure, it may make some licenses incompatible with each other, but that's basically equivalent to whining about somebody releasing their code in GPL and it can't be used in a project that uses MIT... And your argument that the terms are "less understood" really doesn't matter. It's not like people know the Common Public Attribution Licens…

You can totally use GPL code in a MIT-licensed project, by changing the license on the overall work to the GPL. What you can't do is, for example, use GPL code in a CDDL project, or vice versa. The Apache Foundation went through a whole long process to release the Apache License 2 when version 1 was found incompatible with the GPL. License proliferation can be a big deal. In this case it's undesirable but less of a problem. I think.
Post reply on HN