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

Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

moonshotai.github.io

Post reply on HN
Print view View original

361–370 of 442 posts

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#361

Post by glenstein »

nylonstrung wrote: post #83
thedudeabides5 wrote: post #77

great, where does it think taiwan is part of...

I asked it that now and it gave an answer identical to English language Wikipedia When can we stop with these idiotic kneejerk reactions

It's fascinating the degree of defensiveness that shows up in comments on behalf of censorship, especially if it's Chinese. I think the reality is that these models are always going to be critically evaluated in terms of how they tailor AI to respond to topics they deem sensitive.

Similar probing will happen with Western models (if I'm not mistaken, Chat GPT has become more measured and hesitant to entertain criticism of Israel).

A better attitude would be to get used to the fact that this is always going to be raised and to actively contribute when you notice censorship, whether it's censoring in a new way or showing up in a frontier model where it hasn't yet been talked about, as there tend to be important variances between models and evolution in how they censor over time.

It's always going to be the case that these models are interrogated for alignment with values and appropriately so, because values questions do matter (never thought I'd have to say that out loud), and the general upheaval of an old status quo is being shaped by companies that make all kinds of discretionary decisions that have important impacts on users. Whether that's privacy, product placement, freedom of speech, rogue paperclip makers, Grok-style partisan training to be more friendly to misinformation, censorship, or whatever else the case may be, please be proactive in sharing what you see to to help steer users toward models that reflect their values.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#362

Post by ACCount37 »

glenstein wrote: post #357
ACCount37 wrote: post #303

Earlier quoted context omitted.

From what I've heard, Kimi K2 0905 was a major downgrade for writing. So, when you hear people recommend Kimi K2 for writing, it's likely that they recommend the first release, 0711, and not the 0905 update.

Interesting. As others have noted, it has a cut straight to the point non-psychophantic style that I find exceptionally rich in detailey and impressive. But it sounds like you're saying an earlier version was even better.

Again, it's just what I've heard, but the way I've heard it described is: they must have fine tuned 0905 on way too many ChatGPT traces.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#363

Post by glenstein »

thedudeabides5 wrote: post #87
nylonstrung wrote: post #83

Earlier quoted context omitted.

I asked it that now and it gave an answer identical to English language Wikipedia When can we stop with these idiotic kneejerk reactions

just checked, I wouldn't say it's identical but yes looks way more balanced. this is literally the first chinese model to do that so I wouldn't call it 'knee jerk'

And who knows for how long? My experience with very early iterations of Deepseek had direct answers to questions about Hong Kong, but later applied some kind of updates that stopped engaging with the topic. What was especially fascinating to me was some kind of hasty retrofitted layer of censorship, where Deepseek would actually show you an answer and then right in front of your eyes would replace it with a different answer saying it couldn't address the topic.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#364

Post by dzonga »

yanhangyhy wrote: post #249

As a Chinese user, I can say that many people use Kimi, even though I personally don’t use it much. China’s open-source strategy has many significant effects—not only because it aligns with the spirit of open source. For domestic Chinese companies, it also prevents startups from making reckless investments to develop mediocre models. Instead, everyone is pushed to start from a relatively high baseline. Of course, man…

you guys will outperform the US, no doubt.

energy generation multiples of what the US is producing. What does AI need ? Energy.

second - the open source nature of the models - means as you said a high baseline to start with - faster iteration.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#365

Post by naasking »

jstummbillig wrote: post #49
HarHarVeryFunny wrote: post #29

It's good to see more competition, and open source, but I'd be much more excited to see what level of coding and reasoning performance can be wrung out of a much smaller LLM + agent as opposed to a trillion parameter one. The ideal case would be something that can be run locally, or at least on a modest/inexpensive cluster. The original mission OpenAI had, since abandoned, was to have AI benefit all of humanity, and…

> I'd be much more excited to see what level of coding and reasoning performance can be wrung out of a much smaller LLM + agent Well, I think you are seeing that already? It's not like these models don't exist and they did not try to make them good, it's just that the results are not super great. And why would they be? Why would the good models (that are barely okay at coding) be big, if it was currently possible to…

> Why would the good models (that are barely okay at coding) be big, if it was currently possible to build good models, that are small?

Because nobody tried yet using recent developments.

> but there is no reason to assume that people who work on small models find great optimizations that frontier models makers, who are very interested in efficient models, have not considered already.

Sure there is: they can iterate faster on small model architectures, try more tweaks, train more models. Maybe the larger companies "considered it", but a) they are more risk-averse due to the cost of training their large models, b) that doesn't mean their conclusions about a particular consideration are right, empirical data decides in the end.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#366

Post by chaostheory »

dzonga wrote: post #364
yanhangyhy wrote: post #249

As a Chinese user, I can say that many people use Kimi, even though I personally don’t use it much. China’s open-source strategy has many significant effects—not only because it aligns with the spirit of open source. For domestic Chinese companies, it also prevents startups from making reckless investments to develop mediocre models. Instead, everyone is pushed to start from a relatively high baseline. Of course, man…

you guys will outperform the US, no doubt. energy generation multiples of what the US is producing. What does AI need ? Energy. second - the open source nature of the models - means as you said a high baseline to start with - faster iteration.

Going on a tangent, is Europe even close? Mistral has been underwhelming

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#367

Post by naasking »

Aurornis wrote: post #110
cantor_S_drug wrote: post #73

Earlier quoted context omitted.

In CS algorithms, we have space vs time tradeoffs. In LLMs, we will have bigger weights vs test-time compute tradeoffs. A smaller model can get "there" but it will take longer.

> In LLMs, we will have bigger weights vs test-time compute tradeoffs. A smaller model can get "there" but it will take longer. Assuming both are SOTA, a smaller model can't produce the same results as a larger model by giving it infinite time. Larger models inherently have more room for training more information into the model. No amount of test-retry cycle can overcome all of those limits. The smaller models will j…

> No amount of test-retry cycle can overcome all of those limits. The smaller models will just go in circles.

That's speculative at this point. In the context of agents with external memory, this isn't so clear.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#368

Post by taf2 »

Looks really amazing but I'm wondering is this one available to download? I see this: "K2 Thinking is now live on kimi.com under the chat mode [1], with its full agentic mode available soon. It is also accessible through the Kimi K2 Thinking API." but will this be on huggingfaces? Would like to give it a test run locally.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#369

Post by david927 »

dzonga wrote: post #364
yanhangyhy wrote: post #249

As a Chinese user, I can say that many people use Kimi, even though I personally don’t use it much. China’s open-source strategy has many significant effects—not only because it aligns with the spirit of open source. For domestic Chinese companies, it also prevents startups from making reckless investments to develop mediocre models. Instead, everyone is pushed to start from a relatively high baseline. Of course, man…

you guys will outperform the US, no doubt. energy generation multiples of what the US is producing. What does AI need ? Energy. second - the open source nature of the models - means as you said a high baseline to start with - faster iteration.

> will outperform

does outperform

China is absolutely winning innovation in the 21st century. I'm so impressed. For an example from just this morning, there was an article that they're developing thorium reactor-powered cargo ships. I'm blown away.

Re: Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model

#370

Post by LaurensBER »

chaostheory wrote: post #366
dzonga wrote: post #364

Earlier quoted context omitted.

you guys will outperform the US, no doubt. energy generation multiples of what the US is producing. What does AI need ? Energy. second - the open source nature of the models - means as you said a high baseline to start with - faster iteration.

Going on a tangent, is Europe even close? Mistral has been underwhelming

Not anywhere near close.

Europe doesn't have the infrastructure (legal or energy) and US companies offer far better compensation for talent.

But hey, at least we have AI regulation! (sad smile :))

Post reply on HN