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

Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

emergingtrajectories.com

Post reply on HN
Print view View original

341–349 of 349 posts

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#341

Post by anthonypasq »

janalsncm wrote: post #339
anthonypasq wrote: post #335

Earlier quoted context omitted.

because its a matter of public record that the release of Fable was artificially delayed

By Anthropic. The government blocked it after it was already released. Every LLM product goes through testing and alignment after training, maybe even some quick improvements here and there. Kimi probably did something similar. Put another way, if Google says they have the best model in the world but won’t release it in December I will start caring in December, not before.

if Dario and the CEO of Moonshot switched places, Mythos would have been generally available to the public 4 months ago. As a statement of fact, it wasnt, because of Project Glasswing and Dario thinking they created a superweapon that the rubes shouldnt have access to.

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#342

Post by Tuna-Fish »

quotemstr wrote: post #274
Tuna-Fish wrote: post #259

Earlier quoted context omitted.

Taalas only uses SRAM for the KV cache and the activations, the weights are in mask rom in the metal layers. If they designed this right, it means that once they have a model, so long as they keep the hyperparameters fixed they can change the weights much faster than it takes to spin up a completely new chip, essentially at a cost of doing a minor revision.

Is the mask ROM really going to be worth it over Carmack's high-bandwidth-flash concept? I mean, sure, I could be convinced I guess, but it's not obvious.

Yes.

Not because rom is better than flash, but because the critical part is distributing the memory with the compute. Mask ROM is just the densest way of embedding memory with logic. Instead of having a large pool of memory connected to the separate execution units with a bus, each execution unit locally has the rom that it uses. Data movement is >90% of energy use in modern ai accelerators, removing it as far as practical is how they get performance and silicon and energy efficiency.

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#343

Post by mullingitover »

adventured wrote: post #314
mullingitover wrote: post #192

Earlier quoted context omitted.

> the winner will be whoever burns their models to ASICs fastest. It'll obviously be China, and they won't need the bleeding edge of lithography tech to make it happen. Every Chinese smartphone will have something like Sonnet 5, along your car (well, not those of us in the US, but we'll look longingly at pictures of them while we drive whatever the government decides we're allowed to drive in Fortress America). Give…

China won in desktop PCs? Nope. China won in cloud services? Nope, not even close. They had to clone AWS just to try to keep up. China won in mobile? Nope. Although they're very competitive there. China won in search? Nope. Baidu who? China won in ecommerce? Nope. Their dominance is overwhelmingly domestic. China won in software? Nope. Windows is US. Android is US. iOS is US. MacOS is US. Linux is US/Europe/Global. L…

India and Latin America doing some final assembly is kind of a whimper of a counterpoint.

Most of this litany is trivially refuted by pointing out that roughly 40% of the world's declared 5G Standard Essential Patents were developed in China. The only reason they don't completely dominate telecom globally is the panicked invocation of national security fears to erect emergency trade barriers.

"China won in ecommerce? Nope. Their dominance is overwhelmingly domestic."

Their domestic ecommerce market eclipses the US and the next several biggest markets combined. You can argue protectionism, but then you have to face the reality that the US is doing its own flailing protectionism today.

Generally, this is an embarrassing category error of an argument. Running models on ASICs, IoT devices, and EVs are infrastructure and hardware plays: areas where China is inarguably so good that it's a problem.

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#344

Post by tracker1 »

drob518 wrote: post #301
tracker1 wrote: post #256

Earlier quoted context omitted.

Even then... Do we now stop the with to asic DeepSeek and do k3 instead? Assuming such work was happening.

My personal feeling is to not move to ASICs just yet. Things are still pretty frothy right now, so I would probably wait 6-12 months. At some point the froth always calms down. At that point, commit to ASICs. That said, I’m also not totally sure exactly how much the ASIC hard codes vs having some wiggle room. The Taalas site is a bit vague as to exactly how they encode the model.

I would probably wait until a year goes by without any significant advancement, at the very least before making such an effort.

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#345

Post by tracker1 »

daadx wrote: post #290
tracker1 wrote: post #240

Earlier quoted context omitted.

Are you a Chinese national, or otherwise paid by China?

the lot of you sound salty lmao are you american?

So you created a new spam account... I'll answer when you and OP do.

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#346

Post by tacitusarc »

Tuna-Fish wrote: post #259
akiselev wrote: post #161

Earlier quoted context omitted.

The weights are in SRAM, so the LLM architecture is burned in but the weights can be updated.

Taalas only uses SRAM for the KV cache and the activations, the weights are in mask rom in the metal layers. If they designed this right, it means that once they have a model, so long as they keep the hyperparameters fixed they can change the weights much faster than it takes to spin up a completely new chip, essentially at a cost of doing a minor revision.

But attention/KV cache is the heavy bit in long context lengths which is what everyone needs…

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#347

Post by mountainriver »

LarsDu88 wrote: post #110
mountainriver wrote: post #85

Earlier quoted context omitted.

Sol is running on Cerebras right?

Cerebras is not an ASIC. It is a large wafer scale chip that has a load of small SRAM modules paired with tiny compute models. The SRAM is basically a giant cache that is supposed to eliminate the bottleneck between shuttling and materializing big tensors between DRAM and cache, but the actual amount of SRAM is still nowhere near enough to server a frontier model on a single wafer. You still need dozens of wafers, wh…

I know that but they also said Cerebras

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#348

Post by JohnBooty »

tracker1 wrote: post #199
JohnBooty wrote: post #137

Earlier quoted context omitted.

At a 50-100x speedup even a GPT-4o class model could perhaps compete with much newer models simply by thinking deeper, doing harness-controlled Ralph loops, etc. Sure, then it might be "only" ~2-5x faster, but, you wouldn't need to throw all the ASICs into the trash bin. One could also imagine hybrid models, where part of the model is burned into ASICs and part of the model exists in VRAM/HBM2 so it can be updated. I…

Giving a literal monkey the ability to press more keys faster doesn't get a good joke from it. A model will often come up with worse results given more cycles of compute, only because it will tailspin from second guesses, rethinking and literal flip-flopping on concepts. --- edit: to those following the thread below... if you look at the comments from the account replying, it's pretty obviously a pro-China account an…

    Giving a literal monkey the ability to press more 
    keys faster doesn't get a good joke from it.
To some extent, it's because making a joke isn't an iterative exercise with a verifiably correct outcome.

However, a lot of jobs are. Notably, much of software engineering.

Re: Kimi K3, Qwen 3.8, and Anthropic's (Potential) Unravelling

#349

Post by dehugger »

sdfefcxv wrote: post #200
stillpointlab wrote: post #139

Earlier quoted context omitted.

I admit I would like a faster model - but even though I have faster models available I still go to Fable or GPT-5.6 90% of the time. So there is a gap between a potential preference and a revealed preference. Custom AI for things like facial recognition in cameras has existed for decades, before LLMs were a thing. I don't see that getting replaced. And on-device conversational intelligence might go that route as well…

What you think doesn't matter unless its strictly for personal use. Your company will decide what makes economical sense.

Some of us happen to be in positions to make those decisions at our respective companies. What we think does, in fact, matter.
Post reply on HN