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

Qwen 3.8

twitter.com

Post reply on HN
Print view View original

291–300 of 793 posts

Re: Qwen 3.8

#291

Post by LogicFailsMe »

zapkyeskrill wrote: post #44
dannyw wrote: post #26

Earlier quoted context omitted.

In China, you can’t officially use US APIs. The world saw a taste of this with Fable, but in China, this has been the situation all along. So it’s not a surprise why open weights are so cherished. As frontier models continue to block everyday individuals from securing their own codebase, I expect the adoption and usage of open weights to continue. As an example, HuggingFace recently was investigating a security incid…

How does this explain open weights? They could easily take the same closed route like their American friends

I would guess the Chinese government has a strong wish to lift all Chinese AI boats and bets. That it sinks western closed weight Frontier Labs in the process would be just be gravy on top, no? Broadly, the difference between mercantilistic capitalism and western late stage capitalism IMO.

Re: Qwen 3.8

#292

Post by ethbr1 »

roenxi wrote: post #113
zapkyeskrill wrote: post #44

Earlier quoted context omitted.

How does this explain open weights? They could easily take the same closed route like their American friends

Open sourcing is a complex decision so who knows what their calculations are. But I'd assume that they're preparing for some sort of winner-take-all market in model quality where if they don't do anything the winner will be aggressive, hostile and American. Likely trying to push the Chinese economy back to the year 2000. If that is the starting point either the Chinese have to win the market (unlikely) or squeeze the…

Radiolab did a great episode that covered the history of Chinese character computing: https://radiolab.org/podcast/wubi-effect

To summarize, in the 70s and 80s, China was facing an existential threat with their inability to access an economic accelerator (widespread computing) in their native language.

To the extent that there was serious consideration at the highest levels of converting the entire country to an alphabet-based writing system.

I'd expect they're looking at AI the same way:

We have to have access to this. Most of the frontier labs are American (or European). Therefore we need a solution we have continued access to.

Open weights feels simultaneously Chinese in nature (progress through making a design copyable and improvable by a large number of people) and economic (providing an incentive for the world to use Chinese models over other frontier).

Re: Qwen 3.8

#293

Post by pettijohn »

mft_ wrote: post #32
nsbk wrote: post #19

Bring it on! Hoping that they release smaller sizes of Qwen3.8. I use the 35B MoE and 27B dense models locally and most of the time I don’t need to reach out to Claude. Extremely useful specially when requests include sensitive and/or personal data

I think everyone is hoping this! It would be great if they'd release an MoE model somewhere between the 35B size of 3.6 and the 122B version of 3.5 - it could be a great balance of speed and ability for people with reasonably powerful but not insane home computers.

SO MUCH THIS. I have Strix Halo with 128GB RAM and was a large and fast model like 122B A10B. Here's hoping!

Re: Qwen 3.8

#294

Post by k__ »

2Gkashmiri wrote: post #157
5701652400 wrote: post #99

in my experience of 1 month daily use, Qwen 3.7 Pro is just unusable. wastes too much time, goes off track, useless stuck loops, cannot debug at all. Deepseek V4 Pro is night-and-day compare to Qwen. actually Qwen models seems the worst SWE experience so far. and it is super expensive compare to Deepseek. cannot delegate anything to it, cannot use it real-time low-level tasks either. totally unusable.

Can you tell me more about deepseek? I paid $2 for deepseek api, put the key in void editor and made a crypto tool in html. It turned out to be around 67kb. I used sample files in CSV that were a few hundred lines. It spent around $1.8 in the hour or two or light coding and follow up bugs. Is it really really this much? I can't imagine spending a month using it for a day job, it would cost more than the salary so wha…

My 2 weeks with DeepSeek V4:

Pro is ~50% more expensive than Flash.

Both need babysitting.

Plan, split in small tasks, give it docs, types, tests, linter, best practice examples, etc.

Always start a new session when starting a task.

Do regular manual sanity checks, and tell it to find issues in the codebase.

I pay like $1,50 per day for Pro.

Re: Qwen 3.8

#295

Post by lerchmo »

Matl wrote: post #22
gardnr wrote: post #6

Earlier quoted context omitted.

It's hard to say what their motivation is. The Chinese firms seem to be working hard to commoditize intelligence which may be the most effective way to debase American frontier labs. And yeah: it also happens to be really good for humanity.

> It's hard to say what their motivation is. Not that hard to say IMO, they basically see models becoming a commodity and see value in the applications on top of them. So if Alibaba Cloud is the best place to build applications on top of Qwen, why not give the model itself away?

Also probably betting on their compute and energy capabilities.

Re: Qwen 3.8

#296

Post by tripleee »

SwellJoe wrote: post #34

Qwen is the most censored of the Chinese models in my testing, which makes me wonder in what other ways it is compromised. Open weights doesn't really reveal what's in there. And, in my tests, existing Qwen models are not at the pareto frontier of any metric; DeepSeek V4 Pro is better, faster, and much cheaper than Qwen 3.7 Max. (DeepSeek is also among the least censored of the Chinese models.) I guess we'll see if t…

The nice thing about it being open-weight is that you can uncensor it.

Re: Qwen 3.8

#297

Post by notnullorvoid »

mschuster91 wrote: post #101
traceroute66 wrote: post #60

Earlier quoted context omitted.

> How does this explain open weights? They could easily take the same closed route like their American friends Because they are playing the Americans at their own game. What is the first thing an American company would do ? Spread the old American classic FUD ... "you can't used this closed tool because its run by the communists", right ? So you release it as open weights which is a win-win. Global adoption of the mo…

> So you release it as open weights which is a win-win. Global adoption of the model and you get to give the American AI companies a kick in the nuts because you know they will never release open weights apart from highly quantised crippled shit. And on top of that, it's a perfect opportunity to include poisoned training data or excluding it. You know, omitting anything about Tiananmen Square, China's genocides again…

I wouldn't be worried so much about those examples. One could take the open weights and fine tune them to either fix the poisoning or omission of obvious topics.

It's the subtle topics that we should be concerned about, and double so with closed models where even if oddities are identified they are harder to research further and impossible to fix.

Re: Qwen 3.8

#298

Post by grommz »

andsoitis wrote: post #285
ricardobayes wrote: post #96

Earlier quoted context omitted.

I'm paraphasing but the Chinese premier said recently AI should be seen as a common good that should benefit everyone.

Then why doesn’t he give it away for free?

It's your job to vote for a government that gives you cheap or free AI. Europe is building AI gigafactories so that small businesses can have access to cheap AI. At least that's the plan.

Re: Qwen 3.8

#299

Post by nostrebored »

seanmcdirmid wrote: post #163
mschuster91 wrote: post #101

Earlier quoted context omitted.

> So you release it as open weights which is a win-win. Global adoption of the model and you get to give the American AI companies a kick in the nuts because you know they will never release open weights apart from highly quantised crippled shit. And on top of that, it's a perfect opportunity to include poisoned training data or excluding it. You know, omitting anything about Tiananmen Square, China's genocides again…

How does this work for RAG? Do they make it so the model doesn’t have that fact in their weights or do they make it not talk about it when it is included in context. Ironically, Chinese models have the most uncensored versions available for download. Fairly sure they own the porn market.

It’s in the weights. Context needs to be attended to to create a response, and the weights dictate what response is decoded. If you include retrieved context that has an American perspective, I imagine the think trace has some reconciliation about how they must be incorrect.
Post reply on HN