they just released the first small models that i would consider even vaguely articulate for edge inference involving a human. maybe they want to do a mistral and raise a kajillion and work from their home town?
there were no posts in this thread when i linked to the previous. It had the underlying (and additional) information in that reddit thread. I agree though it was of limited utility to do so.
I feel someone just gave them a huge $$$ offer that they couldn't say no too. Given Elon Musk is praising their efforts, and he lost a lot of his original XAI team recently, my money is on Elon.
Is there a leading American AI research organization - big tech or academia - that isn't "full of Chinese Nationals"? If the DoD want an all-American SoTA model, they may have to wait for a while.
This feels like a typical sociopathic corporate scenario with mushroom management - let a bunch of nerds develop something new/exciting outside mainstream corporate culture, then once it becomes good enough jump in, cut them off and harvest whatever they produced while reaping all benefits/credits for yourself, then live off mediocre subsequent releases for a while while blaming remaining team members for future failures.
Somewhat of a devil's advocate here, because I am very familar with corporate idiocy. But how do you define a non-sociopathic corporate scenario where a company makes a lot of money from a good product they develop? Even if done in maximally practically and emotionally intelligent way, this still requires changes from research phase no?
Do you have evidence that they were sacked rather than resigned because they would rather work in a different direction from the one company is taking?
But how do you define a non-sociopathic corporate scenario
Corporate structures are sort of sociopathic by default. Theres no empathy globule on the corporate hierarchy and everyone is motivated to put the corporations interests first.
This isnt even a criticism really, its just the reality. Corporations are like, paperclip maximising AI's, but for shareholder profit.
Which is insane. Obviously he isnt the only lead at alibaba, but qwen as consequence lost many talented people by doing this. It will negatively impact the Qwen team.
Qwen4 is going to flop like Lllama 4.
I hope those who just quit form their own new lab and start building again.
Maybe, maybe not. It's not even clear if there's gonna be a new qwen, and if they'll keep open sourcing it. It also depends on what the team coming from gemini brings to the table. People move around, and things get shared. Happened before with grok, will likely happen with qwen. Everyone wants what the OG teams have.
Mistral was ex llama people. And after their good start, they've kinda plateaued lately. Their latest open models have been quite disappointing. Nothing revolutionary at any rate.
People said about the gemini team that moved to xai that they were "amateurs". And yet they delivered in about 1 year with grok4, was SotA for a few weeks at launch. They now lost some people, and likely will get others.
Round and round the people move around, and everyone gets most of the things that everyone else uses. I have no doubt that the qwen team will get to find a cozy place to call home for a while...
'People said about the gemini team that moved to xai that they were "amateurs". And yet they delivered in about 1 year with grok4, was SotA for a few weeks'
Wait a minute, this is the same company that is sueing OpenAI for... pretty much this?
Now that's the awkward part: qwen was too successful, a team had more influence than Tongyi and even Aliyun, obviously they started to develop multi-modal capabilities overlapping other team's work, even with their own app, very vertical integrated.
But qwen didn't contribute much revenue or DAU/MAU except fame amoung AI communities. Since GPUs were scarce, Alibaba had to balance free open weight models and customer use.
In a better world, alibaba might just as well split qwen into an independent entity and IPO it.
Sounds very plausible to me too. Because even if you refocus the business unit it makes no sense to lay off a highly capable team. Finding new people, integrating them into the team - all that costs a lot of time and money and there is no guarantee for success.
Definitely plenty of people further up the corporate ladder were not happy with the success, while the top is likely too far disconnected to understand.
This is similar to what happened to OnePlus at OPPO. OnePlus was outshining OPPO in the global market - there was absolutely no way the OPPO brand could compete other than price, and its international expansion would have looked like a failure side by side. (Source: Worked there).
Happening to Skoda in Volkswagen - German managers were first forcing Skoda to have dumb design to not threaten VW/Audi, that didn't work, Skoda became the only profitable part of VW, now they are imposing 25% cost reduction at Skoda as well as the company "must save money", despite generating all the profit.
Well, Alibaba hiring a Gemini guy to run Qwen suggests they want to make Qwen into a big consumer / enterprise business like Gemini. I am not sure that I blame them even if it clashes with how their top researchers were hoping Qwen would be run. Most obviously it's natural for a company to want to make money on things they paid for developing. But also, the world needs more competition in AI businesses just like it needs competition in AI research. I wouldn't mind Qwen code to grow into a commercial grade competitor to Claude code that is better, faster and cheaper. I am sure the talented researchers can find a new home in Moonshot AI or even US college or startup.
they just released the first small models that i would consider even vaguely articulate for edge inference involving a human. maybe they want to do a mistral and raise a kajillion and work from their home town?
What does do a mistral mean?
MistralAI is known for their smaller models on the edge, to avoid competing with Gemini & OpenAI directly.
Who knows if OpenAI will do a refresh, but gpt-oss-20B/120B are still some of the best edge models so far.
oh?! what do they handle well? how do they fail?
the 3.5 9b model on my laptop at full fp8 is outlandish in its seeming reasoning capacity, though i haven’t really stress tested it
Subject submitted yesterday fwiw:
https://news.ycombinator.com/item?id=47236390
timing is everything, but linking back to a submission you made with no comments in a thread with more posts is pointless
there were no posts in this thread when i linked to the previous. It had the underlying (and additional) information in that reddit thread. I agree though it was of limited utility to do so.
I feel someone just gave them a huge $$$ offer that they couldn't say no too. Given Elon Musk is praising their efforts, and he lost a lot of his original XAI team recently, my money is on Elon.
FWIW, the article quotes people close to the researchers saying that the departure wasn't their choice.
Nobody with talent wants to work with Elon or xAI
The only people he attracts are h1b candidates who have limited choices
You can broaden that. A lot of talent doesn't want anything to do with the USA.
The US is the number one destination for skilled immigrants on the planet. Nobody else pays as well.
Do you not follow the news?
https://theconversation.com/skilled-migrants-are-leaving-the...
That trend is reversing, because of course it is.
A bit hard for him to swing if he wants to position xAI as a key Defense contractor for AI and his company is full of Chinese Nationals...
Is there a leading American AI research organization - big tech or academia - that isn't "full of Chinese Nationals"? If the DoD want an all-American SoTA model, they may have to wait for a while.
Maybe true but you probably expect a minimum of leaked information directly to the CCP if you accept that.
AI Research orgs voluntarily do their own "leaking" via the publication of research papers, and involuntarily via employee/post-doc churn.
The guilty-until-proven innocent mindset is what kickstarted China's nuclear program - and led to the internment of people with Japanese heritage.
This feels like a typical sociopathic corporate scenario with mushroom management - let a bunch of nerds develop something new/exciting outside mainstream corporate culture, then once it becomes good enough jump in, cut them off and harvest whatever they produced while reaping all benefits/credits for yourself, then live off mediocre subsequent releases for a while while blaming remaining team members for future failures.
Not even corporate, that's just human behaviour/history.
We love to elect chest-beating leaders.
Somewhat of a devil's advocate here, because I am very familar with corporate idiocy. But how do you define a non-sociopathic corporate scenario where a company makes a lot of money from a good product they develop? Even if done in maximally practically and emotionally intelligent way, this still requires changes from research phase no?
Developing a business product and monetizing it doesn't require sacking the research group that created it and would have developed it further.
Do you have evidence that they were sacked rather than resigned because they would rather work in a different direction from the one company is taking?
The article directly mentions it; see the section 'Leaving wasn't your choice'.
Corporate structures are sort of sociopathic by default. Theres no empathy globule on the corporate hierarchy and everyone is motivated to put the corporations interests first.
This isnt even a criticism really, its just the reality. Corporations are like, paperclip maximising AI's, but for shareholder profit.
From what I read, he was fired.
Which is insane. Obviously he isnt the only lead at alibaba, but qwen as consequence lost many talented people by doing this. It will negatively impact the Qwen team.
Qwen4 is going to flop like Lllama 4.
I hope those who just quit form their own new lab and start building again.
Maybe, maybe not. It's not even clear if there's gonna be a new qwen, and if they'll keep open sourcing it. It also depends on what the team coming from gemini brings to the table. People move around, and things get shared. Happened before with grok, will likely happen with qwen. Everyone wants what the OG teams have.
Mistral was ex llama people. And after their good start, they've kinda plateaued lately. Their latest open models have been quite disappointing. Nothing revolutionary at any rate.
People said about the gemini team that moved to xai that they were "amateurs". And yet they delivered in about 1 year with grok4, was SotA for a few weeks at launch. They now lost some people, and likely will get others.
Round and round the people move around, and everyone gets most of the things that everyone else uses. I have no doubt that the qwen team will get to find a cozy place to call home for a while...
Wait a minute, this is the same company that is sueing OpenAI for... pretty much this?
If you're being called incredibly dense by Elon these days, you must be doing something right. :D
Qwen 3.5 was an epic drop.
Qwen3.5 27b is frontier intelligence density.
GPT 20b high used to be top; and only barely inches out 35b.
glm 4.7 flash strong 4th place.
nemotron 30b 4th place.
Its very interesting that GLM 5 didnt manage to get the same density as they were able to with 4.7 flash.
The Qwen team has been putting out great releases lately. I hope that they can continue on that path despite this.
Perhaps forced to move over to building military and intelligence models.
big corp politics.
qwen started as the core team of Tongyi dept, which was part of the algorithm & model offerings from Alibaba-Cloud (aliyun)
https://tongyi.aliyun.com/
- qwen series for nlp/LLM
- wan series for cv/video
- fun series for ASR and TTS
- icss for AI customer service
- lingma for coding
- xingchen for metaverse hype
- tingwu for podcast/notebooklm
Now that's the awkward part: qwen was too successful, a team had more influence than Tongyi and even Aliyun, obviously they started to develop multi-modal capabilities overlapping other team's work, even with their own app, very vertical integrated.
But qwen didn't contribute much revenue or DAU/MAU except fame amoung AI communities. Since GPUs were scarce, Alibaba had to balance free open weight models and customer use.
In a better world, alibaba might just as well split qwen into an independent entity and IPO it.
Sounds very plausible to me too. Because even if you refocus the business unit it makes no sense to lay off a highly capable team. Finding new people, integrating them into the team - all that costs a lot of time and money and there is no guarantee for success.
Definitely plenty of people further up the corporate ladder were not happy with the success, while the top is likely too far disconnected to understand.
This is similar to what happened to OnePlus at OPPO. OnePlus was outshining OPPO in the global market - there was absolutely no way the OPPO brand could compete other than price, and its international expansion would have looked like a failure side by side. (Source: Worked there).
Happening to Skoda in Volkswagen - German managers were first forcing Skoda to have dumb design to not threaten VW/Audi, that didn't work, Skoda became the only profitable part of VW, now they are imposing 25% cost reduction at Skoda as well as the company "must save money", despite generating all the profit.
Hello EST, I saw your rap video yesterday, you should link it here it was pretty good
I hope the people from the Qwen team start their own thing or something... But regardless, the work they did will live on as legendary.
Well, Alibaba hiring a Gemini guy to run Qwen suggests they want to make Qwen into a big consumer / enterprise business like Gemini. I am not sure that I blame them even if it clashes with how their top researchers were hoping Qwen would be run. Most obviously it's natural for a company to want to make money on things they paid for developing. But also, the world needs more competition in AI businesses just like it needs competition in AI research. I wouldn't mind Qwen code to grow into a commercial grade competitor to Claude code that is better, faster and cheaper. I am sure the talented researchers can find a new home in Moonshot AI or even US college or startup.
Well…Gemini sucks so I don’t know why you would hire someone from that team to lead your AI team.
but nano banana is so much fun!
alao discussed here https://news.ycombinator.com/item?id=47249343