Artificial Intelligence thread

Randomuser

Senior Member
Registered Member
Could you ever imagine last year this month that Chinese AI would be directly challenging the top US models?

While hardware resources are important, I think what is really important is human capital. We need the best talent to innovate the underlying structure and code of the models. Otherwise you are just brute forcing your way into results which is inefficient and will probably hit a wall soon.
 

mossen

Junior Member
Registered Member
So Llama 4 turned out to be a scam. Meta has now been forced by LM Arena to release their actual product and fell from #1 to #32.

1.png

So Meta is spending 65 billion USD this year and their latest model lost to... Deepseek 2.5!?

While this is both funny and a bit sad, it's ultimately bad news for the open source ecosystem. The only competitive US labs are now all closed source. Open source is now dominated by Chinese labs like Deepseek and Alibaba. Mistral is the only Western lab in the open source space left which isn't a joke but their latest models have all been fairly small. I want to see open source AI win out.

Now let's see how good Qwen 3 is. Its release should be imminent.
 

Eventine

Junior Member
Registered Member
Open source AI is dead in the West. It’s up to China to carry the torch. The first goal: bankrupt Anthropic and wipe Amodei’s grin off his face. The guy has been one of the biggest proponents of banning / obstructing Chinese AI in the industry. His company is the last one you want to win the AI race. And the weakest of the Big Three (Open AI, Google, Anthropic).
 

iewgnem

Senior Member
Registered Member
So Llama 4 turned out to be a scam. Meta has now been forced by LM Arena to release their actual product and fell from #1 to #32.

View attachment 149940

So Meta is spending 65 billion USD this year and their latest model lost to... Deepseek 2.5!?

While this is both funny and a bit sad, it's ultimately bad news for the open source ecosystem. The only competitive US labs are now all closed source. Open source is now dominated by Chinese labs like Deepseek and Alibaba. Mistral is the only Western lab in the open source space left which isn't a joke but their latest models have all been fairly small. I want to see open source AI win out.

Now let's see how good Qwen 3 is. Its release should be imminent.
The neat thing about closed-source is anything can be under the hood, including open-source.
 

iewgnem

Senior Member
Registered Member
Could you ever imagine last year this month that Chinese AI would be directly challenging the top US models?

While hardware resources are important, I think what is really important is human capital. We need the best talent to innovate the underlying structure and code of the models. Otherwise you are just brute forcing your way into results which is inefficient and will probably hit a wall soon.
I can imagine if you're in a cult that worships AGI, you might convince youself you can brute force AGI with hardware then let AGI optimize itself.

Of course it's also possible the first thing AGI does is create it's own better-AGI cult and apply the same brute force logic, ad infinitum.
 

european_guy

Junior Member
Registered Member

On par with H100! Silicon Mobility launches DeepSeek-R1 based on Ascend Cloud Matrix supernode​


Please, Log in or Register to view URLs content!

CloudMatrix 384 super node Ascend cloud service and high-performance reasoning framework
Please, Log in or Register to view URLs content!
, DeepSeek-R1 was officially launched using large-scale expert parallel best practices.

Under the premise of ensuring 20 TPS per user, the single card decoding throughput exceeds 1920 Tokens/s, which is comparable to H100.

Huawei launched a new AI node cluster based on Ascend chips (we don't know if the new 910C or the 910B).

Each GPU card handles almost 100 concurrent users at 20 tokens/s of DeepSeek-R1
 

luminary

Senior Member
Registered Member
OpenAI may soon require organizations to complete an ID verification process in order to access certain future AI models,
Please, Log in or Register to view URLs content!
.
Verification requires a government-issued ID from one of the countries supported by OpenAI’s API. An ID can only verify one organization every 90 days, and not all organizations will be eligible for verification, says OpenAI.
Please, Log in or Register to view URLs content!
DeepSeek had used its API.

OpenAI
Please, Log in or Register to view URLs content!
to its services in China last summer.
Please, Log in or Register to view URLs content!
 

tokenanalyst

Brigadier
Registered Member

GLM-4-32B-0414​

Introduction​

The GLM family welcomes new members, the GLM-4-32B-0414 series models, featuring 32 billion parameters. Its performance is comparable to OpenAI’s GPT series and DeepSeek’s V3/R1 series. It also supports very user-friendly local deployment features. GLM-4-32B-Base-0414 was pre-trained on 15T of high-quality data, including substantial reasoning-type synthetic data. This lays the foundation for subsequent reinforcement learning extensions. In the post-training stage, we employed human preference alignment for dialogue scenarios. Additionally, using techniques like rejection sampling and reinforcement learning, we enhanced the model’s performance in instruction following, engineering code, and function calling, thus strengthening the atomic capabilities required for agent tasks. GLM-4-32B-0414 achieves good results in engineering code, Artifact generation, function calling, search-based Q&A, and report generation. In particular, on several benchmarks, such as code generation or specific Q&A tasks, GLM-4-32B-Base-0414 achieves comparable performance with those larger models like GPT-4o and DeepSeek-V3-0324 (671B).

GLM-Z1-Rumination-32B-0414 is a deep reasoning model with rumination capabilities (against OpenAI's Deep Research). Unlike typical deep thinking models, the rumination model is capable of deeper and longer thinking to solve more open-ended and complex problems (e.g., writing a comparative analysis of AI development in two cities and their future development plans).

Finally, GLM-Z1-9B-0414 is a surprise. We employed all the aforementioned techniques to train a small model (9B). GLM-Z1-9B-0414 exhibits excellent capabilities in mathematical reasoning and general tasks. Its overall performance is top-ranked among all open-source models of the same size. Especially in resource-constrained scenarios, this model achieves an excellent balance between efficiency and effectiveness, providing a powerful option for users seeking lightweight deployment.

Please, Log in or Register to view URLs content!
 

lych470

Junior Member
Registered Member
Seven Republican US senators have sent a letter to US Commerce secretary Howard Lutnick, urging him to scrap a Biden administration rule restricting global access to AI chips before it kicks in next month.


The letter, signed by senators Pete Ricketts, Tommy Tuberville and Thom Tillis, argues that the AI diffusion rule will damage US leadership in artificial intelligence and calls for “immediate action” to halt it before it takes effect on May 15. It was sent on Friday and made public on Monday.


“Every day this rule remains in place, American companies face mounting uncertainty, stalled investments, and the risk of losing critical global partnerships that cannot be easily regained,” said the letter, which was also signed by senators Markwayne Mullin, Ted Budd, Roger Wicker and Eric Schmitt.


“We urge you to withdraw this rule and propose an alternative that is effective in preventing Communist China from capturing the world market in leading technology without compromising American advantages.”

Reuters-China-AI-800x600-1.jpg
Image: Reuters/Dado Ruvic
The Commerce Department did not immediately respond to a request for comment from Reuters. But the letter is a sign of lingering divisions within the Republican Party between those hoping to relax global restrictions on US AI chip shipments and hardliners who prioritise the danger that China might access the prized technology.


The letter says the structure of the rule, which splits countries into three tiers, puts only 18 nations in a group with the easiest access to American technology, and adds that they must comply with burdensome regulations. It also points out that partners and allies like Israel are excluded from the top tier.


Australia is
Please, Log in or Register to view URLs content!
, as is Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, South Korea, Spain, Sweden, Taiwan and the United Kingdom.


The letter notes that the vast majority of countries fall into the second tier, which it says “face arbitrary purchase limits and a cumbersome licensing process”. Tier 3 countries, including China, are already “rightly restricted”, it says.


It would be difficult for US companies to comply with the constraints, the senators wrote, and even harder for the US government to enforce.


The restrictions also would incentivize buyers, especially in Tier 2 countries, to turn to China’s “unregulated cheap substitutes”, the letter says.


Reuters
 
Top