Artificial Intelligence thread

[solarz]

the question is why is anyone still using the free version of ChatGPT when there are better Chinese models around that are free?

https://twitter.com/i/web/status/2042252748050067805

Anyways, ByteDance with new end to end speaking model called Seeduplex. All the description and measures look good. Curious, how it turns out.

Because they’re obviously trained in different datasets. Chinese LLM assume by default your context is Chinese, while GPT is moronic when it comes to Chinese subject matters.

Also the biggest advantage of ChatGPT is the voice recognition. No other models I’ve used comes close.

 

[tphuang]

Because they’re obviously trained in different datasets. Chinese LLM assume by default your context is Chinese, while GPT is moronic when it comes to Chinese subject matters.

Also the biggest advantage of ChatGPT is the voice recognition. No other models I’ve used comes close.

you can just type out your prompt.

Most of my requests are in English and the Chinese models can handle them with no problem.

Please, Log in or Register to view URLs content!

数据中心是用电增量大户。据中国信息通信研究院数据，2025年我国算力中心用电量已达1960亿千瓦时，同比增长300亿千瓦时，增速高达18.1%。预计到2030年，我国数据中心用电量将超7000亿千瓦时，占全社会用电量5%以上。

196 TWh of electricity was used by data center in China in 2025.

 

[solarz]

you can just type out your prompt.

Most of my requests are in English and the Chinese models can handle them with no problem.

Typing is too inefficient for the way I use Chat GPT. GPT is also able to keep my location in context while Chinese LLMs keep reverting to assumptions about my nationality and location.

 

[tphuang]

Typing is too inefficient for the way I use Chat GPT. GPT is also able to keep my location in context while Chinese LLMs keep reverting to assumptions about my nationality and location.

Which Chinese LLMs have you tried out recently? have you tried GLM5.0-Turbo. I'm using it now and it's all responding to me in English.

 

[solarz]

Which Chinese LLMs have you tried out recently? have you tried GLM5.0-Turbo. I'm using it now and it's all responding to me in English.

It’s not about responding to me in English or Chinese, it’s about the content. For example, if I ask ChatGPT how to grow a fig tree, it will take into account my location and the appropriate climate. A Chinese LLM will instead give me a response about different Chinese regions.

 

[tphuang]

It’s not about responding to me in English or Chinese, it’s about the content. For example, if I ask ChatGPT how to grow a fig tree, it will take into account my location and the appropriate climate. A Chinese LLM will instead give me a response about different Chinese regions.

I don't get this at all. This is my most recent query on kimi

"Please analyze the effect on economy of Iran if they can make $90B per year for strait of Hormuz toll booth, increase of oil export to 3 million bpd of crude at full brent price and 1 million bpd of refined oil production, elimination of sanctions from Western countries and China making Iran an import rail transit hub to Europe."

And it gave me a great result and generated nice charts. What are you asking that Chinese AI cannot do? This is really weird.

 

[solarz]

I don't get this at all. This is my most recent query on kimi

"Please analyze the effect on economy of Iran if they can make $90B per year for strait of Hormuz toll booth, increase of oil export to 3 million bpd of crude at full brent price and 1 million bpd of refined oil production, elimination of sanctions from Western countries and China making Iran an import rail transit hub to Europe."

And it gave me a great result and generated nice charts. What are you asking that Chinese AI cannot do? This is really weird.

I guess you use AI to generate content?

I’m using AI to gather information. It would be extremely inefficient if I had to write prompts like yours everytime.

My prompts are things like:
- how do I grow asparagus?
- where are the best fishing spots in X?
- how do I fix a leaking faucet?
- what does this indicator mean on my car?

This is why context memory and voice parsing is very important.

 

[jli88]

I don't get this at all. This is my most recent query on kimi

"Please analyze the effect on economy of Iran if they can make $90B per year for strait of Hormuz toll booth, increase of oil export to 3 million bpd of crude at full brent price and 1 million bpd of refined oil production, elimination of sanctions from Western countries and China making Iran an import rail transit hub to Europe."

And it gave me a great result and generated nice charts. What are you asking that Chinese AI cannot do? This is really weird.

Have you tried coding? I can assure you that Claude models are absolutely the best (followed by latest OpenAI models) when it comes to general coding use cases, hence their widespread use by coders. If they could escape by using much cheaper Chinese models they would have, but unfortunately not.

By size of the models, Opus is rumored to be 5T, Mythos to be > 10T. And these are being trained on upwards of 100 T tokens.

By comparison, Chinese models are much smaller. Kimi K2.5 for example is 1.04 Trillion params, with training done on 15 T tokens. It is fantastic for its size. But China needs to scale right now!

 

[tphuang]

I guess you use AI to generate content?

I’m using AI to gather information. It would be extremely inefficient if I had to write prompts like yours everytime.

My prompts are things like:
- how do I grow asparagus?
- where are the best fishing spots in X?
- how do I fix a leaking faucet?
- what does this indicator mean on my car?

This is why context memory and voice parsing is very important.

I use AI for everything and kimi is pretty good for pretty much everything I ask it for.

Questions I've asked recently.
What is the appropriate age for the lion, the witch and the wardrobe?

How many books in this series?

what's the best way to take train from london city center to harry potter studio?

How many la liga did cristiano Ronaldo win?

never any issues with these questions.

 

[tphuang]

Have you tried coding? I can assure you that Claude models are absolutely the best (followed by latest OpenAI models) when it comes to general coding use cases, hence their widespread use by coders. If they could escape by using much cheaper Chinese models they would have, but unfortunately not.

By size of the models, Opus is rumored to be 5T, Mythos to be > 10T. And these are being trained on upwards of 100 T tokens.

By comparison, Chinese models are much smaller. Kimi K2.5 for example is 1.04 Trillion params, with training done on 15 T tokens. It is fantastic for its size. But China needs to scale right now!

yes, did you see my comment right above this?

seriously, have people actually used GLM-5.0-Turbo or GLM-5.1 before they write crap like this?

Today, my prompt on GLM-5.0-Turbo.

I need a python script that can access hyperliquid's spot api

Please, Log in or Register to view URLs content!

. I need it to be able to get a list of all the dexes and then access all the instruments on those dexes. + all the public endpoints that can be accessed with user address.

Give me the scripts, json files and everything in 5 minutes. Almost fully worked on a first pass.

See Design Arena

https://twitter.com/i/web/status/2041550551855595622