How much does the number of models by region represent the impact.? Deepseek is in Hangzhou in Zhejiang which has a vibrant tech scene(Alibaba and Qwen I guess would also be Hangzhou )but a relatively low percentage of models
Non-Generative AI = Analyzer (processes existing data)
Both are shaping the future, and together they’re making technology smarter, faster, and more human-like. Whether you want to generate a story or predict tomorrow’s stock price, AI has a tool for you.
Learning them now can give you a future-proof career in one of the fastest-growing industries worldwide.
AI models generalize well if they are trained on a large amount of data. If we talk about the text-generating AI models, they need an enormous amount of text to better understand all the possible contexts and hence generate accurate responses. These texts have to be processed first and then used in a deep neural network model (e.g., transformers having billions of parameters). Training such a model would require lots of expensive hardware resources and electrical power.
Additionally, challenges such as inference latency, model size, hallucinations, and deployment constraints make optimization difficult.
Not at all a tech/AI person so a lot of this went over my head, but I’ve heard rumblings in the China policy and Sinosphere about how Chinese AI like DeepSeek also heavily pick up on government censorship and factual inaccuracies that are innate to the state-controlled Chinese Internet. This isn’t necessarily caused by data used to train the model originally but since DeepSeek’s release to the public, due to actual usage and interactions with existing state controls on the web, it’s becoming more and more influenced by it. Not sure how true this is at all and could just be people talking smack. Would love an actual expert opinion.
How much does the number of models by region represent the impact.? Deepseek is in Hangzhou in Zhejiang which has a vibrant tech scene(Alibaba and Qwen I guess would also be Hangzhou )but a relatively low percentage of models
Generative AI = Creator (makes new content)
Non-Generative AI = Analyzer (processes existing data)
Both are shaping the future, and together they’re making technology smarter, faster, and more human-like. Whether you want to generate a story or predict tomorrow’s stock price, AI has a tool for you.
Learning them now can give you a future-proof career in one of the fastest-growing industries worldwide.
AI models generalize well if they are trained on a large amount of data. If we talk about the text-generating AI models, they need an enormous amount of text to better understand all the possible contexts and hence generate accurate responses. These texts have to be processed first and then used in a deep neural network model (e.g., transformers having billions of parameters). Training such a model would require lots of expensive hardware resources and electrical power.
Additionally, challenges such as inference latency, model size, hallucinations, and deployment constraints make optimization difficult.
Not at all a tech/AI person so a lot of this went over my head, but I’ve heard rumblings in the China policy and Sinosphere about how Chinese AI like DeepSeek also heavily pick up on government censorship and factual inaccuracies that are innate to the state-controlled Chinese Internet. This isn’t necessarily caused by data used to train the model originally but since DeepSeek’s release to the public, due to actual usage and interactions with existing state controls on the web, it’s becoming more and more influenced by it. Not sure how true this is at all and could just be people talking smack. Would love an actual expert opinion.
I wrote sth on the process of how all models in China are evaluated (censored) by the government. Should be somewhat helpful: https://ocpl.substack.com/p/172affb0-9a50-4007-99ac-4953a82e99ee