Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Indian billionaire allegedly calling for indian "AI independence"
No, Gemini's attention can only process 32k of context; 1M is a lie.
64k is good but not enough
I wrote up a long thread here on the effect of Minimax and Zai models and why they are such a big deal on software industry.
People that are angry that China isn't importing a lot of GPUs should look at the price of RAM and GPUs and think for at least a moment that if their current Phone/PC broke and they needed to replace it/replace the RAM or GPU would they have enough money to actually afford it, since if China start buying GPUs like it's candy for building data centers just for the purpose of scaling their models to 10's of trillions of parameters with the belief that just scaling will get them to "Artificial General Intelligence" will increase the price of both GPUs and RAM even more than the prices we have now. They should be happy that China is thinking rationally and not like the billionaires in Silicon Valley who are desperately trying to build "God" in the form of "AGI" thinking it would save them from the eventual uprising of the working class.
Pardon for OOT.
But honestly they believed that if China doesn't parameter and performance maxxing, Trump would unleash latest opus model to cyber attack whole China. Having 80-90% performance with 10% price isn't enough
There is a fast take off scenario, however small probability, where AI can perform original research, make new AI-related discoveries and then self-improve. China absolutely needs some teams working on this, but it's likely a different path than what Anthropic or OpenAI is doing with their large mainstream models requiring GW-scale data centers.
Like Minimax focusing on small coding models, these teams can build a specialized model with custom training data that focuses on AI research to self-improve. I believe Bytedance Seed and DeepSeek are already working on research-specific models
Like Minimax focusing on small coding models, these teams can build a specialized model with custom training data that focuses on AI research to self-improve. I believe Bytedance Seed and DeepSeek are already working on research-specific models
Even the best AI hallucinates like crazy with even little bit of poking. These LLM are not even remotely reliable without heavy human involvement in coding.
AGI is not possible with LLMs. A fundamental rethinking will be needed.