... and now another one when a (not so cheap tbh) Chinese model matched OpenAI GPT Sol and Anthropic Fable/GPT 4.8 (some scores below, some scores above) ...
Oh and exceeded Google's best internal model capability enough to have them cancel the next Gemini release. Probably. But I'm 95% certain that's exactly what happened. This is the first time I really fear for the future of Google, because OpenAI matched or exceeded the experience of Google search ... and now a Chinese company matched OpenAI ... followed by Google failing to catch up. That really made me take pause.
It indicates the the "secret sauce" of OpenAI and Anthropic is ... nothing, really. Or none of it matters, except access to the hardware needed for 1.5 trillion parameter models, which Chinese firms now have as well. It means there is no "AI takeoff" where other firms can't match OpenAI/Anthropic models. And Google, inventors of transformers just missed the takeoff. There's more secret sauce, but clearly a great engineering firm can find it in less than 3 months.
The fact that this Chinese model has similar cost to OpenAI and ballpark cost compared to Anthropic while we can be quite sure they're minimizing the cost (that's what China is doing everywhere else) also raises questions about the true cost of serving comparable models and with that about the future profitability of OpenAI and Anthropic.
January 2025, release of DeepSeek R1, the first open reasoning model. There was a lot of panic then that it was done with very few resources.
I guess when cheap Chinese model was very close to SoTA models
... and now another one when a (not so cheap tbh) Chinese model matched OpenAI GPT Sol and Anthropic Fable/GPT 4.8 (some scores below, some scores above) ...
Oh and exceeded Google's best internal model capability enough to have them cancel the next Gemini release. Probably. But I'm 95% certain that's exactly what happened. This is the first time I really fear for the future of Google, because OpenAI matched or exceeded the experience of Google search ... and now a Chinese company matched OpenAI ... followed by Google failing to catch up. That really made me take pause.
It indicates the the "secret sauce" of OpenAI and Anthropic is ... nothing, really. Or none of it matters, except access to the hardware needed for 1.5 trillion parameter models, which Chinese firms now have as well. It means there is no "AI takeoff" where other firms can't match OpenAI/Anthropic models. And Google, inventors of transformers just missed the takeoff. There's more secret sauce, but clearly a great engineering firm can find it in less than 3 months.
The fact that this Chinese model has similar cost to OpenAI and ballpark cost compared to Anthropic while we can be quite sure they're minimizing the cost (that's what China is doing everywhere else) also raises questions about the true cost of serving comparable models and with that about the future profitability of OpenAI and Anthropic.