Yup, Google is probably almost gleeful, for now, that the likes of deepseek and qwen are helping it clear out space from its competitors....
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
Did Deepseek use data from western sources abt this thing or something?
For English prompts, it will have to be trained on English data, which naturally has mostly western sources
Reasoning is the big one to await for. In their paper they mentioned specifically how their R1 reasoning model helped them a lot on generating synthetic data.
If R1 was so helpful, which I don't rate it too highly tbh, then I can only imagine the leap forward with a R2 reasoning model fine tuned on a DeepSeek V3 LLM
It's fascinating how American are so devoid of individuality and mental agency, even after witnessing objective demonstration to the contrary they're still convinced of their indoctrination to "worry" about DeepSeek.
an indian OpenAI employee lol
I guess when you're outcompeted by a scale of 300x the only thing left to do is cope.
Well guess what, if you had done GPT o1 correctly, maybe you would have also gotten the o1 visa you want.
When you can't compete, just change the benchmarks:
I guess when you're outcompeted by a scale of 300x the only thing left to do is cope.
Well guess what, if you had done GPT o1 correctly, maybe you would have also gotten the o1 visa you want.
When you can't compete, just change the benchmarks:
ooh lol you can just change the AI benchmark, if that isn't a "scam VC out of money, for big bonus" tech then i don't know what is.
Something is fishy with the results. Gemini 1.5 flash and Grok-beta are actually hot garbage.
Also no way Gemma 2 9b is better
Whole thing smells. Smells bad