Anyone use the Deepseek R1 model yet? I tried it yesterday and it didn’t seem all that better than existing LLMs. Is that a limitation of the free model?

Reply to this note

Please Login to reply.

Discussion

Front running the China censorship slew of comments

It’s not great. I get too many hallucinations with it

note1wsmtalu5ql7uwu80pu0xzyevx8rc4ssmjg7txvhew6strmah8nkq8fwqzg

With bitcoin questions I’ve only had decent results with spirit of Satoshi, the training data out there about bitcoin is garbage, but it’s an interesting test case 💪

Yeah the preface rambling before the real answer is annoying

I guess it’s called reinforced learning, but it’s more like reinforced confusion

It’s also passive aggressive sometimes. “The user seems to be under the impression that I somehow have access to this information…” 😂

I haven’t. From what I’ve heard it’s supposed to be on par with existing LLMs. Given how much they spent on training. I.e fraction of the cost

I think it's more focused on logic reasoning like the O1 models for example for coding tasks. Just as an answer machine it's not the right tool for the jop.

Y3ah I find it pretty amazing to he honest. 32B model on laptop is honestly amazing. Not quite as good as o1, but not as annoying either