The Illusion of Thinking: Strengths and Limitations of Reasoning Models

"ChatGPT killer" sounds really scary the first time. But once everyone has reacted to it and figures it's just as disappointing as ChatGPT, then they won't have any reason to react the next time it's lauded: https://en.wikipedia.org/wiki/China's_final_warning

kamikazechaser•7h ago

There is a defacto shunning of it. But people are using it, MAI-DS-R1 https://huggingface.co/microsoft/MAI-DS-R1 among others are alternatives of the same thing with all the blocked topics removed.

atemerev•7h ago

What? It is on OpenRouter and HuggingFace. You can download it and run it locally on your own machine. Unless you work for the government, nobody will deny you that.

starship731•6h ago

What utter and complete rubbish. Deepseek is available on all the major LLM API hosts, from TogetherAI, to Replicate, to Fireworks, to OpenRouter, to NVIDIA NIM, to Groq, to Hugging Face, to GitHub, to Cloudflare Workers AI. Hell, even Scaleway, a FRENCH provider, offers it.

Pls do a bit of investigation before posting nonsense. Low information commenters like you detract from the conversation.

starship731•6h ago

Deepseek R1, the Reasoning Model, never had a 'moment' when it was released.

It was DeepSeek, the Mobile App that consumers (including stock traders, analysts, & investors who don't follow the day-to-day developments within the LLM field) could download from App Stores that had the 'moment'.

And it was the AI chat app that led the professional talking heads to learn about the Language Model (DeepSeek R1) powering the Chat App, the Scientific Paper explaining the techniques used to build the core model (Deepseek V3), and the said paper's claims of the model being developed at a far cheaper amount than comparable Western models.