Using DeepSeek from Italy

5 hours after Italy shuts down access to DeepSeek from the country as a whole, we're giving you a backdoor using DeepSeek-R1 through HuggingFace. You can try it below. Change the type in the select dropdown to "deepseek-r1" to use DeepSeek instead of OpenAI.

• Try DeepSeek

Is it good?

Deepseek-R1 is really good. On benchmarks it scores on pair with OpenAI's O1, and sometimes even outperforms it. It's not as easy to make it correctly execute functions as OpenAI's models, but in general it's very good, and I suspect this is a prompt engineering issue. It's also sloooow, but I suspect that's a HuggingFace problem and not a DeepSeek problem. If you want to watch me play with it, you can watch the following video.

Is it Spying on You?

Even though Italy's government apparently doesn't like DeepSeek due to its ties to China, the paradox is that our solutions probably contains much less spying than OpenAI. This is because we're actually using DeepSeek through HuggingFace, with options to run it through Ollama, allowing us to deliver self-hosted solutions for clients with deep pockets to purchase GPUs.

This is only possible because of DeepSeek publishing their LLM as open source under the terms of the MIT license. This allows anyone to download the LLM, fine tune it, and host it on their own infrastructure. This completely eliminates all spying, including spying done by OpenAI.

This is a really big deal, and crucial for small independent software vendors such as us, since it allows us to deliver high quality services, through the entire stack, with quality on pair with the top dogs from Nasdaq, including Microsoft, Amazon, OpenAI, and Google.

The importance of what DeepSeek did here cannot be underestimated. It basically completely levels the playing field, allowing companies such as AINIRO to deliver AI Agent-based systems, on pair and sometimes better than what giants such as Google and Microsoft can do, and deliver the LLM itself too. Sure, you'll need deep pockets to host your own LLM, and we're talking several thousands per month - But this is still "peanuts" compared to the 100 million dollars OpenAI spent to train GPT-4, and possibly billions of dollars they spent on O1.

Thank you DeepSeek 🥰

If you want to use DeepSeek-R1 on your own infrastructure, and/or through HuggingFace or Ollama, feel free to toss us a message. If you're an Italian company we would especially enjoy it if you contacted us, for obvious reasons 😊

And yes, we are GDPR compliant!

Have a Custom AI Solution

At AINIRO we specialise in delivering custom AI solutions and AI chatbots with AI agent features. If you want to talk to us about how we can help you implement your next custom AI solution, you can reach out to us below.

• Contact AINIRO