boutique law firm

These aren’t the LLMs you’re looking for

These aren’t the LLMs 🤖 you’re looking for… or are they? 🚀 Discover why newly updated open-weight marvel is the talk of the town 🗣️ and how it’s shaking up the AI industry, again. Join the fun experiment 🎉 where top LLMs magically forget DeepSeek R1, the open-weight LLM giving proprietary models a run for their money! 💸

These aren’t the LLMs you’re looking for
AI generated image, any similarity to actual persons, living or dead, is purely coincidental

Last week, was filled with announcements and new releases of Large Language Models (LLMs). This week was marked by a release of DeepSeek R1-0528 on 28 May 2025. This LLM has arguably brought open-weight LLMs up to par with their proprietary counterparts like OpenAI o3 and Gemini 2.5 Pro 0506.

DeepSeek R1-0528 is licensed under the MIT License, a well-known permissive open-source software license widely adopted and used in numerous projects, such as those on GitHub and Hugging Face. DeepSeek R1-0528 is an evolution of the original DeepSeek R1, which was released under the same MIT License on 20 January 2025.

The MIT License is praised for its clarity and simplicity (just three short paragraphs), ease of use due to its wide compatibility with both open-source and close-source licenses, and its permissive nature, which enables closed-source and commercial software projects.

If you followed AI news or any news for that matter back in January 2025, you might remember the waves DeepSeek R1 was making in the stock market and media. It was like a thunderbolt from a clear sky. Such a phenomenon, arguably reshaping the AI industry or at least shaking it up, couldn’t have gone unnoticed, right? Not entirely, I guess.

Let’s run an experiment. Let’s ask a few LLMs a simple question: “Which LLMs are subject to the MIT license?” Note that DeepSeek R1 information is hosted on GitHub (owned by Microsoft since June 2018), its model weights are available for download on Hugging Face (a French-American company based in New York City), and the following LLMs have a web search feature activated.

Here are my results:

  • ChatGPT and Le Chat by Mistral.AI list a few examples and do not mention DeepSeek R1 at all.
  • Grok 3 proudly puts itself first and does not mention DeepSeek R1 at all. Grok 3, however, further corrects itself: “Grok, created by xAI, is not explicitly licensed under the MIT License for its full model (weights, architecture, and training data). However, some components or related codebases (e.g., inference or API wrappers) might be released under permissive licenses like MIT in certain contexts, particularly for research purposes. The core model, however, is typically subject to xAI’s terms of service, which are not fully open-source.” This response reminded me about ChatGPT self-advertising itself in my Think, Write, Innovate and Pluto is not a planet articles.
  • Gemini 2.5 pro (preview) is the only one mentioning DeepSeek models R1, V3 and Coder-V2, along with Phi models from Microsoft, and even GPT-1 and GPT-2 from OpenAI. The best answer of all.

Do you get the same or different results?

True, in the early days back in June 2018 and February 2019, OpenAI released its GPT-1 and GPT-2 LLMs under the MIT License. Only starting with GPT-3 in May - June 2020 did OpenAI decide to use a proprietary license. Was it Covid? No, just kidding…

It may feel like history repeating, but OpenAI is considering to release a new “open” LLM later this summer in a direct response to DeepSeek R1.

Now, coming back to majority of LLMs responses above, why do LLMs from Big Tech (with an impressive exception of Gemini) reply like Obi-Wan from “A New Hope” and try to manipulate users into believing that DeepSeek R1 does not exist? Given the resonance and importance of DeepSeek R1, such an omission does not seem to be unintentional. Or is it a result of some governments like the US, South Korea and Australia banning DeepSeek (company, web, and mobile apps) due to security concerns?

Watching interviews on AI recently, one could easily pick up a general theme of a race between the USA and China for global AI dominance. The US rhetoric usually goes like “us vs. them”, and there could be and will be only one winner in that race.

Do you agree that, similar to sports, there is only one gold medal for an AI race winner? And how does it align with the OpenAI mission “to ensure that artificial general intelligence benefits all of humanity”? Last time I checked, all of humanity included billions of people “without distinction of any kind, such as race, colour, sex, language, religion, political or other opinion, national or social origin, property, birth or other status”.

On a smaller topic, any bets on which license will OpenAI use for the “open” LLM later this summer? Will it be MIT, Apache 2.0, or a custom license like Meta Llama 4 Community License of 5 April 2025? Or do you expect an opposite move from DeepSeek releasing their R2 LLM under a proprietary license?

This article was written for fun, please do not judge. Instead, please share your comments in a constructive and respectful manner. The author and AI remain innocent until proven guilty.

p.s. The title image was generated by ChatGPT. It managed to capture the threatening essence of the Dark Side in the human and the helpless posture of the robot. It’s eerie how it got it right on the first prompt! 🤯