lon Musk’s xAI released Grok-1’s AI model code and weights, taunting OpenAI in the process.
They shared it on GitHub and BitTorrent, giving folks all over the place the chance to work with it. Grok-1 is huge, with 314 billion bits of info to play with – that’s a whole lot more than what GPT-3 has.
xAI wants everyone to have a chance to work with the latest big language model tech. They’re giving away a version of Grok-1 that’s not polished yet, but it’s open for anyone to try out and use, even to make money.
This week, @xAI will open source Grok
— Elon Musk (@elonmusk) March 11, 2024
Musk couldn’t help but throw in some playful jabs about making Grok’s details public. When the ChatGPT X account said Grok’s announcement copied their style, Musk shot back with a cheeky comment asking for more details on what makes OpenAI truly “open.”
Tell us more about the “Open” part of OpenAI …
— Elon Musk (@elonmusk) March 17, 2024
Elon Musk, along with OpenAI’s founders Sam Altman and Greg Brockman, find themselves in a legal tug-of-war. They’re debating over OpenAI’s shift from a non-profit, open-source research group to a profit-driven branch of Microsoft.
Grok adds to OpenAI’s challenges, especially with new AI models like Anthropic’s Claude 3 Opus and Google’s Gemini entering the scene. Apple has also stepped into the game with its MM1 model.
Yet, Grok-1 isn’t quite ready to handle chat applications out of the box.
First off, it lacks the fine-tuning for dialogue systems. This means it’ll take more work and tools to get Grok-1 ready for chatting, presenting a hurdle for those eager to dive into conversational AI.
Also, with its massive size of 296GB, running Grok-1 demands top-notch computing power that not everyone has.
Still, there’s hope within the AI community that future tweaks could make Grok-1 easier to handle, possibly through quantization. This would trim down both its size and the need for heavy-duty computers, making it more user-friendly.
Grok-1 is open-source
One key point about Grok-1’s launch is xAI’s decision to use the Apache 2.0 license, similar to Mistral’s 7B model.
This choice is important because, unlike other licenses that can be quite strict about how you can use and share the software, the Apache 2.0 license gives people a lot of freedom. You can change the software, use it in your projects, and share it with others pretty freely.
Open-source models like Grok-1 could change how money flows in the AI world. They offer an alternative to the pay-to-use models from companies like OpenAI and Anthropic, making it a bit of a challenge for those businesses.
Grok weights are out under Apache 2.0: https://t.co/9K4IfarqXK
— Sebastian Raschka (@rasbt) March 17, 2024
It's more open source than other open weights models, which usual come with usage restrictions.
It's less open source than Pythia, Bloom, and OLMo, which come with training code and reproducible datasets. https://t.co/kxu2anrNiP pic.twitter.com/UeNew30Lzn
Including commercial projects, Grok-1 offers a solid base for both companies and people wanting to add the model to their own offerings.
By sharing Grok-1’s inner workings with everyone, xAI moves Elon Musk’s idea of open AI forward and challenges the wider AI world.
OpenAI likely feels the heat with new moves from Anthropic, Google, and now xAI.
The AI community is buzzing, expecting something like GPT-5 or Sora to come out, aiming to get ahead once again.
Sources:
https://www.theverge.com/2024/3/17/24097810/xai-open-source-grok-musk-generative-ai-llm
