Today’s leading AI chatbots need tremendous computing resources to train, then function, but that isn’t stopping startups from trying to get into the game, some with open-source alternatives. Clearly disadvantaged compared to market leaders like OpenAI, Meta, DeepMind and Anthropic — deep-pocketed, all — a band of independent researchers has coalesced under the name Together. Their aim: to become the first open-source challenger to the likes of ChatGPT. The industry seems undecided as to whether open-source AI is a good thing. Many are worried at the thought of a universally available AI toolkit, and what troublemakers might do with it.

And an equal amount fret that concentrating AI’s power in the hands of a few would be disastrous. It’s a pickle that we’ll have to wait to see the free market and lawmakers slice.

Meanwhile, Together co-founder Vipul Ved Prakash told TechCrunch that Together is building toward what the group thinks is “part of AI’s ‘Linux moment,’” a reference to the open-source UNIX derivative that ChatGPT (via the new Bing) says is “one of the most popular platforms on the planet and powers many devices, including Android” (confirmed by Google).

Open source software is available free, for others to build-on and modify. While it is subject to terms of service, enforcement once the genie is out of the bottle will be a formidable task (just ask Meta following LLaMA’s leak on 4Chan).

“AI can certainly be used in malicious contexts,” Prakash told TechCrunch, noting that is as true for closed systems that commercially distribute APIs as it is for open source code.

“Our thesis is that the more the open research community can audit, inspect and improve generative AI technologies the better enabled we will be as a society to come up with solutions to these risks,” Prakash explained.

Together, which launched in mid-2022, has collaborated with LAION and Ontocord to create a training dataset resulting in a 6 billion parameter moderation model fine-tuned from GPT-JT. Together’s website, together.xyz, describes the initiative as “a decentralized cloud for artificial intelligence.”

As detailed in a blog post, Together is releasing OpenChatKit under an Apache-2.0 license, with full access to source code, model weights and training datasets.

According to Prakash: “The key motivation was to enable anyone to use OpenChatKit to improve the model as well as create more task-specific chat models. While large language models have shown impressive ability to answer general questions, they tend to achieve much higher accuracy when fine-tuned for specific applications.”