- A new language model called Dolly is available for free and open source
- Trained by humans and equipped with a database integrating several billion parameters, it could overshadow ChatGPT
- We can also think of it as a starting point. The fact that it is open source can give rise to other fruitful initiatives
We may have just witnessed a new turning point in the race for AI. The American company Databricks indeed launched this Wednesday a new language model called Dolly 2.0. What distinguishes the latter from competitors such as ChatGPT or Bing Chat is that it is accessible for free and the API is open source.
In fact, this language generator seems extremely promising. With its 12 billion built-in parameters, it is based on EleutherAI’s pythia model. And to make sure that the AI offers user-friendly answers, Databricks has put its employees to work by having them participate in an internal competition.
A starting point for unleashing AI
5,000 employees have calibrated this tool by asking it closed and open questions, by making it work on creative writing exercises, or even as an assistant and brainstorming aid.
On its blog, the company is full of praise for its initiative: “To our knowledge, this dataset is the first open-source, human-generated instructional dataset specifically designed for large language models to exhibit the magical interactivity of ChatGPT.”
If the launch of Dolly remains relatively confidential to the general public, AI specialists are very enthusiastic. Like the researcher Simon Willison who told our colleagues d’ArsTechnica : “Even if Dolly 2 isn’t good, I expect we’ll soon see a bunch of new projects using this training data. And some of them might produce something really useful. »
It’s hard to say what will happen next, but we could therefore be dealing with a major change. We can in any case note that the race for AI is reshuffling the cards a bit and that companies like Databricks, which did not play in the same court as GAFAM, are trying to hold their own. That said, the giants of Tech have the financial mass and sufficient resources to counterattack, and the initiatives of Google, Microsoft, and Meta are there to attest to this.
