Meta Tuesday announcement That he will host his very first DEV conference dedicated to a generative AI.
Called Llamacon After Meta’s Llama Family of Generative IA Models, the conference should take place on April 29. Meta said she was planning to share “the last on [its] Open Source developments to help developers […] Create incredible applications and products. »»
Additional details will be available soon, said Meta. The annual conference of the company’s developers, Meta Connect, will be held later in the year, in September – its typical window.
Several years ago, Meta adopted an “open” approach to develop AI technologies in order to promote an ecosystem of applications and platforms. It has not been revealed how many applications or services have been built in addition, but it is previously noted that Goldman Sachs, Nomura Holdings, AT&T, Doordash and Accenture Use Llama. The company claims to have hundreds of millions of model downloads, and at least 25 partners hosting Llama, notably Nvidia, Databricks, Groq, Dell and Snowflake, some of which have built additional tools which, for example, refer to the data of the data of owners and allow them to operate at lower latencies.
But Meta would have been captured flat by ramp -up of the Chinese AI company Deepseek, which managed to release “Open” AI to compete with the Meta.
Would haveMeta believes that one of the new Deepseek models could surpass the next version of Llama, which should be released in the coming weeks. Meta would have been blurred to set up war rooms to decipher the way Deepseek lowered the cost of the race and the deployment of models, so that he can apply these learning to the development of Llama.
Meta recently said that he would spend up to $ 80 billion in AI -related projects this year, including IA hiring and the construction of new AI data centers.
Meta-chef Mark Zuckerberg previously announced That the company plans to launch several Llama models in the coming months, including “reasoning” models in the direction of the O3-Mini of Openai and models with multimodal capacities natively. He also alluded to “agent” capacities, suggesting that future Llama models will be able to take certain actions independently.
“I think it could very well be the year when Llama and Open Source become the most advanced and used AI models,” Zuckerberg said when Meta’s quarter 2024 results in January. “”[O]your goal for [Llama this year] is to direct. “”
Meta is also in the middle of a trial which accuses the company of training its models on the book equipment protected by copyright without authorization. In another challenge to the ambitions of Meta Llama, several EU countries have forced the company to postpone – and in some cases completely cancel – its model launching plans on data confidentiality problems.
Techcrunch has a newsletter focused on AI! Register here to get it in your reception box every Wednesday.
