Giovedì Mattina Meta has announced its latest AI model, Llama 3, touted as the most powerful open-source LLM in circulation. And at the time, the giant’s chief scientist, Yann LeCun, said the company had given birth to the Flame Erede, an even more powerful piece of software. The formula that could potentially outperform the world’s synthetic (closed-source) intelligence models comes from OpenAI’s GPT-4 and Google’s Gemini.
It was brought about by several versions of Flame 3, one with 8 million parameters (a set of parameters that corresponds to a giant line of force of a style) and another with 70 million parameters. Ma LeCun added that the Menlo Park colossus has been executing even grander styles; At present, the maximum difficulty of this Llm, which has more than 400 billion parameters, is in the addition phase. “We’ve probably used various diversifications of those styles,” LeCun said at the Imagination in Action course, a convention at the Massachusetts Institute of Technology that is committed to generative synthetic intelligence. It’s unclear when the style of another 400 million parameters can be distributed.
Martedì Meta has also announced a new Llama 3-based assistant, Meta AI (currently not available in Italy), and says you can test its amazing old open-source Llm versions with several similar configurations. Although lately most people who occupy AI, such as Google and OpenAI, use their own technology, LeCun prevents open-source models from advancing faster, potentially accelerating the progression of synthetic intelligence to the human level.
Second, LeCun in the age of open source generation is destined to triumph over this company. The open technique is reserved for core infrastructure software, as it allows for faster sharing and more accurate code improvement and review, while maintaining Meta’s clinical approach. “This increases when no one tests the code anymore,” he continues. The infrastructure has to be open source – I’m progressing faster. “
In fact, Meta’s open-source AI strategy has helped fuel the existing frenzy for generative synthetic intelligence. With the launch of Llama 2 in July 2023, the company has allowed many startups, creators, and marketers to experiment and build with very hard tools. AI models.
by Giorgio Papavero
by Sara Carmignani
by Francesco Del Vecchio
There’s a part of GPT-4 and Gemini that’s only accessible via the API for payment (and in February, Google will upload a lighter style than Gemma), a third-party organization has the option to stick with Meta and distribute prospectives. Styles. Open Source Artificial Intelligence. For just one month, the LLM has the ability to further increase traffic in the era of startup Databricks. But if the results of the comparative tests were reported by Meta fossero corretti, Llama 3 will be even higher.
Semper Martedì LeCun also underestimated Meta Point’s strategy of making available all the synthetic intelligence models highly prized at the extremes of race and computing power. General Mark Zuckerberg, founder and CEO of the colossus, told US network CNBC that, in 2024, he spent billions of dollars to get the GPU needed to load his device’s learning algorithms. Today, the company’s number one also said that Meta’s ultimate purpose is the same effort as OpenAI and Google: to expand the cosiddetta of synthetic, on-device general intelligence into each and every step of every single thing that happens in the capacity of the human intellect. On Instagram, Zuckerberg was able to notice that Meta has incorporated the new Meta AI assistance into products such as Whatsapp, Instagram, Facebook, and Messenger.
In its special blog post published to announce Llama 3, Meta writes this about updating the data used to upload its new Llm. The company said the design is forced through a much larger amount of data than its predecessor, Llama 2. Some AI experts have noted the knowledge spread through meta-evidence, as well as the creation of Llama 3, which is rich in massive amounts of force to force the server.
The expanding features of open-source AI models have subject matter experts that allow LLM to expand knowledge of the computational, chemical, or biological army or add products that can stand up to humans. , Meta has brought tools that prevent the flame from generating potentially harmful results.
by Giorgio Papavero
by Sara Carmignani
by Francesco Del Vecchio
In everyone’s synthetic intelligence box, there are also those who help llama models that are as smart as they are open, at a time when Meta’s open-source license imposes restrictions on which vendors and developers can build.
Stella Biderguy, director of AI who collaborates with EleutherAI, an open-source nonprofit AI organization, showed that Llama 3’s license seems even more restrictive than its predecessor’s. In fact, Meta warned that the guy had more than 700 million active users and that he had a special license, a clause devised to prevent his task from competing with a giant.
In any case, Llos angelesma 3 seems destined to dare to use new exconsistent knowledge about synthetic intelligence. Clément Delos angelesnge, CEO of startup HuggingFace, reported that developers have created more than 30 million variants of Llos angelesma 2. “I’m sure we also have a raffia of new models founded in Llos angelesma 3,” he says. It’s a thing in the Meta component consistent with the angels community. “
This article appears to be from Wired US.
?️ Date announced: June 14-16, the Wired Next Fest in Milan. Text: Admission is free
? Ascolta Grande Giove, Wired’s anniversary podcast in Italy, with stories of innovation, generation and science
?Mettetetevi alla prueba with Wired’s new questionnaire
? The new Wired factor at the Cervelló school
?? ?? Le notizie consistent with capire thing is happening in the Middle East
?? The War in Ukraine: Wired’s Improvements in the Conflict
? Wired has opened the Whatsapp channel: iscriviti subito!
? Watch the Wired video: on YouTube
? Scopri le nostre newsletter: the latest technology, gadget, environment, health and diritti. Iscriviti subito
? News, Reviews & Guide to Buying Your Best Gadgets of the Moment
? Need to do it safely with Wired’s editorial team? United States
by Roberto Pizzato
From the Editorial Board
by Fernanda Gonzalez
by Michele Chicco
SCOPRI THE LAST WARNING
CONDÉ NAST ITALY