Meta needs to win over AI developers at its first LlamaCon

Tuesday, Meta is hosting Its first Llamacon The AI ​​developer conference is at his Menlo Park headquarters, where the company will try to pitch applications about creating applications with its Open Lama AI models. Just a year ago, it was no solid sales.

However, in recent months, both “Open” AI Labs, like the Meta DEPSEC, and the rapidly developed AI Reese, have fought to continue “Open” AI labs, like closed commercial competitors. Llamacon comes at an important moment for Meta in search of a broad Lama ecosystem.

Winning developers can be as easy as shipping to better open model. However, it can be more difficult to achieve than listen.

Start a committed initial

Meet Lama 4 turned on the beginning of this month Underhalemed developers with several benchmark scores in models like DIPSC R1 and V3. It was far from what the llama was once: a boundary-pushing model lineup.

When Meta launched its Lama 3.1 405B model last summer, CEO Mark Zuckerberg It marked as a big winThe A Blog postMeta Lamama 4.1 405 BK has called “the most able Foundation Model”, the best model of the opening with the performance competes with GPT -4O.

It was a fascinating model to confirm – and there were other models of the Meta Lama 3 family. Jeremy Nixon, who has hosted the hackathon at the San Francisco AGI House for the past several years, has launched “Historic Tihasik moments” known as Lama 3.

Lama 3 rationally made Meta the AI ​​developers, the Models wherever they chose supplies with the freedom to host-Ez performance with the freedom to host. Today, the Mater Lama 4.1 model has been downloaded more than Lama 4, said the Hug Product and Growth Head, Jeff Boudia, an interview.

The opposite and the difference with the Meta Lama 4 family reception and the difference is complete. However, Lama 4 has been controversial from the beginning.

Benchmarking

Meta is one of its Lama 4 models, a version of the Lama 4 Mavarick favors for a “conversation” that it helped to gain the top spot in the crowded benchmark LM Arena. Meta never published this model, however – Mavarick’s version that has been rolled out extensively Ending In LM Arena

LM Arena’s behind the team said Meta Should have been “clean” About significance. Ion Stoika, an LM Arena co-founder and Professor of UC Berkeley, who co-founded companies, including anyscale and databrix, told TechCrunch that this incident had damaged the confidence of the developer community.

“[Meta] Should have been more clear that the Mavarick model was turned on [LM Arena] The model that was published was different, “Stoika told TechCrunch in an interview. Of course, they could recover it by publishing better models.”

No reason

Lama was an AI rational model excluded from the family 4 family. Reasonable models can work carefully through questions before answering. Last year, many AI has published the art argument modelWhich performs better on certain criteria.

Meet Teasing a Lama 4 argument modelHowever, the company did not indicate when it would expect it.

AI2 researcher Nathan Lambert says that no argument model has been published with Meta Lama 4, it implies that the company can run to the launch.

“Everyone is releasing an argument model and it looks so good to their models,” said Lambert. “Why couldn’t [Meta] Wait to do it? I don’t have the answer to this question. It looks like the strangeness of the general organization. “

Lambert mentions that rival open models are closer to the border than ever before and they now come into more size and size – increase the pressure in the meta. For example, Monday, Alibaba has published a collection of models, Queen 3It has been alleged that a programming benchmark, OpenAI and Google’s best coding models have surpassed the codesmark, CodeFource.

To regain the leadership of the Open model, Meta only needs to provide a higher model, according to AI researcher Ravid Schwarz Jive at the NYO Center for Data Science. He told TechCrunch that there may be more risk to hire new strategies.

It is unclear whether the meta is in a position to take a big risk now. Current and former employees said earlier Fortune Meta’s AI Research Lab “slow death dies.” VP of AI Research Company, Joel Pino, This month has announced that he is leavingThe

Llamacon Open, Google, XAI and others like AI labs, are cooking to show what is cooking to lose. If it fails to supply it, the company may fall further behind in the super-competitive place.