with the actions of Meta Platforms experiencing a recent 0.5% dropThe company of mark zuckerberg it is forced to develop new concepts and more revolutionary trends if it wants to survive. In this sense, AI has become the philosopher’s stone of the big technology companies to find there a market niche to exploit and that can generate significant economic benefits.
In this sense, Meta is creating a new high-end product group focused on generative artificial intelligence (AI), in which many company teams have been working to create pleasant experiences in this regard. This was announced by Zuckerberg himself in an Instagram post.
The goal that the CEO of Meta is now considering, in the long term, is none other than to develop AI characters that can help people in multiple ways. For all this, they are experiencing the application of this efficient technology at text options (chat on WhatsApp and Messenger), With Images (Instagram creative filters and new banner ad formats) and with videoas well as others multimodal experiences.
Meta does not cease in its efforts to lead the market, and in this sense, Ahmad Al Dahle will be in charge of leading the new products team that will report to Chris Coxproduct manager at Meta.
And it is that after the launch of ChatGPT by Open AIat the end of last year, and now supported by Microsoft, Meta only has to hit the metaverse table if he does not want to be expelled from the field of play. In fact, Microsoft has made use of great technological almighties such as alphabet either Baidu to advertise your own AI offerings.
LlaMA, your last bet
At the end of February, Facebook’s parent company already publicly announced that it was devising a new language model called Large Language Model Meta AI (Calls)capable of competing against OpenAI’s ChatGPT.
The new core software of an AI system that would be available under non-commercial license and exclusively for researchers and entities affiliated with government, academia and civil society. All this is due to the fact that it is sought maintain AI integrity and prevent it from being used improperly and immorally by any user.
Zuckerberg himself assures that his new large-scale language model will have great potential in text generationwill allow you to hold conversations, summarize written material and perform more complex tasks such as predicting protein structures or solving mathematical theorems.
FLAME has been released with text in 20 different languages, although focused on Latin and Cyrillic alphabets, being able to predict the next lines of text. Unlike other AI language models, Meta claims that LLaMA is a small model that requires less power and computing resources to work, although it may work in a broader scope of application. In addition, its performance will be extraordinary.
Similarly, Meta has decided to launch it in various sizes 65B, 33B, 13B and 7B, trained with different amounts of tokens and texts. So that, FLAME -13B outperforms GPT-3 (175B) in most of the tests performed, while FLAME-65B It competes with the best models like the Chinchilla 70-B and PaLM-540B, although the correct answer rate is still quite low.
Nonetheless, LlaMA is not free of errors and the ability to issue untrue information, since toxic and biased comments have been detected in the tests being carried out. But it is still a long way from being able to generate quality texts such as essays or poems through AI and within everyone’s reach, as ChatGPT is achieving.
Previously, last November, its AI research area, led by Yann LeCunlaunched a researcher-oriented AI tool also based on the Large Language Model, LLM: galacticalthough it did not pass the testing phase and failed.
AI’s fierce competition
Given the alliances of ChatGPTbacked by Microsoft, with Bing and Edge, Google has announced that it will soon launch Bard, its own linguistic artificial intelligence. Meta’s strategy with LLaMA is none other than to rival Microsoft and also keep their programming code secret, but once it is developed, they will make it open so that researchers can test new approaches more easily and be able to eliminate problems.
goal intended democratize user access to AI language modelsbecause only in this way will it be possible to overcome the obstacles encountered in the generation of quality information.
