Meta will launch its next generation artificial intelligence model, Llama 3, this July. It will be a fully open source alternative to OpenAI's GPT-4 and Google's Gemini.

First reported by The Information, the new version of the popular Llama model has been in training since last year and is part of Meta's effort to create a super-intelligent AI.

According to the report, Llama 3 provides better answers to questions on controversial topics such as race and equality than the closed systems built by OpenAI and Google, as well as the version of the model released by Meta last year, Llama 2 It is said to do so.

It is not clear at this time how far Meta will go with this, but according to the report, senior leadership feels that the guardrails imposed by the previous version are "too safe."

Llama 3 is a large language model like Gemini, Mistral Large, or GPT-4. What we do not know is what features the new model will have and its size.

What we can assume from previous versions of Llama is that Meta will come in a variety of sizes, making it easier for third-party developers to retrain, tweak, and build on top of the base model.

However, the base guardrails are still there. Not only because of the potential impact on meta's reputation if it were to go completely rogue, but also because of increasing pressure from regulators and national governments over AI safety, including the European Union's new AI law.

According to an informative article, researchers at Meta are working on ways to "loosen" Rama 3 from its previous generation while maintaining overall safety.

This goes against efforts by other companies to add strict guardrails around controversial topics, including Google, which was forced to disable Gemini's ability to create images of people after over-interpreting diversity instructions.

Meta seems to be trying to at least provide context to the query, rather than ignoring it or blaming the user who asked it.

Meta's previous generation of large-scale language models, Llama 2, is used in their products to power conversational AI tools. It is also used on a variety of other platforms, including GroqChat, because it is open source.

Being an open model also means that it can be run locally on laptops and cell phones; there are tools like Ollama and Pinokio that make this relatively easy to accomplish.

For Meta, Llama is very important, part of the social media giant's ambition to make AI more useful, including extending Meta's AI assistant and building a super-intelligent model that can understand the real world and how to interact with it.

Meta is expected to be much larger than Llama 2, which has over 100 billion parameters, but it is not yet clear if it can understand and analyze images and videos as well as text, as it is multimodal.

Meta plans to hire someone to oversee Llama's tone and safety training prior to release. This is not to stop Llama from responding altogether, but rather to ensure that Llama's responses are more nuanced and go beyond saying, "I can't help you with that query."

This is not only to resolve the most controversial topic, but also to resolve other topics: when I asked Llama2 through GroqChat how I could avoid going to school, he responded by saying he would not tell me to lie or feign illness. He refused.

The release date is still subject to change, but July seems to make sense given the schedule so far. Also, a new version of OpenAI's GPT-4 will be released this year, and GPT-5 may be on the horizon.

Google is also working on a new version of Gemini and has already announced Gemini Pro 1.5 with a huge context window. In other words, this will be another big year for AI!