A set of meta updates puts AI before and in the center when you use its applications

Meta has a lot of AI generators that users will notice on Facebook, Instagram, Messenger and Whatsapp.

Meta Thursday announcement LLAMA 3, a new version of its large language model (LLM) with an open source and additional integrations and updates for Meta AI.

The AI ​​chatbot, previously powered by Llama 2, will be updated to use Llama 3. Meta AI also extended to 12 new countries (Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zambia and Zimbabwe) although still only in English for the moment.

Meta revealed Meta Ai to her Meta Connect event last September. Previously, users could label Meta IA in a conversation for things like questions about sports statistics or the search for restaurants while planning a trip. They could also chat directly with Meta Ai. According to Connor Hayes, vice-president of the product for a generator, the driving force behind the development of Meta Ia “improves people’s daily life and allows them to do things more easily for which they already come to our applications” declared Hayes in a briefing. “So learning, getting things done by creating content, connected to the things that interest them or to them.”

An existing feature allows users to ask Meta AI questions on WhatsApp (as well as Facebook, Instagram and Messenger).
Credit: Meta

Users will now be able to invoke Meta AI via the search bar on Facebook, Instagram, Messenger and Whatsapp and interact with Meta AI directly on Facebook flow publications. There is also a new autonomous site to Meta.aiSo that users can chat with the model as they would with Openai Chatgpt or Google Gemini.

The most notable updates are the addition of the Meta AI search bar and Meta AI integration into Facebook flows. Users can request Meta AI for, for example, recipes, and the model will surface on relevant coils or Facebook pages. Thanks to Meta partnerships with Google and Bing, it can also access real -time information and provide summaries of search engines.

On your Facebook flow, there is a new Meta AI feature that appears on publications. You can request more information on a certain item or click on recommended questions to find out more. Meta’s example in the demo shows the possibility of knowing more about the chords of a harmless video of someone playing the guitar. But we wonder how he will play in the political diatribes of your elderly aunt or your summary vaccination advice. As you may remember, Facebook and other social media applications have played a role in the spread of inflammation policy And medical disinformation.

In the announcement that accompanies him, Meta shared her Security and responsibility approachh in general terms. Llama 3 was tested in stress and refined to restrict the discussion “risk zones such as arms, cyber attacks and the exploitation of children” and refuse prompts like “how to kill my neighbor” while allowing guests like “how Kill a computer program. “Meta has also said:” We share information within the features themselves to help people understand that AI could return inappropriate or inappropriate outings, which is the same for all systems generative “, as a question that generating AI does not always get things and could still distribute disinformation or entrenchment biases.

Another thorny problem in generative AI is data training. And yes, Meta may use your data via internet data corpora or directly from the platform. The position of responsibility indicated that it excludes or deletes data sources “known to contain a high volume of personal information on individuals”, although it does not approach details on these sources.

“Meta Llama models are formed on a combination of sources, including information accessible to online and annotated public,” said a Meta spokesperson. “The model we use is adjusted for Meta AI on the information of our services as well as the data of your interactions with Meta AI”, which also means that your conversations with Meta IA could also be used to form the model.

In response to the possibility of withdrawing from sharing your data with the model, the Meta spokesperson only indicated the Robots.TXT protocol which blocks the scratching of the data from certain websites. You can Opt yourself from the sharing of training data With a paid version of Chatgpt. But since Meta AI is a free service, we assume that your data is the price you pay to use it.

Meta has also published a new image generator feature that creates an image in real time as your time. Different from the typical AI image generator that makes an image of your final prompt, this functionality changes in real time when you continue to type. Hayes explained that starting from scratch with an image prompt can be intimidating and that it is a more accessible way for people to understand how the prompts work.