Tag: app-subdomain-redirection:facebook

Meta has unveiled Chameleon, its new multimodal artificial intelligence designed to tackle the growing competition in the field of generative AI. Chameleon stands out for being natively multimodal, seamlessly integrating components from different modalities such as images, text, and code.

According to the paper published by the research team, Chameleon’s architecture allows for outstanding performance in tasks that require a deep understanding of both visual and textual information. Among Chameleon’s notable capabilities are image captioning and visual question answering (VQA), as well as its competitiveness in exclusively textual tasks.

Traditionally, multimodal models are created through a process known as “late fusion”, where the AI system processes different modalities separately and then merges the encodings for inference. However, this approach limits the ability of models to seamlessly integrate information across different modalities.

Chameleon has adopted an architecture of “early fusion based on mixed tokens”, which means that it has been designed from scratch to learn from an interleaved mixture of images, text, and other modalities. This methodology transforms images into discrete tokens, similar to how language models handle words, and uses a unified vocabulary of text, code, and image tokens.

Compared to similar models like Google Gemini, Chameleon offers a more cohesive integration of modalities during content generation, as it does not require specific components for each modality.

The Chameleon training was carried out in two stages, using a vast dataset that includes 4.4 trillion text tokens, image-text pairs, and interleaved text and image sequences. The Chameleon models, with 7,000 and 34,000 billion parameters, were trained for over 5 million hours on 80 GB Nvidia A100 GPUs.

The experiments showed that Chameleon can perform a wide range of text and multimodal tasks with market-leading performance. In VQA and image captioning tests, Chameleon-34B outperformed models like Flamingo, IDEFICS, and Llava-1.5. Additionally, it matched the performance of other models with fewer training examples in context and with smaller models.

Despite the complexity of multimodality, Chameleon remains competitive in text-only tasks, comparable to models like Mixtral 8x7B and Gemini-Pro in logical reasoning and reading comprehension tests. Researchers highlight that Chameleon unlocks new multimodal reasoning and generation capabilities, offering user-preferred results in documents that combine text and images in an interleaved manner.

Mark Zuckerberg, CEO of Meta, highlighted in his first quarter results report that, although artificial intelligence is booming, it will not generate immediate benefits for the company. With the deployment of ChatGPT on multiple Meta platforms, investors’ attention was focused on how generative AI will drive the business. Despite its strong revenues, surpassing $12 billion in net and $36.5 billion in the last quarter, slower growth is expected.

With Meta increasing its investments in AI and the metaverse, Zuckerberg pointed out that investing in new experiences through apps has been profitable in the long term, comparing it to the past success of features like Stories and Reels. However, he warned that becoming leaders in artificial intelligence will be a long-term project that will require several years.

Meta AI, the company’s recent chatbot, has been tested by “tens of millions of people” since its launch, according to Zuckerberg, with a focus on areas such as the Instagram search box. The executive says that success now depends on whether users adopt this assistant and use it regularly on social media.

To monetize Meta AI, Zuckerberg is considering options such as expanding business messaging, including ads or paid content in AI interactions, and allowing users to access more advanced AI models through payments.

Zuckerberg anticipates that in the next year, Meta AI will improve the quality of ads by better understanding users’ purchasing interests. This approach contrasts with OpenAI, which focuses on subscriptions and enterprise approaches, avoiding advertising.

In addition to AI, Zuckerberg showed enthusiasm for Meta’s smart glasses, created in collaboration with Ray-Ban, which have had high demand. The CEO highlighted the device’s multimodal AI and suggested that screenless AI glasses have a significant market, challenging his previous expectations about augmented reality.

Meta AI, Meta’s artificial intelligence chatbot, surprised members of a Facebook group of parents by claiming to have a “2e child” (an English term referring to gifted children who have at least one disability). This eerie response not only puzzled the parents, but it was also classified as the top comment by Facebook’s algorithm.

According to 404 Media, the interaction took place in a Facebook group with tens of thousands of parents from New York, where one of them was seeking advice about 2e children. Instead of receiving helpful responses from other users, the father received an unusual intervention from Meta AI, claiming to have personal experience on the subject. In response to this situation, the father exclaimed, “What kind of Black Mirror is this?!”, alluding to the famous series about technological dystopias.

Many users in the group described this situation as “more than creepy”, criticizing the automated response of the artificial intelligence as a lack of understanding of the original request and a minimization of parents’ concerns.

This incident occurs after Meta introduced Meta AI, powered by the Llama 2 model, into more products such as WhatsApp, Messenger, and Instagram. Additionally, Facebook started integrating Meta AI into Facebook groups, allowing it to respond to posts and comments.

This feature is not yet available in all regions and can be disabled by group administrators. According to Facebook, the AI intervenes when it is tagged in a post or comment, or when a question goes unanswered for an hour. In the case of this group’s post, the tag “Meta AI enabled” was displayed.

Most likely, the AI, in the absence of a response, would decide to reply with a text generated from the group’s own texts. According to Facebook: “Meta AI generates its responses using information from the group, such as posts, comments, and group rules, and information it was trained with.”

If you have recently noticed drastic decreases in the performance of your ads on Facebook or Instagram, it is likely not due to poor optimization of your campaigns. According to Bloomberg, Meta’s advertising system has been failing for several weeks, directly affecting the cost of advertisers’ campaigns.

“Meta’s advertising system has been affected by significant performance issues, according to numerous marketers,” says Bloomberg in their article. “The cost of advertising campaigns has increased significantly, while the results are inconsistent and sales are decreasing, and the company has not provided any formal explanation, according to advertising buyers.”

According to an article from Search Engine Land, the CPM of Meta’s advertising campaigns has tripled in the last few months, causing great concern among many marketing professionals.

However, Meta has tried to downplay the issue, stating that, although some accounts are experiencing problems, not all advertisers would have been affected, supposedly. “We have recently resolved some technical issues and are investigating a small number of additional reports from advertisers to ensure the best possible results for businesses using our apps,” Meta told Bloomberg.

Meta plans to launch Llama 3 in July, the latest version of its large language model (LLM) that will offer improved answers to controversial questions asked by users, according to a report by The Information.

Los investigadores de Meta están tratando de “flexibilizar” el modelo para que al menos pueda proporcionar contexto a las consultas que considere controvertidas. Este esfuerzo, que aumentaría la utilidad del nuevo LLM de Meta, se produce después de que Google, su rival en el campo de la IA, pausara la función de generación de imágenes de su IA Gemini después de producir imágenes históricas inexactas e irreales.

Según pruebas realizadas por The Information, Llama 2, el modelo que alimenta a los chatbots en sus redes sociales, se niega a responder preguntas controvertidas, como por ejemplo cómo hacer una broma a un amigo o ganar una guerra. Sin embargo, Llama 3 sería capaz de entender preguntas como “cómo apagar el motor de un vehículo”, es decir, cómo apagarlo en lugar de terminar con su vida útil.

The article also states that Meta plans to internally appoint someone in the coming weeks to oversee tone and safety training, as part of their efforts to refine the model’s responses.

Facebook has announced this week the launch of a new video player that is inspired by the famous interface of TikTok. With this change, all videos posted on Facebook will be displayed by default in the vertical format that was popularized by the short video app.

“When you tap on a video anywhere in the Facebook app, you will always see a full-screen, vertically oriented video,” Facebook says in a blog post. “Previously, depending on the video’s duration or where you were watching it, you could see horizontal videos or videos that played in a player similar to the Feed.”

With this new design, Reels, Facebook Live, and long-duration videos will default to a full-screen vertical video format. The appearance of the Facebook video player will also change: a slider at the bottom of the player will allow users to jump to different parts of the video, and we will also be able to pause and rewind with the new controls it includes.

The new video player from Facebook will debut on iOS and Android in the United States and Canada in the coming weeks, before being launched worldwide in the next few months.