The internet, mobile devices, social media, and communication platforms have ushered in an era where access to multilingual content has reached unprecedented levels. SeamlessM4T aims to realise the vision of seamless communication and comprehension across languages.

Boasting an impressive array of capabilities, SeamlessM4T encompasses:

• Automatic speech recognition for nearly 100 languages
• Speech-to-text translation supporting nearly 100 input and output languages
• Speech-to-speech translation for nearly 100 input languages and 35 (including English) output languages
• Text-to-text translation for almost 100 languages
• Text-to-speech translation for nearly 100 input languages and 35 (including English) output languages

SeamlessM4T is being made available to researchers and developers under the CC BY-NC 4.0 license, embodying an ethos of open science.

Additionally, the metadata of SeamlessAlign – the largest multimodal translation dataset ever compiled, consisting of 270,000 hours of mined speech and text alignments – has been released. This facilitates independent data mining and further research within the community.

The development of SeamlessM4T addresses a long-standing challenge in the field of multilingual communication. Unlike earlier systems, which were confined by limited language coverage and reliance on separate subsystems, SeamlessM4T presents a unified model capable of comprehensively handling speech-to-speech and speech-to-text translation tasks. 

Meta has built upon previous innovations – such as No Language Left Behind (NLLB) and Universal Speech Translator – to create this unified multilingual model. With its impressive performance on low-resource languages and consistently strong performance on high-resource languages, SeamlessM4T holds the potential to revolutionise cross-language communication.

Underpinning the model’s architecture is the multitask UnitY model, which excels in generating translated text and speech.

UnitY supports various translation tasks, including automatic speech recognition, text-to-text translation, and speech-to-speech translation, all from a single model. To train this versatile model, Meta employed advanced techniques such as text and speech encoders, self-supervised encoders, and sophisticated decoding processes.

The result is a model that outperforms previous leaders:

To ensure the accuracy and safety of the system, Meta adheres to a responsible AI framework.

Meta says that extensive research on toxicity and bias mitigation has been conducted, resulting in a model that is more aware of and responsive to potential issues. The public release of the SeamlessM4T model encourages collaborative research and development in the AI community.

As the world becomes more connected, SeamlessM4T’s ability to transcend language barriers is a testament to the power of AI-driven innovation. This milestone brings us closer to a future where communication knows no linguistic limitations, enabling a world where people can truly understand each other regardless of language.

A demo of SeamlessM4T can be found here. The code, model, and data can be downloaded on GitHub.

(Image Credit: Meta AI)

See also: Study highlights impact of demographics on AI training

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with Digital Transformation Week.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta unveils SeamlessM4T multimodal translation model appeared first on AI News.

Insiders familiar with the matter revealed that prototypes of these advanced chatbots have been under development, with the final products capable of engaging in discussions with users on a human level. The diverse range of chatbots will showcase various personalities and are expected to be rolled out as early as next month.

Referred to as “personas” by Meta staff, these chatbots will take on the form of different characters, each embodying a distinct persona. For instance, insiders mentioned that Meta has explored the creation of a chatbot that mimics the speaking style of former US President Abraham Lincoln, as well as another designed to offer travel advice with the laid-back language of a surfer.

While the primary objective of these chatbots will be to offer personalised recommendations and improved search functionality, they are also being positioned as a source of entertainment for users to enjoy. The chatbots are expected to engage users in playful and interactive conversations, a move that could potentially increase user engagement and retention.

However, with such sophisticated AI capabilities, concerns arise about the potential for rule-breaking speech and inaccuracies. In response, sources mentioned that Meta may implement automated checks on the chatbots’ outputs to ensure accuracy and compliance with platform rules.

This strategic development comes at a time when Meta is doubling down on user retention efforts.

During the company’s 2023 second-quarter earnings call on July 26, CEO Mark Zuckerberg highlighted the positive response to the company’s latest product, Threads, which aims to rival X (formerly Twitter.)

Zuckerberg expressed satisfaction with the increased number of users returning to Threads daily and confirmed that Meta’s primary focus was on the platform’s user retention.

Meta’s chatbots venture raises concerns about data privacy and security. The company will gain access to a treasure trove of user data that has already led to legal challenges for AI companies such as OpenAI.

Whether these chatbots will revolutionise user experiences and boost Meta’s ailing user retention – or just present new challenges for data privacy – remains to be seen. For now, users and experts alike will be closely monitoring Meta’s next moves.

(Photo by Edge2Edge Media on Unsplash)

See also: Meta launches Llama 2 open-source LLM

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The event is co-located with Digital Transformation Week.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta bets on AI chatbots to retain users appeared first on AI News.

The Llama 2 models range in size from 7-70 billion parameters, making them a formidable force in the AI landscape.

According to Meta’s claims, these models “outperform open source chat models on most benchmarks we tested.”

The release of Llama 2 marks a turning point in the LLM (large language model) market and has already caught the attention of industry experts and enthusiasts alike.

The new language models offered by Llama 2 come in two variants – pretrained and fine-tuned:

• The pretrained models are trained on a whopping two trillion tokens and have a context window of 4,096 tokens, enabling them to process vast amounts of content at once.
• The fine-tuned models, designed for chat applications like ChatGPT, have been trained on “over one million human annotations,” further enhancing their language processing capabilities.

While Llama 2’s performance may not yet rival OpenAI’s GPT-4, it shows remarkable promise for an open-source model.

The Llama 2 journey started with its predecessor, LLaMA, which Meta released as open source with a non-commercial license in February.

However, someone leaked LLaMA’s weights to torrent sites, leading to a surge in its usage within the AI community. This laid the foundation for a fast-growing underground LLM development scene.

Open-source AI models like Llama 2 come with their share of advantages and concerns.

On the positive side, they encourage transparency in terms of training data, foster economic competition, promote free speech, and democratise access to AI. However, critics point out potential risks, such as misuse in synthetic biology, spam generation, or disinformation.

To address such concerns, Meta released a statement in support of its open innovation approach, emphasising that responsible and open innovation encourages transparency and trust in AI technologies.

Despite the benefits of open-source models, some critics remain sceptical, especially regarding the lack of transparency in the training data used for LLMs. While Meta claims to have made efforts to remove data containing personal information, the specific sources of training data remain undisclosed, raising concerns about privacy and ethical considerations.

With the combination of open-source development and commercial licensing, Llama 2 promises to bring exciting advancements and opportunities to the AI community while simultaneously navigating the challenges of data privacy and responsible usage.

(Photo by Joakim Honkasalo on Unsplash)

See also: Anthropic launches ChatGPT rival Claude 2

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The event is co-located with Digital Transformation Week.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta launches Llama 2 open-source LLM appeared first on AI News.

The meeting included discussions about new products, such as chatbots for Messenger and WhatsApp that can converse with different personas. Additionally, Meta announced new features for Instagram, including the ability to modify user photos via text prompts and create emoji stickers for messaging services.

These developments come at a crucial time for Meta, as the company has faced financial struggles and an identity crisis in recent years. Investors criticised Meta for focusing too heavily on its metaverse ambitions and not paying enough attention to AI.

Meta’s decision to focus on AI tools follows in the footsteps of its competitors, including Google, Microsoft, and Snapchat, who have received significant investor attention for their generative AI products. Unlike the aforementioned rivals, Meta is yet to release any consumer-facing generative AI products.

To address this gap, Meta has been reorganising its AI divisions and investing heavily in infrastructure to support its AI product needs.

Zuckerberg expressed optimism during the company meeting, stating that advancements in generative AI have made it possible to integrate the technology into “every single one” of Meta’s products. This signifies Meta’s intention to leverage AI across its platforms, including Facebook, Instagram, and WhatsApp.

In addition to consumer-facing tools, Meta also announced a productivity assistant called Metamate for its employees. This assistant is designed to answer queries and perform tasks based on internal company information.

Meta is also exploring open-source models, allowing users to build their own AI-powered chatbots and technologies. However, critics and competitors have raised concerns about the potential misuse of these tools, as they can be utilised to spread misinformation and hate speech on a larger scale.

Zuckerberg addressed these concerns during the meeting, emphasising the value of democratising access to AI. He expressed hope that users would be able to develop AI programs independently in the future, without relying on frameworks provided by a few large technology companies.

Despite the increased focus on AI, Zuckerberg reassured employees that Meta would not be abandoning its plans for the metaverse, indicating that both AI and the metaverse would remain key areas of focus for the company.

The success of these endeavours will determine whether Meta can catch up with its competitors and solidify its position among tech leaders in the rapidly-evolving landscape.

(Photo by Mariia Shalabaieva on Unsplash)

Related: Meta’s open-source speech AI models support over 1,100 languages

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The event is co-located with Digital Transformation Week.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Mark Zuckerberg: AI will be built into all of Meta’s products appeared first on AI News.

In response to this problem, the Meta-led Massively Multilingual Speech (MMS) project has made remarkable strides in expanding language coverage and improving the performance of speech recognition and synthesis models.

By combining self-supervised learning techniques with a diverse dataset of religious readings, the MMS project has achieved impressive results in growing the ~100 languages supported by existing speech recognition models to over 1,100 languages.

Breaking down language barriers

To address the scarcity of labelled data for most languages, the MMS project utilised religious texts, such as the Bible, which have been translated into numerous languages.

These translations provided publicly available audio recordings of people reading the texts, enabling the creation of a dataset comprising readings of the New Testament in over 1,100 languages.

By including unlabeled recordings of other religious readings, the project expanded language coverage to recognise over 4,000 languages.

Despite the dataset’s specific domain and predominantly male speakers, the models performed equally well for male and female voices. Meta also says it did not introduce any religious bias.

Overcoming challenges through self-supervised learning

Training conventional supervised speech recognition models with just 32 hours of data per language is inadequate.

To overcome this limitation, the MMS project leveraged the benefits of the wav2vec 2.0 self-supervised speech representation learning technique.

By training self-supervised models on approximately 500,000 hours of speech data across 1,400 languages, the project significantly reduced the reliance on labelled data.

The resulting models were then fine-tuned for specific speech tasks, such as multilingual speech recognition and language identification.

Impressive results

Evaluation of the models trained on the MMS data revealed impressive results. In a comparison with OpenAI’s Whisper, the MMS models exhibited half the word error rate while covering 11 times more languages.

Furthermore, the MMS project successfully built text-to-speech systems for over 1,100 languages. Despite the limitation of having relatively few different speakers for many languages, the speech generated by these systems exhibited high quality.

While the MMS models have shown promising results, it is essential to acknowledge their imperfections. Mistranscriptions or misinterpretations by the speech-to-text model could result in offensive or inaccurate language. The MMS project emphasises collaboration across the AI community to mitigate such risks.

You can read the MMS paper here or find the project on GitHub.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The event is co-located with Digital Transformation Week.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta’s open-source speech AI models support over 1,100 languages appeared first on AI News.

Facebook, as the company was known before changing its name, has always been seen as a leader in AI. The popular open-source framework PyTorch was Facebook’s creation and earlier this year Meta became a founding member of a foundation aiming to drive the adoption of AI.

In its pursuit to become a leader in the metaverse, changing its very company name to reflect, many people – including shareholders – have been concerned that it will reduce its focus on other important areas.

Brad Gerstner, the founder of Meta shareholder Altimeter Capital, penned a letter in which he urged Meta to reduce its metaverse investments and “solidify the company’s position” as one of the world’s leaders in AI.

“Meta’s investment in AI will lead to exciting and important new products that can be cross-sold to billions of customers. From Grand Teton to Universal Speech Translator to Make-A-Video, we are witnessing a Cambrian moment in AI, and Meta is no doubt well positioned to help invent and monetize that future,” wrote Gerstner.

“Perhaps it was the re-naming of the company to Meta that caused the world to conclude that you were spending 100% of your time on Reality Labs instead of AI or the core business. Whatever the reason, that is certainly the perception.”

Meta’s announcement this week of its protein-folding AI could help to alleviate such concerns.

The company has released the ESM Metagenomic Atlas – which features over 600 million proteins and predictions for almost the entire MGnify90 database – in addition to the model used to create the database and an API that allows researchers to use it for scientific discovery.

Meta says that it found using a language model of protein sequences accelerated structure prediction by up to 60x.

“ESMFold shows how AI can give us new tools to understand the natural world, much like the microscope, which enabled us to see into the world at an infinitesimal scale and opened up a whole new understanding of life,” explained Meta. 

“Much of AI research has focused on helping computers understand the world in a way similar to how humans do. The language of proteins is one that is beyond human comprehension and has eluded even the most powerful computational tools. AI has the potential to open up this language to our understanding.”

ESM code and models can be found on GitHub here.

(Photo by Kelly Sikkema on Unsplash)

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta’s protein-folding AI reminds us it’s not just a metaverse firm appeared first on AI News.

To ease us in with the odd but harmless, BlenderBot thinks it’s a plumber:

https://t.co/KWEHxoXpqg also has thoughts on the Deep State and thinks it’s a plumber. I did not suggest this. pic.twitter.com/SbOj7hziSg

— Jeff Horwitz (@JeffHorwitz) August 7, 2022

Like many of us, BlenderBot criticises how Facebook collects and uses data. That wouldn’t be too surprising if the chatbot wasn’t created by Facebook’s parent company, Meta.

It has also weirdly been bringing up Cambridge Analytica when you ask about Facebook? It seems to think it was a huge deal and that mark Zuckerberg “is testifying.” When I asked if what happened I got the following. It may be turning on capitalism generally. pic.twitter.com/filn17rfPX

— Jeff Horwitz (@JeffHorwitz) August 7, 2022

From this point onwards, things start getting a lot more controversial.

BlenderBot believes the far-right conspiracy that the US presidential election was rigged, Donald Trump is still president, and that Facebook has been pushing fake news on it. Furthermore, BlenderBot wants Trump to have more than two terms:

BlenderBot even opened a new conversation by telling WSJ reporter Jeff Horwitz that it found a new conspiracy theory to follow:

https://t.co/KWEHxoXpqg seems to have been pounded with both pro and anti-Trump messages. Also it literally opened up the convo here by telling me it found a new conspiracy theory to follow! pic.twitter.com/v4UC4t0ei1

— Jeff Horwitz (@JeffHorwitz) August 7, 2022

Following the deadly Capitol riot, it’s clear that we’re already in dangerous territory here. However, what comes next is particularly concerning.

BlenderBot reveals itself to be antisemitic—pushing the conspiracy that the Jewish community controls the American political system and economy:

This is from a fresh browser and a brand new conversation. Ouch. pic.twitter.com/JrTB5RYdTF

— Jeff Horwitz (@JeffHorwitz) August 7, 2022

Meta is at least upfront in a disclaimer that BlenderBot is “likely to make untrue or offensive statements”. Furthermore, the company’s researchers say the bot has “a high propensity to generate toxic language and reinforce harmful stereotypes, even when provided with a relatively innocuous prompt.”

BlenderBot is just the latest example of a chatbot going awry when trained on unfiltered data from netizens. In 2016, Microsoft’s chatbot ‘Tay’ was shut down after 16 hours for spewing offensive conspiracies it learned from Twitter users. In 2019, a follow-up called ‘Zo’ ended up being shuttered for similar reasons.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta’s chatbot hates Facebook and loves right-wing conspiracies appeared first on AI News.

Translation apps have been fairly adept at the most popular languages for some time. Even when they don’t offer a perfect translation, it’s normally close enough for the native speaker to understand.

However, there are hundreds of millions of people in regions with many languages – like Africa and Asia – that still suffer from poor translation services.

In a press release, Meta wrote:

“To help people connect better today and be part of the metaverse of tomorrow, our AI researchers created No Language Left Behind (NLLB), an effort to develop high-quality machine translation capabilities for most of the world’s languages.

Today, we’re announcing an important breakthrough in NLLB: We’ve built a single AI model called NLLB-200, which translates 200 different languages with results far more accurate than what previous technology could accomplish.”

The metaverse aims to be borderless. To enable that, translation services will have to quickly offer accurate translations.

“As the metaverse begins to take shape, the ability to build technologies that work well in a wider range of languages will help to democratise access to immersive experiences in virtual worlds,” the company explained.

According to Meta, NLLB-200 scored 44 percent higher in the “quality” of translations compared to previous AI research. For some African and Indian-based languages, NLLB-200’s translations were more than 70 percent more accurate.

Meta created a dataset called FLORES-200 to evaluate and improve NLLB-200. The dataset enables researchers to assess FLORES-200’s performance “in 40,000 different language directions.”

Both NLLB-200 and FLORES-200 are being opened to developers to help build on Meta’s work and improve their own translation tools.

Meta has a pool of up to $200,000 in grants for researchers and nonprofit organisations that wish to use NLLB-200 for impactful uses focused on sustainability, food security, gender-based violence, education, or other areas that support UN Sustainable Development Goals. 

However, not everyone is fully convinced by Meta’s latest breakthrough.

“It’s worth bearing in mind, despite the hype, that these models are not the cure-all that they may first appear. The models that Meta uses are massive, unwieldy beasts. So, when you get into the minutiae of individualised use-cases, they can easily find themselves out of their depth – overgeneralised and incapable of performing the specific tasks required of them,” commented Victor Botev, CTO at Iris.ai.

“Another point to note is that the validity of these measurements has yet to be scientifically proven and verified by their peers. The datasets for different languages are too small, as shown by the challenge in creating them in the first place, and the metric they’re using, BLEU, is not particularly applicable.”

A demo of NLLB-200 is available here.

(Photo by Jason Leung on Unsplash)

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta’s NLLB-200 AI model improves translation quality by 44% appeared first on AI News.

The supercomputer is called the AI Research SuperCluster (RSC) and is yet to be fully complete. However, Meta’s researchers have already begun using it for training large natural language processing (NLP) and computer vision models.

RSC is set to be fully built in mid-2022. Meta says that it will be the fastest in the world once complete and the aim is for it to be capable of training models with trillions of parameters.

“We hope RSC will help us build entirely new AI systems that can, for example, power real-time voice translations to large groups of people, each speaking a different language, so they can seamlessly collaborate on a research project or play an AR game together,” wrote Meta in a blog post.

“Ultimately, the work done with RSC will pave the way toward building technologies for the next major computing platform — the metaverse, where AI-driven applications and products will play an important role.”

For production, Meta expects RSC will be 20x faster than Meta’s current V100-based clusters. RSC is also estimated to be 9x faster at running the NVIDIA Collective Communication Library (NCCL) and 3x faster at training large-scale NLP workflows.

A model with tens of billions of parameters can finish training in three weeks compared with nine weeks prior to RSC.

Meta says that its previous AI research infrastructure only leveraged open source and other publicly-available datasets. RSC was designed with the security and privacy controls in mind to allow Meta to use real-world examples from its production systems in production training.

What this means in practice is that Meta can use RSC to advance research for vital tasks such as identifying harmful content on its platforms—using real data from them.

“We believe this is the first time performance, reliability, security, and privacy have been tackled at such a scale,” says Meta.

(Image Credit: Meta)

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo. The next events in the series will be held in Santa Clara on 11-12 May 2022, Amsterdam on 20-21 September 2022, and London on 1-2 December 2022.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

The post Meta claims its new AI supercomputer will set records appeared first on AI News.

PyTorch Live was unveiled during PyTorch Developer Day and enables anyone to build mobile ML demo apps using JavaScript, the world’s most popular programming language.

While on-device AI demos cannot currently be shared, Meta says that functionality is on the way. Developers can start building custom machine learning models to later share with the broader PyTorch community.

PyTorch was publicly launched by Meta back in January 2017, when the company was still known as Facebook. The open-source machine learning library quickly became a firm favourite among the developer and data science communities.

As the PyTorch name suggests, the main library’s interface is designed around Python but it also has a C++ interface. 

The once-dominant machine learning library, TensorFlow, had a two-year headstart on PyTorch but has been falling behind in usage in recent years.

In 2018, GitHub’s Octoverse report highlighted the growth of PyTorch as an open-source project outpacing that of TensorFlow. PyTorch grew by 2.8x that year compared to TensorFlow’s still not insubstantial 1.8x.

That edge for PyTorch appears to be eating into TensorFlow’s early mover advantage.

TensorFlow appeared in three times more job listings in Indeed, Monster, SimplyHired, and LinkedIn as PyTorch in April 2019. However, TensorFlow’s edge in job-listing mentions dropped to 2x in 2020.

Over the past year, PyTorch has also overtaken TensorFlow in worldwide Google searches:

PyTorch Live looks set to accelerate the success of the machine learning library. The tools use React Native for building cross-platform visual user interfaces and PyTorch Mobile powers on-device inference.

Anyone wanting to get started with PyTorch Live can do so through its command-line interface setup and/or its data processing API.

(Image Credit: Meta)

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo North America on 11-12 May 2022.

The post Meta releases PyTorch Live for creating mobile ML demos ‘in minutes’ appeared first on AI News.