Latest In Web3, AI & Emerging Tech

Hello, and welcome to the latest edition of my weekly newsletter, where I bring you updates on the latest happenings in the world of Web3, AI, and Emerging Technologies.?

There are just so many things happening all at the same time in this industry and it’s often a job in itself to keep up to date with the latest, so it’s my pleasure to digest it all for you and share with you some of my favourites or most important developments in the space.

So, without further ado, let's dive into this week's news!

Empowering Multilingual Communication: Meta AI Introduces Speech-to-Text and Text-to-Speech for 1,100+ Languages

Meta AI takes a significant step toward fostering language inclusivity and accessibility with its groundbreaking Massively Multilingual Speech (MMS) project. By harnessing the power of self-supervised learning and a comprehensive dataset covering over 1,100 languages, Meta AI revolutionizes the field of multilingual speech recognition and generation. This initiative aims to bridge communication gaps and preserve linguistic diversity across the globe.

Overcoming the challenge of limited labelled speech data, the MMS project pioneers an innovative approach. Leveraging a rich collection of religious texts, including translated versions of religious scriptures, Meta AI extracts publicly available audio recordings in numerous languages. This dataset encompasses readings of religious texts, such as the New Testament, in over 1,100 languages, with an average of 32 hours of audio per language. Furthermore, including unlabeled recordings from diverse Christian religious sources expands language coverage to nearly 4,000 languages. Notably, the models demonstrate equal proficiency across genders and remain unbiased toward religious language.

To ensure data quality and alignment, the MMS dataset undergoes meticulous preprocessing. Employing alignment models and efficient forced alignment algorithms, Meta AI aligns audio recordings with their corresponding transcriptions, enhancing accuracy and usability. Building upon their pioneering work in self-supervised learning, Meta AI's MMS project leverages the wav2vec 2.0 framework. By training on approximately 500,000 hours of speech data across 1,400 languages, the models achieve remarkable performance in multilingual speech recognition, language identification, and text-to-speech synthesis.

The results of the MMS project showcase its transformative impact on multilingual communication. The models surpass existing systems regarding language coverage and accuracy while preserving linguistic diversity. With support for over 1,100 languages, the MMS models achieve substantially lower error rates compared to previous benchmarks, rivalling systems like OpenAI Whisper. Additionally, the MMS project excels in language identification accuracy, facilitating information accessibility in diverse languages. Moreover, the models produce high-quality synthesized speech for over 1,100 languages, empowering individuals to communicate and engage in their preferred languages. Meta AI's commitment to collaboration within the AI community further ensures responsible development and continuous improvement in speech accessibility and inclusivity.

OpenAI Debuts Official ChatGPT App for iOS, Android Launch on the Horizon

OpenAI has made a significant move by releasing the official version of its AI-powered chatbot, ChatGPT, on Apple's App Store. The company has also hinted at an upcoming Android version in the works.

The ChatGPT app offers users the ability to sync their chat history with the web version and introduces voice input, leveraging OpenAI's speech recognition model, Whisper. Initially, the app is available to iPhone and iPad users in the United States, with plans for expansion to other countries in the coming weeks.

A successful global rollout of the ChatGPT app would grant access to a significant portion of Apple's 2 billion active devices, further propelling the adoption of AI technology.

Enthusiasts have long sought ways to utilize ChatGPT's capabilities on mobile devices, often resorting to complex workarounds to integrate the chatbot with Apple's virtual assistant, Siri. While some have relied on Microsoft's Bing app, which provides access to their GPT-4-powered chatbot, OpenAI's official app brings a native ChatGPT experience to iOS users.

Both Apple's App Store and the Google Play store have seen an influx of apps claiming to offer similar services to ChatGPT. However, caution is advised, as some of these apps have turned out to be copycats charging exorbitant subscription fees. OpenAI's official app ensures a trusted and reliable experience for users.

Existing ChatGPT Plus subscribers will enjoy the benefits of GPT-4's enhanced capabilities, along with early access to new features and faster response times on their iOS devices. OpenAI's commitment to expanding the app's availability underscores the growing demand for AI-powered chatbots in the mobile space.

Instagram's New Text-Based App: A Twitter Competitor Emerges

Instagram is set to introduce a new text-based app, which could potentially rival Twitter's functionality. Leaked details shared by Lia Haberman shed light on the upcoming release, revealing a leaked marketing slide and some key features of the app.

Referred to as "Instagram's new text-based app for conversations," the app is codenamed P92 or Barcelona. Users will be able to seamlessly sign in using their existing Instagram credentials, with their followers, handle, bio, and verification carrying over to the new app.

The app will feature a feed where users can create text posts of up to 500 characters, accompanied by links, photos, and videos. The leaked marketing slide suggests that the app's interface combines elements of Instagram and Twitter, with Meta aiming to provide robust moderation controls from the start. Users will have settings to manage replies and account mentions, and blocked accounts on Instagram will remain blocked in the new app.

An intriguing aspect of the Instagram text app is its element of decentralization. According to the leaked slide, the app will be compatible with other platforms like Mastodon. This compatibility will enable users from these platforms to search, follow, and interact with profiles and content, provided the user's profile is public or has approved the follower request for private accounts. Meta is likely exploring ActivityPub, a decentralized social networking protocol, for this compatibility.

If the app gains widespread release, it has the potential to further cement Instagram's popularity. As Twitter faces ongoing challenges, users are actively seeking alternative platforms for sharing tweet-like updates. While platforms like Mastodon and Bluesky have gained traction, the introduction of a text-based Instagram app would offer existing users a seamless transition without the need to rebuild their following in yet another space.

Meta, the parent company of Instagram, has declined to comment on the leaked details. However, their previous statement about exploring a standalone decentralised social network for text updates aligns with the rumoured Instagram text-based app, highlighting the company's focus on providing creators and public figures a dedicated space for sharing timely updates.