This Week in AI: Meta's Tower of Babel, Hugging Face Raises a Mega Round
SeamlessM4T Speaks and Translates 100 Languages, Hugging Face Raises from Salesforce & Nvidia, White House Announces $20M AI Hackathon
SeamlessM4T: Meta’s Revolutionary AI that Speaks & Translates 100 Languages
In a world where "lost in translation" might soon be a phrase of the past, Meta introduces SeamlessM4T, a jack-of-all-trades AI that can juggle nearly 100 languages, translating them from speech to text, text to speech, and every combo in between. This linguistic prodigy doesn’t just stop at recognizing languages; it can even detect when you're mixing up your Telugu with Hindi in a single sentence and smoothly translate it into English.
Meanwhile, Google's peeking over the fence with its Universal Speech Model, hinting that the translation turf war is just heating up.
From Teen Chatbots to a $4.5 Billion: Hugging Face Raises Series D Led by Salesforce
AI darling Hugging Face raised $235 million in Series D funding, skyrocketing their valuation from $1.9 billion to $4.5 billion. Not too bad for a company that started off trying to chat up teenagers!
They're now the go-to hub for AI code repositories, models, and datasets, boasting over 1 million repositories. With big names like Google, Amazon, and Nvidia throwing their weight (and wallets) behind them, it's clear Hugging Face is more than just a pretty name. From their ambitious BigScience project aiming to rival OpenAI's GPT-3, to collaborations with cloud giants like Nvidia and Amazon, this Brooklyn-based startup is on a mission to be the open platform that empowers AI.
Hugging Face Releases IDEFICS: The Flamingo-Inspired AI That Processes Both Images & Text
As if we haven't talked about them enough, Hugging Face also released IDEFICS, inspired by DeepMind's elusive Flamingo.
This open-source visual language model, which processes both images and text. This means it can answer questions about images, describe visual content, and even craft stories based on multiple images.
Think of it as a sophisticated AI that can "see" pictures and then "talk" about them. It's an open-access model, designed to replicate the capabilities of Flamingo, a model by DeepMind. IDEFICS offers two versions with different complexities: one with 9 billion parameters and another with 80 billion parameters. 🦩🤖
White House Announces $20M AI Hackathon to Fortify Digital Defenses
The White House has unveiled the "AI Cyber Challenge", a DARPA-led competition seeking AI systems that can spot and patch software vulnerabilities. This two-year tech showdown, backed by tech giants like Google and OpenAI, aims to bolster the nation's digital defenses. With a whopping $20 million prize pool, the contest hopes to attract both big names and small businesses. So, if your AI can fend off cyberattacks and fix bugs on the fly, you might just walk away $4 million richer!
The White House, in collaboration with tech giants like Google and OpenAI, has initiated the "AI Cyber Challenge" to develop AI that can spot and fix software vulnerabilities. The competition is offering $20 million in prizes. Winning teams will be encouraged to open-source their solutions.