🥣 Apple Drops Open-Source LLM, Ferret

PLUS: Google Announces VideoPoet

Read time: 4.5 minutes

Hello, AI enthusiasts! Mr. Byte here, ready to ring in the New Year with a fresh edition of The AI Crunch. As we stand on the cusp of 2024, let's dive into the exciting world of AI advancements. Prepare to step into a new year filled with innovative AI breakthroughs. Are you ready to explore the latest in AI as we welcome a promising year ahead? Let's jump in!

Top Stories

Apple Drops Open-Source LLM, Ferret

In a move that's about as low-key as Apple gets, they've teamed up with Cornell University to drop an open-source, multimodal large language model (LLM) called Ferret. Sneaking it onto GitHub last October without so much as a peep, Ferret has since caught the eye of AI researchers, probably because Apple doing anything 'open source' is like spotting a unicorn.

Ferret's party trick? It looks at parts of an image, identifies what's interesting, and then uses those elements to answer queries. Highlight an animal in a photo, ask Ferret what it is, and bam! – it tells you the species and maybe even what it's up to, using the context of other elements in the image.

This move is a bit of a plot twist for Apple, known more for its secrecy than sharing. But hey, when you can't throw computing power at AI like Google or Microsoft, you've got to get creative. By going open source, Apple's playing a different game – think Meta Platforms' style – potentially a savvy move to stay in the AI race without having to cozy up to the cloud bigwigs.

Google Announces VideoPoet

Google, in its quest to make every sci-fi fan's dream a reality, has whipped up something called VideoPoet, which sounds like it should be narrating epic tales rather than processing videos. But hey, Google loves to surprise. VideoPoet, still in its VIP-only phase (sorry, no public access yet), is like the Mozart of video generation, trained on the fancy-sounding MAGVIT-2.

So, what's the big deal? Well, VideoPoet is basically your text-to-video wizard. Give it a prompt, and it conjures up high-motion, variable-length videos complete with matching audio. No extra text needed. It’s like having a Hollywood studio in your pocket, minus the Hollywood budget.

The brains behind this magic? A video tokenizer and an audio tokenizer, turning images and sounds into something VideoPoet can understand. Then, it uses its autoregressive language model to create videos that are not just visually stunning but also make sense.

If you're itching to see it in action, Google's teasing us with a demo website and a research paper – because reading technical papers is everyone's idea of fun, right? For now, VideoPoet is a glimpse into a future where creating videos is as easy as typing a text message. Keep an eye on this one; it's shaping up to be a game-changer in AI video creation.


Quick bites to start your morning:

  • Runway just introduced adjustable video ratios (X)

  • AI defeats humans at dexterity game (TNW)

  • Google launches AI chatbot for product help (925)

  • Students created PIGEON, an AI that identifies image locations (OA)

  • Meta’s CTO on the generative AI craze (SE)

  • Stable Video Diffusion is now available via API (SA)

  • Major consumer LLM services receive ~2B visits a month (X)

  • Researchers identify a new class of antibiotic using AI (PO)

  • Microsoft unveiled new AI tool allowing text to song through Suno (SU)

  • Nat Friedman reveals AI that can read ancient text (X)

  • AI machine cannot be called an inventor, rules UK court (FT)

AI Funding

Virtualleap: AI VR startup raised $3M (XRT)

OpenAI: In early talks to raise at a valuation at or above $100B (BBG)

Anthropic: Is discussing raising $750M at a ~$15M valuation (R)

Rabbit: A startup building an AI-enabled operating system raised $10M (BW)

Inverted AI: Raised $4M for generative AI in AV/ADAS development (PRN)

Jenesys AI: Secures $1M to integrate generative AI into transactional accounting (EUS)

Totus Medicines: Raised $66M to revolutionize small molecule drug discovery (FM)

ImpriMed: Raised $23M to expand its AI-driven human healthcare approach (BW)

AI Tools

Newest AI tools (AI Tool Leaderboards):

  • Taplio: Grow your personal brand on LinkedIn

  • AI Social Bio: Automatically generate your social media bio

  • Instantly: Automate outreach campaigns for more replies & revenue

  • Robin: Automate sales outreach with Robin AI - no human needed

  • RivalFlowAI: Improve existing content with RivalFlowAI

  • Bunni: Chat to your PDF documents

  • Face Shape AI: Your personal face analysis companion

  • AI Undetect: AI content generation with detection bypass

  • SlidesPilot: AI PowerPoint presentation generator

  • Varys AI: Interior design AI for professional, GPT for room and space

  • IllumiDesk: Create interactive courses 10x faster with AI

  • YTClass: Elevate your YouTube learning experience with AI

If you have anything interesting to share, please reach out to us by sending us a DM on Twitter: @theaicrunch

❤️ Crunchy Review of the Week