. .AI: Agent startup Adept launches powerful new model, and more (1.24.24)

Friends, four stories led the AI community conversation today, two technical and two business stories.

Thanks for joining us here. If you’re in a podcast listening mood, my mind was expanded in some big exciting ways by friend of the newsletter Gianni Giacomelli’s appearance on Ross Dawson’s Amplifying Cognition podcast last week. I listened to it while cleaning my kitchen last night and have already upgraded by ChatGPT prompts to a whole new level.

Now here’s today’s news…

-Marshall Kirkpatrick, Editor

Fuyu-Heavy: Adept's New Contender Against Google's, X's, and Other Generalist Models

First impacted: AI developers, AI researchers, Automation Engineers
Time to impact: Short

AI agent startup Adept has launched a new generalist model named Fuyu-Heavy. They claim the model outperforms others in its class and is only surpassed by models 10 to 20 times its size. Despite challenges in training it on both text and image data, Fuyu-Heavy has shown impressive abilities in understanding user interfaces. It reportedly also has the capability to excel in long-form conversations, reportedly matching the larger Claude 2.0 model in chat evaluations, even while managing image modeling tasks. [Adept Fuyu-Heavy: A new multimodal model] Explore more of our coverage of: Digital Agent Model, Adept Fuyu-Heavy, MMMU Benchmark. Share this story by email

Google Research Launches an Impressive New Video Generation Model, Lumiere

First impacted: Content Creators, Video Editors, Artists
Time to impact: Short

Google has launched Lumiere, a text-to-video diffusion and an image-to-video model that it says produces realistic, coherent motion videos directly from the source content. The company says Lumiere leverages a pre-trained text-to-image diffusion model to generate full-frame, low-resolution videos in one pass, and can be used for tasks such as image-to-video conversion, video inpainting, and stylized generation. It does this in part by creating the entire span of the video at once. This differs from previous models that first create individual keyframes and then use temporal super-resolution to blend images. This new method enables Lumiere to maintain a consistent global temporal sequence. [Lumiere - Google Research] Explore more of our coverage of: Google Research, Text-to-Video Diffusion, AI Video Generation. Share this story by email

Spellbook Raises $20 Million in Series A Funding for Legal Tech

First impacted: Lawyers, In-house Legal Teams
Time to impact: Short

Canadian legal AI firm Spellbook has secured $20 million in Series A funding, with Inovia Capital leading the round. The company provides an AI tool for contract drafting and review, for customers ranging from solo lawyers to large in-house legal teams. [Legal AI startup Spellbook raises $20 mln as sector draws more investments] Explore more of our coverage of: AI Legal Technology, Series A Funding, Contract Drafting AI. Share this story by email

Voltron Data Acquires Claypot for Real-Time Analytics Expansion

First impacted: Data analysts, AI engineers
Time to impact: Short to Medium

San Francisco-based startup Voltron Data ($110M raised) has acquired Claypot, a real-time AI platform. Claypot's team will join Voltron's team, enhancing their offering with real-time analytics capabilities. This addition is aimed at helping enterprises unlock value from fresh data, aiding use cases like fraud detection and personalization. The acquisition follows Voltron's recent launch of Theseus, a distributed query engine designed to streamline data preprocessing and AI tasks. With the integration of Claypot, Voltron's modular systems, previously focused on local and batch data, will now be able to deliver analytics from streaming data. [Exclusive: Voltron Data acquires Claypot to unlock real-time AI with modular data systems] Explore more of our coverage of: AI Acquisition, Real-Time Analytics, Data Processing. Share this story by email

That’s it! More AI stories tomorrow!