UCLA researchers develop AI model that analyzes medical scans 5,000x faster
Plus Perplexity AI is in talks to raise $500M+ despite facing lawsuits
Today’s Highlights:
📰 News: UCLA researchers develop AI model that analyzes medical scans 5,000x faster + Archetype AI releases groundbreaking AI that learns physics from scratch
💰 Funding: Perplexity AI is in talks to $500M+ and OpenAI’s former CTO is raising $100M for a new venture
⚡️ Top News Stories:
1. UCLA researchers have developed SLIViT, a deep-learning AI model that analyzes 3D medical scans like MRIs and CTs with expert-level accuracy, operating 5,000 times faster than human specialists, while requiring fewer training samples, making it more practical and scalable for real-world healthcare applications.
NVIDIA GPUs (T4 and V100 Tensor Core) powered the research, enabling SLIViT to deliver expert-level analysis in a fraction of the time taken by human specialists.
SLIViT’s adaptability spans various imaging modalities, including retinal scans, MRIs, ultrasounds, and CTs, offering versatility across different organs and diseases.
SLIViT's scalability and ease of deployment could democratize medical imaging analysis, particularly in areas with a shortage of clinical experts, improving patient outcomes.
2. Archetype AI unveils Newton, an AI model that learns physical laws from data alone, predicting behaviors without prior equations or retraining.
Archetype AI’s Newton model can predict and encode physical behaviors using sensor data without prior knowledge of physical laws, representing a breakthrough in AI’s ability to learn from raw data.
Unlike traditional models trained on specific equations, Newton's zero-shot forecasting allows it to generalize across various physical systems, from fluid flows to electrical grids.
Newton’s foundation model was trained on 0.59 billion data points from open-source datasets, enabling it to predict complex behaviors, like electrical grid performance, without needing extensive retraining.
This AI model could have broad applications, from autonomous systems that adapt to new environments to revolutionizing fields like robotics, automotive systems, and scientific discovery.
3. Meta FAIR introduced new AI models and datasets, including updates to image segmentation, multimodal language models, and performance-boosting tools for LLMs.
Spirit LM, a multimodal language model, integrates text and speech, generating more expressive and natural-sounding speech.
SAM 2.1 offers enhanced image and video segmentation, improving on its predecessor’s capabilities with over 700,000 downloads in 11 weeks.
Layer Skip accelerates LLM generation times by nearly 2x, providing a significant performance boost without requiring specialized hardware.
New tools also include SALSA for cryptography security testing, Meta Lingua for language model training, and Open Materials 2024 for AI-driven materials discovery.
4. News Corp has filed a lawsuit against Perplexity AI, accusing it of massive copyright violations by using content from *The Wall Street Journal* and *New York Post* without permission.
The lawsuit claims Perplexity copies and scrapes articles, reproduces content verbatim, and drives traffic away from original sources, harming News Corp’s revenue. Perplexity is also accused of encouraging users to "skip the links" to original sources, competing directly with publishers.
News Corp seeks up to $150,000 per infringement and demands the destruction of databases containing its content, with similar accusations from other outlets like *Wired* and *Forbes*.
5. Meta has partnered with Blumhouse and select filmmakers, to test its AI-powered Movie Gen, a suite of models that allows users to create custom HD videos and sounds from text inputs, with plans for a public launch in 2025.
6. Microsoft is launching Copilot Studio in public preview, enabling businesses to build autonomous AI agents for automating key functions like lead generation, sales, service, finance, and supply chain management, with ten new agents available in Dynamics 365.
7. Elon Musk’s xAI has launched the Grok API, starting with the grok-beta model, priced at $5 per million input tokens or $15 per million output tokens.
8. Midjourney is rolling out a new tool next week for select users that will allow AI-based image editing and retexturing, enabling users to modify uploaded images by changing textures and colors based on captions.
9. Anthropic has developed new AI sabotage evaluations to test potential risks of AI systems, including manipulating human decisions, inserting bugs, and undermining oversight.
10.Apple's in-house AI technology is reportedly two years behind competitors like Google, OpenAI, and Meta, with internal tests showing ChatGPT is 25% more accurate and answers 30% more questions than Apple's new Siri.
11. Bain & Company has expanded its partnership with OpenAI, establishing a Center of Excellence to accelerate AI implementation and provide advanced AI-driven services across industries like retail and healthcare, building on their 2022 collaboration, with a focus on co-designing tailored AI solutions using OpenAI’s latest innovations, such as GPT-4o and OpenAI o1.
12. Penguin Random House has updated its copyright pages globally to explicitly prohibit AI firms from using its books for training AI models, becoming the first major publisher to adopt this policy. The new language reserves PRH titles from the text and data mining exception in the European Parliament directive, as the publisher aims to balance defending authors' intellectual property while responsibly integrating AI tools in publishing.
💰 Top Funding News:
1. Perplexity AI is in talks to raise $500M, potentially increasing its valuation to over $8Bn
The AI startup's annualized revenue is estimated at around $50 million, based on recent sales.
Perplexity AI is supported by Jeff Bezos and Nvidia, which has helped boost its growth.
2. Mira Murati, former CTO of OpenAI, is raising over $100M for her new AI startup, which aims to develop proprietary AI products and models.
Murati left OpenAI in September, citing a desire for personal exploration after helping the company transform AI systems’ learning and reasoning capabilities.
Murati was instrumental in developing ChatGPT and DALL-E and played a key role in OpenAI's partnership with Microsoft, solidifying her position as an AI leader.
Barret Zoph, another prominent ex-OpenAI researcher, may also join the venture, as Murati is reportedly recruiting OpenAI employees.
3. SandboxAQ, an Alphabet Inc. spin-off, which specializes in AI and quantum technology, is raising funds at a valuation over $5Bn, pre-money.
The startup offers simulation software for drug discovery, cybersecurity, and quantum sensor applications.
4. Carbon Robotics, which uses AI-powered computer vision and deep learning in its LaserWeeder™ to identify and eliminate weeds with precision, significantly reducing the need for chemical herbicides, manual labor, and soil disruption, raised a $70M Series D, led by BOND, w/ NVentures (NVIDIA’s venture capital arm), Anthos Capital, Fuse Venture Capital, Ignition Partners, Revolution, Sozo Ventures, and Voyager Capital.
5. Neysa, which provides AI and ML infrastructure as a service (IaaS) and platform as a service (PaaS), raised a $30M Series A, co-led by NTTVC, Z47 (formerly Matrix Partners India), and Nexus Venture Partners.