TLDR AI 2023-12-20
Harvey raises $80M 💰, Anthropic Terms of Service 🧑⚖️, Google’s VideoPoet ✍️
A Novel Multi-Modal Model for Object Tracking (12 minute read)
Introducing a multi-modal visual prompt tracking model, this project overcomes the limitations of single-modal object tracking by dynamically harnessing the strengths of various modalities like RGB and infrared.
Google’s VideoPoet (4 minute read)
VideoPoet, a groundbreaking language model, is revolutionizing video creation with its unique ability to handle various tasks like text-to-video, video stylization, and even video-to-audio conversion. This approach stands out by combining multiple video generation techniques into one model.
Deepfake Detection Using CNNs (9 minute read)
This study presents a deep learning method for identifying deepfake faces in videos utilizing four pre-trained CNN models for high accuracy.
👨💻
Engineering & Research
Save up to 90% on every flight for life (Sponsor)
For the next 12 hours, get 93% off Dollar Flight Club’s lifetime membership for $129 (normally $1,690) and try it risk-free (3-day money-back guarantee). Fly roundtrip to Paris from $299, Hawaii from $161, and more dream destinations discounted up to 90%.
Start Exploring with $1500 off now.
Enhanced Real-Time Rendering (GitHub Repo)
This project introduces Space-time Supersampling (STSS), a framework that significantly improves high-resolution, high-frame-rate content in real-time rendering.
Distil Whisper (GitHub Repo)
Distil-Whisper is a distilled version of Whisper that is 6 times faster, 49% smaller, and performs within 1% WER on out-of-distribution evaluation sets.
LLMLingua (GitHub Repo)
LLMLingua uses a well-trained small language model after alignment to detect unimportant tokens in a prompt and enable inference with the compressed prompt in black-box LLMs, achieving up to 20x compression with minimal performance loss.
Pilotless FedEx, Reliable Robotics Plane Completes Flight (5 minute read)
Reliable Robotics Corp. has flown a small cargo plane without a human on board. The 12-minute flight, made with a plane on loan from FedEx, was Reliable Robotics' second automated flight. The startup is working to gain full approval from the FAA. Its system will restrict remote pilots to only supervising one aircraft at a time rather than managing multiple autonomous flights. Remote piloting will boost efficiency and allow planes to be repositioned more easily to match where demand is strongest.
A Comprehensive 3D Instruction-Following Dataset (4 minute read)
M3DBench, a new extensive dataset, is set to transform 3D understanding in AI, bridging the gap in multimodal language model research. It features over 320,000 diverse instruction-response pairs, integrating text, images, and 3D objects, paving the way for AI to perform a broader range of real-world 3D tasks.
Tokenize Anything (GitHub Repo)
This new model surpasses previous capabilities by simultaneously performing image segmentation, recognition, and captioning.
Get the most interesting AI stories and breakthroughs delivered in a free daily email.
Join 920,000 readers for
one daily email