May 10, 2024

• Roman snail dye, known as Tyrian purple, valued more than gold, has been discovered in the UK for the first time at a cricket club, marking a significant archaeological find.

• An investigation revealed a nationwide scam on Airbnb, involving fake hosts and nearly 100 property listings across eight cities, exploiting the platform's policies for financial gain.

• Consistency Large Language Models (CLLMs) introduce a new method for parallel decoding, significantly reducing inference latency by efficiently decoding an $n$-token sequence per inference step, leveraging Jacobi trajectories for training. Read the paper for detailed insights.

• ElevenLabs Music introduces an early preview showcasing songs generated from a single text prompt, demonstrating the platform's innovative approach to music creation.

• Wilson Lin mapped and analyzed 40 million Hacker News posts and comments using text embeddings to create a semantic map, aiming to enhance search, recommendations, and topic sentiment analysis. Data and code available here.

• Datatype99 introduces type-safe, portable, and predictable algebraic data types (ADTs) for C99, enhancing safety and developer productivity without external dependencies.

• AdVon, an AI company known for creating fake writers at Sports Illustrated, has been found to produce vast amounts of AI-generated content for prominent publications, blurring the lines between journalism and advertising.

• Multimodal models like CLIP and Stable-Diffusion do not exhibit true "zero-shot" generalization but instead require exponentially more data for linear improvements in performance on new concepts.

• Muddy is a browser designed for collaborative work, automatically organizing project files and web apps into a unified, personalized workspace for each user, as demonstrated in their YouTube demo.

• Ellipsis is an AI-powered development tool that automates code reviews, generates code from GitHub comments, and addresses pull request comments with working, tested code, enhancing the speed of software development.

• The author's paper was accused of plagiarism due to a 92% match with a pre-print version of the same paper on Arxiv, highlighting a significant oversight in the plagiarism detection process.

• CS388 offers a comprehensive online master's course in Natural Language Processing (NLP), featuring a curriculum that spans from linear sentiment classification to advanced topics like transformers and large language models, accessible through a YouTube playlist.

• AlphaFold 3 by DeepMind introduces marginal improvements in protein folding accuracy and more than doubles accuracy in protein antibodies, marking a significant advancement in computational biology.

• Interaction with large language models (LLMs) like GPT-4 in text-adventure games enables emergent behaviors, allowing players to co-create game narratives beyond the original storyline.

• Vision Transformers Need Registers introduces additional tokens to Vision Transformer inputs, eliminating artifacts in feature maps and setting new benchmarks for self-supervised visual models. Read more