Search the archive

FISH SPEECH

Powerful voice generation and single-shot voice cloning. Completely open source and easy to get running.

1X GENIE

Genie is a video generation for world model systems. 1x Robotics has open-sourced a version that mirrors the one it trained internally.

OPENAI SAYS IT'S FIXED ISSUE WHERE CHATGPT APPEARED TO BE MESSAGING USERS UNPROMPTED

A Reddit user reported that OpenAI's ChatGPT initiated a conversation unprompted, leading to speculation about new engagement features. OpenAI acknowledged the issue and issued a fix, attributing it to a glitch with unsent messages. Debate continues over the authenticity of the incident, with similar reports from other users.

ANNOUNCING PIXTRAL 12B

Pixtral 12B excels in multimodal tasks, maintaining state-of-the-art performance on text-only benchmarks, and supports variable image sizes in a 128K token context window. Its architecture includes a new 400M parameter vision encoder and a 12B parameter multimodal decoder based on Mistral Nemo. Pixtral outperforms many open and closed models in multimodal reasoning and instruction following without compromising on text capabilities.

SCALING: THE STATE OF PLAY IN AI

LLMs like ChatGPT and Gemini are becoming increasingly capable as they scale up in size, data, and computing power, leading to improved performance across various tasks. Current Gen2 models like GPT-4 and Claude 3.5 are leading the market, with upcoming Gen3 models expected to further escalate capabilities and costs. The discovery of a new scaling law in AI, pertaining to increased "thinking" during inference, promises further advancements in AI performance beyond just model training.

OVERLAP (PRODUCT LAUNCH)

Overlap (YC S24) is a new AI-powered iOS app that curates the best short video clips on literally any topic you're interested in - built for those quick work or study breaks.

MISTRAL LAUNCHES A FREE TIER FOR DEVELOPERS TO TEST ITS AI MODELS

Mistral AI has launched a free tier to let developers fine-tune and build test apps with its models and slashed API prices by over 50%.

A PROMPTABLE RETRIEVAL MODEL

Promptriever is the first retrieval model that can be prompted like a language model.

Add WorkOS to your app in minutes ↗️

SALESFORCE'S NEW AI STRATEGY ACKNOWLEDGES THAT AI WILL TAKE JOBS

Salesforce is overhauling its AI strategy, introducing generative AI tools for tasks without human supervision and revising its pricing model to $2 per AI-driven conversation. This shift aims to address investor concerns about AI-induced job losses impacting subscription-based revenues. The new AI tools offer higher efficiency and autonomy compared to traditional copilots and chatbots.

QWEN 2.5

An impressive array of open models that approach the frontier of performance. Specifically, they have strong performance on code, math, structured output, and reasoning. The Qwen team has also released a suite of sizes for a variety of use cases.

CREATE FULL WEB APPS WITH LLAMACODER

Together AI and Meta have partnered to make a tool that enables users to build entire apps from just a prompt on the LlamaCoder platform. It is akin to Claude Artifacts but purely created to show off the speed of Together AI's inference engine.

BITNET WITHOUT RETRAINING

The Hugging Face team, using a new technique called quantization warm up, fine-tuned Llama 3 8B and maintained Llama 1 performance while quantizing it to use 1.58 Bits per parameter.

1X WORLD MODEL

1x, a robotics company, has trained a video generation model that can be used to simulate first person views of robotic activity. This can be helpful for offline data generation and robot training.

DIAGRAM OF THOUGHT: ITERATIVE REASONING IN LANGUAGE MODELS

The Diagram of Thought (DoT) framework introduces a new way for large language models to reason by organizing ideas into a directed acyclic graph (DAG). This method allows models to propose, critique, refine, and verify ideas, improving logical consistency.

ZML INFERENCE

ZML is a lightning fast inference engine written in Zig that is designed with performance in mind. It can be used for a variety of models, although it requires some customization to get working for a new architecture.

ADVERSARIAL ATTACKS ON NAVIGATION AGENTS

This repository introduces a new attack method for embodied navigation agents that attaches transparent patches with learnable textures to target objects.

GRAPH ANOMALY DETECTION: A SURVEY

This paper offers an in-depth review of deep learning methods, particularly graph neural networks (GNNs), for detecting anomalies in graph data. Researchers introduce a novel taxonomy of approaches, analyzing GNN designs, proxy tasks, and anomaly measures.

SOCIALAI OFFERS A TWITTER-LIKE DIARY WHERE AI BOTS RESPOND TO YOUR POSTS

SocialAI, a new iOS app, offers a social media experience solely populated by AI-powered bots, eliminating human interaction. Users can share thoughts and receive endless, tailored AI-generated responses, choosing from categories like "supporters" or "critics." Founder Michael Sayman aims to provide a private, interactive space leveraging large language models for diverse feedback.

ACADEMIC TEXT PARSING

AceParse is a dataset designed to improve the parsing of structured texts in academic papers, including formulas, tables, and complex sentences.

JUMPING OVER AI'S UNCANNY VALLEY

This article explores the Uncanny Valley theory, which suggests that discomfort with almost-human AI could hinder its adoption. It examines recent advancements in AI that bring this psychological phenomenon into focus, questioning its potential impact on AI's future. The piece concludes by suggesting that AI's best use may be in a complementary role rather than as a human replacement.

WellSaid

MERCOR'S $30M SERIES A

Mercor raised $30M from Benchmark to build an AI recruiter.

CANCER DETECTION

SkinMamba is a hybrid model that combines CNN and Transformer techniques to improve skin lesion segmentation for early cancer detection.

AMAZON ALEXA CAN NOW BE CONTROLLED BY THOUGHT ALONE - THANKS TO THIS BRAIN IMPLANT

Synchron has enabled an ALS patient to control Amazon's Alexa with a brain implant, bypassing the need for voice or touch.

🌉 Register to attend @ The San Francisco Marriott Marquis

MISTRAL FREE API AND PRICE UPDATE

Mistral has released a free API tier, dramatically reduced its costs, improved the performance of its small model, and put its vision model in Le Chat.

CHALLENGERS ARE COMING FOR NVIDIA'S CROWN

Nvidia's dominance in AI chips has propelled it to immense market value, largely thanks to its GPU capabilities and CUDA software ecosystem. However, competitors like AMD, Intel, Cerebras, and SambaNova are developing innovative solutions to challenge Nvidia's supremacy in AI hardware. While Nvidia's lead remains secure for now, the landscape is dynamic, with multiple players striving to carve out their own niches in the AI market.

JINA EMBEDDINGS V3 (HUGGING FACE HUB)

The Jina series of embeddings are a high quality and powerful suite of models that can be used for embedding and retrieval. Its development team has released the next version of their model with improved performance and training.

TRUSTWORTHINESS OF RAG SYSTEMS

This work introduces a framework to evaluate the trustworthiness of Retrieval-Augmented Generation (RAG) systems across six key areas: factuality, robustness, fairness, transparency, accountability, and privacy.

ENHANCING RECOMMENDER SYSTEMS WITH BEEFORMER

The new beeFormer framework improves sentence Transformers by incorporating interaction data, making them more effective for recommender systems.

AI COMIC UNDERSTANDING

The last frontier of Visual Language Models is the ability to understand and reason about comics. This project is a survey and a call for research.

WORD LLAMA

A lightweight toolkit for fuzzy deduplication, reranking, and other NLP-based tasks. Optimized to run on the CPU.

SYLLABLE SEGMENTATION IN SPEECH MODELS

This project enhances speech representation learning by separating syllabic structures from speaker information in self-supervised models. By fine-tuning the HuBERT model with speaker perturbation techniques, researchers improved syllable segmentation, leading to better syllabic unit organization.

DATA PIPELINES ARE THE NEW AI SECRET SAUCE

With models being somewhat commoditized, much of the advantage in AI comes from the data. It also, by extension, comes from the pipeline that ingests and creates the data. This post discusses the challenges and opportunities associated with data pipelines in the modern age.

WHY COPILOT IS MAKING PROGRAMMERS WORSE AT PROGRAMMING

AI tools like GitHub Copilot enhance programming productivity but risk eroding essential coding skills. Over-reliance on AI-generated code can lead to quality, security, and maintainability issues and reduce learning opportunities. These tools may also limit creative problem-solving and foster a false sense of expertise among developers.

SURVEILLANCE VIDEO SUMMARIZATION

A custom-trained Florence 2-based model and system that can summarize CCTV and Surveillance videos and give accurate updates about what is occurring at any time.

TIKTOK'S OWNER WANTS TO DESIGN ITS OWN AI CHIPS

ByteDance is developing its own AI chips with TSMC to reduce reliance on Nvidia GPUs amid U.S. export controls.

SALESFORCE UNLEASHES ITS FIRST AI AGENTS

Salesforce has debuted Agentforce, its effort to create generative AI bots capable of taking action on their own - within established limits.

55% off for TLDR readers! Pick a language and get started

REFLECTION 70B MODEL MAKER BREAKS SILENCE AMID FRAUD ACCUSATIONS

Matt Shumer, CEO of OthersideAI, faced criticism after third-party researchers couldn't replicate the performance of his new large language model, Reflection 70B. Shumer attributed the discrepancies to issues during the model's upload process and has apologized for getting ahead of himself. The AI community remains skeptical and awaits further clarification.

HOW MEMPHIS BECAME A BATTLEGROUND OVER ELON MUSK'S XAI SUPERCOMPUTER

Elon Musk's xAI is building "Colossus," the world's largest supercomputer, in Memphis to support its AI chatbot, Grok. The project has faced criticism for environmental oversight issues and its substantial energy and water demands. Despite these concerns, xAI aims to rapidly advance its AI capabilities while impacting the local community.

RUNWAY ANNOUNCES AN API FOR ITS VIDEO-GENERATING AI MODELS

Runway announced an API to integrate its Gen-3 Alpha Turbo video-generation model into third-party platforms, with pricing at one cent per credit. Concerns over copyrighted training data persist, with Runway refusing to disclose sources and similar claims impacting competitors like OpenAI and Nvidia. Legal uncertainties continue, but AI-driven video tools are expected to disrupt the film and TV industry significantly.

INSTANTDRAG: IMPROVING INTERACTIVITY IN DRAG-BASED IMAGE EDITING

Editing images can be challenging due to the continuous nature of pixels. This work builds on the great line of research on drag-based editing. It takes a set of control points given by a user and updates images accordingly. Previous works are quite slow - this paper speeds things up considerably.