Search the archive

MODAL BUYS TIDBYT

The first acquisition by the elastic scaling GPU company. Modal bought Tidbyt, an NYC-based hardware company, for the in-house talent of the team around infrastructure and containerization.

OPENAI FACES AI IMPROVEMENT SLOWDOWN

OpenAI's next major model, code-named "Orion," reportedly shows less improvement over previous models, suggesting a slowdown in AI advancement. To tackle this, OpenAI has formed a foundations team focused on enhancing models with alternative methods, such as synthetic data training and post-training adjustments, as the availability of new data has decreased.

WORLD'S LARGEST OPEN-SOURCE MODEL

Near Protocol has announced plans to create a massive 1.4 trillion parameter open-source AI model. It aims to surpass existing models like Meta's Llama.

FRONTIER MATH BENCHMARK

Epoch AI has released a new challenging mathematics benchmark. Most frontier models are unable to solve more than 2% of the problems.

BITNET A4.8: 4-BIT ACTIVATIONS FOR 1-BIT LLMS

One of the key challenges with 1.58bit LLMs was the lack of hardware acceleration support. This work proposes 4.8bit activations to take advantage of the INT4/FP4 kernels in new hardware. It comes at no runtime cost.

SUPERCHARGING CLIP WITH LLMS

LLM2CLIP combines CLIP's visual and textual alignment with the advanced language understanding of LLMs.

TORCH COMPATIBLE MUON OPTIMIZER

Muon was the optimizer used for the GPT-2 training record. It is a momentum adapted SGD style method. This repository contains an implementation that can be dropped in for AdamW.

MOCHI VIDEO MODEL WITH OPTIMIZED INFERENCE

Mochi is the best open source text-to-video model. On launch, it took 8 H100s to run. Now, thanks to the community, it can be run on a single 48GB L40 with no loss in quality.

TRAINABLE PYTORCH REPRODUCTION OF ALPHAFOLD3

Protenix is a working and trainable reproduction of AlphaFold3, a protein folding project from DeepMind. It was written by Bytedance's 'AI for Science' team.

LLAMAPREVIEW

LlamaPReview is an AI assistant for GitHub that offers one-click installation and auto-reviews of pull requests with context-aware analysis. It supports multiple programming languages and integrates with GitHub Actions, providing insightful feedback directly on PRs. Currently available for free, it enhances code quality by identifying issues and suggesting optimizations.

SMOLLM2

Hugging Face's SmolLM2 is a compact language model family with sizes ranging from 135M to 1.7B parameters trained on 11 trillion tokens. These models efficiently run on-device and support various tasks, with weights available under the Apache 2 license. Quantized models like the 1.7GB and 138MB versions provide diverse flexibility for different computational needs.

EMBEDDINGS ARE UNDERRATED

Machine learning embeddings can transform technical writing by enabling mathematical comparisons of arbitrary text, improving features like recommendation systems with semantic similarities. They place text within a multi-dimensional space, providing intuitive semantic relationships, useful for tasks like identifying related content. Docs site owners who offer embeddings for their content could foster innovative applications from their communities.

SAMSUNG DEBUTS AI-POWERED 'NEXT-GENERATION BIXBY,' BUT YOU CAN'T USE IT YET

Samsung has launched a "next-generation Bixby" with enhanced AI capabilities on the Galaxy W25 and W25 Flip in China.

EVEN MICROSOFT NOTEPAD IS GETTING AI TEXT EDITING NOW

Microsoft is integrating AI-powered text editing in Notepad, enabling users to rephrase, adjust tone, and modify text length through a feature called Rewrite.

AI FOR REAL-TIME FUSION PLASMA BEHAVIOR PREDICTION AND MANIPULATION

A new multimodal machine learning methodology enhances super-resolution data for better analysis of complex fusion plasma phenomena, such as Edge Localized Modes (ELM), aiding in the stabilization of future fusion reactors.

your compliance program into gear

MISTRAL LAUNCHES MODERATION API

Mistral has launched batch processing and a moderation API as part of its broader API platform.

PALANTIR AND ANTHROPIC PARTNERSHIP

Palantir and Anthropic have partnered to bring the Claude suite of models to AWS for US intelligence agencies and defense operations.

CHATGPT CAN NOW CONTROL A ROBOT ARM

Researchers at UC Berkeley and ETH Zurich used GPT-4 to train affordable robot arms to clean spills. They achieved this by integrating a multimodal agent called LangChain to convert LLM inputs into robot actions. The research showcases a novel proof-of-concept for human-robot interaction and democratizes robotics with open-source technology.

MAGNETIC ONE: GENERALIST MULTI AGENT MODEL

Microsoft has released a multi-agent system built on top of the company's Autogen library. It uses GPT-4o as a backend model for agentic behavior.

COSMO TOKENIZER

Nvidia has released state-of-the-art image and video tokenizer and compression models. These can be used to train video generation, VLMs, and other multimodal models. Nvidia has released the inference code, a paper, and model weights.

3D INSTANCE SEGMENTATION

SA3DIP is a new method for improving 3D instance segmentation that integrates additional 3D priors beyond standard 2D models.

best solution. Read more on the Appfire blog

G SPLAT

G Splat is a powerful package and studio for doing Gaussian splat research.

A NEW METHOD TO IMPROVE VISION-LANGUAGE MODELS

RaVL is a new method that boosts the accuracy of vision-language models by focusing on local image features rather than the entire image, aiming to avoid misleading correlations.

3D SCENE SYNTHESIS

SCGaussian is a novel approach to 3D scene synthesis that maintains structural consistency even with sparse input.

A FASTER, BETTER WAY TO TRAIN GENERAL-PURPOSE ROBOTS

MIT researchers developed Heterogeneous Pretrained Transformers (HPT) to train robots using diverse data from different domains and modalities. This method enables easier adaptation to new tasks without starting training from scratch, improving performance by over 20% compared to traditional training. Inspired by LLMs, HPT combines vast datasets to enhance robot learning capabilities across varied environments.

THE RISE OF THE AGENTIC WEB

The Agentic Web is advancing AI agents with onchain functionalities, facilitating the creation, ownership, and transaction capabilities for these agents. Platforms like Replit, VIRTUALS.io, and Wayfinder enable AI integration with blockchain for activities like asset management, data retrieval, and decentralized applications. This shift supports AI-driven automation in payments, trading, and decentralized finance onchain.

THE PRESENT FUTURE: AI'S IMPACT LONG BEFORE SUPERINTELLIGENCE

More powerful AI models will soon surpass human intelligence, enabling transformational advancements in work and society. Current AI models like Claude are already impacting industries by automating tasks, providing safety monitoring, and facilitating interactions using multimodal inputs and outputs. Organizations must navigate ethical considerations to ensure AI enhances rather than replaces human capabilities.

VLMS ARE IN-CONTEXT LEARNERS

Visual Language Models (VLMs) can acquire skills via prompts.

OPENAI IN TALKS WITH REGULATORS TO BECOME A FOR-PROFIT COMPANY

This shift is significant as it involves valuing OpenAI's lucrative LLM technology while maintaining a nonprofit arm to uphold its original mission.

GOOGLE'S BIG SLEEP LLM AGENT DISCOVERS EXPLOITABLE BUG IN SQLITE

The agent used variant analysis and human-like workflows to identify the vulnerability, which traditional fuzzing missed.

ANTHROPIC HIKES THE PRICE OF ITS HAIKU MODEL

Anthropic's new AI model, Claude 3.5 Haiku, outperforms Claude 3 Opus but costs significantly more. The model lacks image analysis capabilities but excels in coding, data extraction, and content moderation. The price increase raises questions about Anthropic's future pricing strategy.

OPENAI ACQUIRED CHAT.COM

OpenAI acquired the domain Chat.com, which now redirects to ChatGPT. The domain was previously purchased for $15.5 million by HubSpot's co-founder. OpenAI's acquisition likely involved shares, but financial details remain undisclosed.

A SCALABLE COMMUNICATION PROTOCOL FOR NETWORKS OF LARGE LANGUAGE MODELS

Agora is a cross-platform, dead-simple protocol for efficient communication between LLM agents. It enables very different agents to communicate with each other at a fraction of the cost. Agora can be easily integrated with existing multiagent frameworks, such as Camel AI, LangChain, and Swarm.

REDUCING HALLUCINATIONS IN VISION-LANGUAGE MODELS

Vision-guided Direct Preference Optimization (V-DPO) addresses hallucination issues in large vision-language models (LVLMs) where text responses can stray from visual input due to an over-reliance on language.

A SIMPLE APPROACH TO VISION-LANGUAGE CLASSIFICATION

SuperClass is a straightforward classification model for vision-language tasks that skips the need for a text encoder, unlike contrastive models like CLIP. SuperClass avoids complex text filtering and large batch requirements by directly using tokenized raw text as classification labels.

🗣️Talk with Flow instantly (no signup required!)

ENHANCING RAG WITH HTML DATA

HtmlRAG is a new approach that improves retrieval-augmented generation (RAG) by retaining the HTML structure of retrieved web content instead of converting it to plain text.

HIGH-RESOLUTION VIDEO SEGMENTATION

LiVOS is a new lightweight video object segmentation (VOS) model that reduces memory load, allowing for the segmentation of long, high-resolution videos with less hardware demand.

ADAM ALTERNATIVE FOR DEEP LEARNING OPTIMIZATION

PUSHING THE FRONTIERS OF AUDIO GENERATION

Google has advanced its speech generation technology, enabling more natural and dynamic multi-speaker dialogues in digital assistants and AI tools. Innovative models like SoundStream and AudioLM now support applications such as NotebookLM and Illuminate, enhancing accessibility through AI-generated content. The technology achieves high-quality audio with fast inference times and safeguards against misuse via watermarking.

OCTOVERSE: AI LEADS PYTHON TO TOP LANGUAGE AS THE NUMBER OF GLOBAL DEVELOPERS SURGES

There has been a rapid increase in AI project engagement driven by the rise in data science and machine learning activities. Python has surpassed JavaScript as the most popular language. The developer community is expanding globally, with significant growth in Africa, Latin America, and Asia fueled by tools like GitHub Copilot. There's a notable trend where smaller AI models are being developed for efficient use. Generative AI projects have nearly doubled worldwide.

HOW TO CREATE SOFTWARE DIAGRAMS WITH CHATGPT AND CLAUDE

This article explores how developers can use ChatGPT and Claude to create software architecture diagrams. It highlights the iterative process of improving diagrams using multimodal AI and tools like Mermaid and Whimsical. The author demonstrates the benefits of LLMs for diagramming by showing how they can work with images and provide real-time feedback.

building with Writer AI Studio