Search the archive

DEEPSEEK V3-0324 (HUGGING FACE HUB)

DeepSeek has uploaded new model weights to its Hugging Face repository.

OPENAI TO START TESTING CHATGPT CONNECTORS FOR GOOGLE DRIVE AND SLACK

OpenAI is beta testing ChatGPT Connectors, enabling business users to link apps like Slack and Google Drive to ChatGPT for informed responses.

YC-BACKED REACTWISE IS APPLYING AI TO SPEED UP DRUG MANUFACTURING

ReactWise, a Y Combinator-backed startup, uses AI to accelerate chemical manufacturing, aiming to improve the process by leveraging pre-trained models to reduce trial-and-error.

Built on Licensed Data, Not Scraped (a.k.a stolen) Data.

* Visual AI Cookbooks (GitHub projects) For Visual Generative AI Model Customization

* Explore All Bria APIs for Content Creation

GET YOUR FREE API TOKEN TO PLAY WITH BRIA TODAY, WITH 1000 CREDITS ON YOUR NEW ACCOUNT.

OPENAI AND MIT EXPLORING AI AND EMOTIONAL WELL-BEING

MIT Media Lab and OpenAI are exploring how users engage emotionally with ChatGPT, examining its impact on social and emotional well-being. The study highlights the diverse ways people use AI conversational models and sets the stage for further research on responsible AI interactions.

HUGGING FACE REAL-TIME ENDPOINT ANALYTICS

Hugging Face has revamped its analytics dashboard, which now features real-time updates for monitoring AI inference endpoints. The improved system ensures faster data loading and instant insights into request latency, error rates, and performance metrics.

PERPLEXITY'S PLAN FOR REBUILDING TIKTOK IN AMERICA

Perplexity AI envisions a TikTok that prioritizes deep content discovery and truth-seeking powered by an advanced answer engine. It plans to enhance the platform's utility while maintaining its core function as a hub for creative expression.

SYNTHETIC CITIES

SynCity is a novel system that generates detailed 3D worlds from text prompts without any training. It combines a 2D image generator for artistic quality and a 3D generator for accurate geometry, building scenes tile-by-tile and seamlessly blending them into a navigable environment.

NEMOTRON-H

Nvidia has released a new set of hybrid Mamba style models trained on 20T tokens. They achieve strong performance, especially on long context tasks where they have extreme sequence scaling benefits. This represents the first successful fp8 large scale training of a hybrid attention model.

LARGE-SCALE OUTDOOR SCENE GENERATION

NuiScene introduces a new method for generating expansive outdoor environments that utilizes uniform vector-set encoding and explicit outpainting for better compression and faster generation.

TOKENIZE AN IMAGE AS A SET

"Tokenize Image as a Set" is a new framework for image generation that uses set-based tokenization and a novel discrete diffusion method. The approach represents images as unordered token sets, enabling a unique and invertible generative modeling process.

LATENT MULTI-HOP REASONING

This research investigates whether Large Language Models (LLMs) perform multi-hop reasoning – a process of connecting multiple pieces of knowledge – when presented with complex prompts. The study found evidence of this latent reasoning process, particularly in the initial step, but utilization of the connected knowledge varied and showed a scaling trend with model size, suggesting both challenges and opportunities for future LLM development.

LARGE ANIMATABLE HUMAN RECONSTRUCTION MODEL FOR SINGLE IMAGE TO 3D IN SECONDS

Code release for a method that takes a single image as input and creates a fully animatable 3D avatar. It is permissively licensed and seems to work quite well.

OPENAI EXEC LEAVES TO FOUND MATERIALS SCIENCE STARTUP

OpenAI's VP Liam Fedus is leaving to start a materials science AI company with backing from OpenAI. His new firm will compete with Google DeepMind and Microsoft in the AI materials science sector. Some experts doubt current AI can achieve groundbreaking scientific discoveries.

PEOPLE ARE USING GOOGLE'S NEW AI MODEL TO REMOVE WATERMARKS FROM IMAGES

Google's new Gemini 2.0 Flash AI model can remove watermarks from images, raising copyright concerns. The tool, available in Google's AI Studio, is experimental and not suited for production use, yet shows advanced capabilities in watermark removal. Google states that using its generative AI tools for copyright infringement violates its terms of service.

APPLE'S AI STRATEGY PLAGUED BY DELAYS, SIRI UPGRADE REMAINS IN LIMBO

Apple is facing delays and challenges with its AI features for Siri, initially planned for iOS 18.5 but now possibly pushed to 2026. Despite showcasing features at WWDC, internal assessments reveal the technology isn't ready for public release yet. Apple remains optimistic in enhancing Siri, aiming for significant improvements by 2027 with a new infrastructure to stay competitive.

IMAGE TO SVG: STARVECTOR (HUGGING FACE HUB)

A set of models that can turn icons and basic images into SVGs.

EFFICIENT REMOTE SENSING MODEL

DynamicVis is a foundation model for dynamic visual perception in remote sensing that offers efficient parsing of ultra-large images with drastically reduced memory and computation requirements.

GOOGLE'S IMPROVING GMAIL'S SEARCH WITH AI

Gmail's updated search now uses AI to prioritize your most-clicked emails and frequent contacts for better results.

👉 Check out all the recipes here (free, ungated).

READY TO START COOKING? Build visual GenAI with Bria's fully licensed datasets, source-available models, and flexible developer tools. Start

with 1K free API calls.

LOOKING FOR INSPIRATION? Join 1000's Visual GenAI devs on the Bria discord [6].

CLAUDE WEB SEARCH

Anthropic has enabled a new feature that allows Claude to search the web as part of its chat offering.

OPENAI NEXT-GEN AUDIO MODELS

OpenAI has introduced new speech-to-text and text-to-speech models in its API, improving transcription accuracy and enabling customizable voice outputs for more natural and expressive speech agents.

HUGGING FACE OPEN R1 GUIDE

A guide to setting up and using the OlympicCoder 7B model locally in VS Code with LM Studio for optimal coding assistance, including quantized inference for efficiency.

DEEPMESH: AUTO-REGRESSIVE ARTIST-MESH CREATION WITH REINFORCEMENT LEARNING

Autoregressive creation of meshes from point clouds that are artist friendly and well topologized. It uses deep reinforcement learning to improve the quality of the meshes with some clever regularization rewards.

END-TO-END 3D SEGMENTATION

A new object-aware lifting approach for 3D segmentation using Gaussian representations and a learnable object-level codebook.

COMPRESSION FOR 3D GAUSSIAN SPLATTING

HAC++ achieves over 100X size reduction compared to 3D Gaussian Splatting (3DGS) while enhancing fidelity, using advanced quantization and adaptive masking for superior compression in view synthesis tasks.

ROBLOX FOUNDATION MODEL

Roblox has released a novel 3D generation and understanding foundation model that can be used for research applications.

EFFICIENT VIDEO TRAINING WITH TOKEN OPTIMIZATION

FluxViT introduces a novel augmentation tool, Flux, to optimize video token selection, improving accuracy while significantly reducing computational cost in large-scale video pre-training.

STATE SPACE MODELS FOR 3D OBJECT DETECTION

DEST3D applies an interactive state-space model for 3D object detection in point clouds, leveraging novel bidirectional scanning and inter-state attention mechanisms for efficient scene understanding.

SYNCHRON'S BRAIN-COMPUTER INTERFACE NOW HAS NVIDIA'S AI

Neurotech company Synchron's Nvidia-powered brain-computer interface enables individuals with paralysis to control their home environment using the Apple Vision Pro headset. Synchron's BCI implants translate brain activity into commands, with improved speed and accuracy through Nvidia's Holoscan platform. By developing a cognitive AI based on brain data, Synchron aims for a scalable BCI system that offers more autonomous control for users with minimal training.

WHAT DOES OPENAI REALLY WANT FROM TRUMP?

OpenAI's proposal to the White House seeks to bypass state AI laws like California's SB 1047 by leveraging federal trade laws against China and requesting significant AI infrastructure investments. With nearly 900 state AI bills under consideration, OpenAI argues that diverse state regulations could cause chaos, while a lack of federal framework leaves the industry vulnerable. Critics point out the absence of a national governance model in OpenAI's plan, which relies heavily on national security concerns to advance its agenda.

ANTHROPIC SUPPORTS AI TRANSPARENCY IN POLICY DISCUSSIONS

Anthropic has responded to Governor Newsom's AI Working Group report, advocating for increased transparency in AI development and security practices. The company highlighted the need for standardized disclosure of safety measures to build public trust without stifling innovation.

DAPR'S MICROSERVICES RUNTIME NOW SUPPORTS AI AGENTS

Microsoft's Dapr Agents are optimized for building scalable AI agents with efficient orchestration and statefulness.

AI SLOP IS A BRUTE FORCE ATTACK ON THE ALGORITHMS THAT CONTROL REALITY

AI-generated content is flooding social media with spam-like brute force tactics, targeting algorithms rather than human engagement.

CHINA WILL ENFORCE CLEAR FLAGGING OF ALL AI GENERATED CONTENT STARTING FROM SEPTEMBER

China will mandate explicit labeling of all AI-generated content by September, with the rules covering all data types including text, images, and videos.

ANTHROPIC'S RED TEAM WARNS OF AI SECURITY RISKS

AI models are rapidly advancing in cybersecurity and biology, approaching expert-level knowledge in some areas. While current risks remain manageable, the company warns that stronger safeguards will be needed as capabilities progress.

GOOGLE'S AI HEALTH INNOVATIONS ANNOUNCED AT THE CHECK UP 2025

Google highlighted how AI is driving health advancements globally, from diagnostic tools to research collaborations, at The Check Up 2025. The company aims to enhance medical accessibility and improve patient outcomes through AI-driven innovations.

INSIDE ZOOM'S AI EVOLUTION: FROM BASIC MEETING TOOLS TO AGENTIC PRODUCTIVITY PLATFORM POWERED BY LLMS AND SLMS

Zoom is expanding beyond video conferencing by developing an agentic AI infrastructure that transforms meetings into action-oriented workflows. This includes AI Companion 2.0, with features like task management and document creation, and customized AI agents via the new AI Studio. Zoom's federated approach combines the company's custom small language model with larger LLMs, offering enterprises efficient, customizable AI solutions.

PLUG-AND-PLAY EXTERNAL KNOWLEDGE FOR LLMS

A novel approach to integrate knowledge into an LLM without retraining, enabling online and realtime learning.

THE KOLMOGOROV TEST: COMPRESSION BY CODE GENERATION

Meta has introduced a new benchmark for testing a language model's reasoning and knowledge. The test presents the language model with a long sequence of data and requires the model to output the shortest program that regenerates the sequences and halts. This is known as Kolmogorov compression and is uncomputable in polynomial time. The test requires strong reasoning abilities.

MEASURING AI ABILITY TO COMPLETE LONG TASKS

A new “mores law” for agents' task horizon ability shows that the length of tasks an agent can complete has been doubling every 7 months. This suggests that in the next two years, agents will be able to complete multi-hour-long tasks with many complicated steps.

STABILITY'S 3D VIDEO GENERATION

Stable Virtual Camera is a new multi-view diffusion model that generates immersive 3D videos from single or multiple 2D images, following user-defined or pre-set camera trajectories.

AAPM 2025 CHALLENGE

This repository provides code and tutorials designed to help participants get started with developing dose prediction models for the GDP-HMM Challenge at AAPM 2025.

HUGGING FACE'S R1 REPLICATION

The Open-R1 effort is designed to be slightly more robust and feature complete, but still minimal and hackable. It does more SFT steps and data distillation.