Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Sponsored
The Predibase team recently launched LoRA Land, a collection of 25+ fine-tuned Mistral-7b models that rival GPT4 [11] on a set of task-specific applications. In this hands-on workshop, you'll learn how to build your own [11] LoRA Land with popular open-source LLM frameworks, Ludwig and LoRAX. Become a fine-tuning pro. Join the live
Stanford has released a neat research system called Storm that uses retrieval guided language models to create reports for specific topics.
Accurately segmenting road lines and markings is crucial for autonomous driving but challenging due to occlusions caused by vehicles, shadows, and glare. The Homography Guided Fusion (HomoFusion) module uses video frames to identify and classify obscured road lines by leveraging a novel surface normal estimator and a pixel-to-pixel attention mechanism.
Vision Language Models (vLLMs) often struggle with processing multiple queries per image and identifying when objects are absent. This study introduces a new query format to tackle these issues, and incorporates semantic segmentation into the training process.
Anyscale's latest release of Ray, Ray 2.10, adds support for Intel Gaudi 3. Developers can now spin up and manage their own Ray Clusters, provisioning Ray Core Task and Actors on a Gaudi fleet directly through Ray Core APIs, tap into Ray Serve on Gaudi through Ray Serve APIs for a higher level experience, and configure Intel Gaudi accelerator infrastructure for use at the Ray Train layer.
Code Qwen 1.5 is a new set of 7B models trained on 3T tokens of code related data. It performs well on HumanEval, with a non-zero score on SWE-bench. The chat variant specifically shows promise for long context retrieval tasks up to 64k tokens.
Extreme low-bit quantization for small pre-trained models, like Llama2-7B, is challenging, but fine-tuning just 0.65% of parameters significantly improves performance. Newly fine-tuned 1-bit models outperform 2-bit Quip# models, while 2-bit models with specialized data can exceed full-precision counterparts. This research suggests that proper fine-tuning and quantization may enhance efficiency without compromising model quality, potentially shifting focus from training smaller models to optimizing larger, quantized ones.
Sponsored
Apache Airflow is playing a key role in the AI/ML initiatives of modern enterprises. Find out how to start delivering production-ready AI with Airflow in this free ebook by Astronomer [18].
Google DeepMind's ALOHA Unleashed is a program that pushes the boundaries of dexterity with low-cost robots and AI.
The pressure is immense right now to have a plan to factor AI into existing products. This short step-by-step guide will help you take the first step.
Personalized AI app and wearable powered by what you've seen, said, and heard.
Sponsored
Most of the successful applications of genAI are currently in B2B. But to win enterprise-sized contracts, AI startups need to meet stringent security and compliance requirements. Against this backdrop, WorkOS has seen tremendous growth [3] with AI companies using its enterprise-readiness platform. This includes heavy-hitters like Jasper, Copy.ai, AI21 Labs, and Perplexity.ai.
OpenAI has opened an office in Japan. ChatGPT Enterprise is being adopted in Japan by businesses such as Daikin, Rakuten, and TOYOTA Connected. The Japanese government, focusing on AI policies for societal benefit, has collaborated to share ChatGPT best practices.
Adobe is developing an AI model to generate video, which will be integrated into Premiere Pro later this year. It will feature object addition, removal, and generative extend capabilities. Adobe is collaborating with third-party vendors and addressing deepfake concerns through Content Credentials.
OpenAI and Meta are teasing the next iterations of their AI models, expected to feature enhanced reasoning and planning capabilities. Dubbed GPT-5 and Llama 3, the models aim to advance toward artificial general intelligence, with vague release timelines and application details. The tech community remains skeptical given the history of overhyped AI promises with limited substantive evidence.
T5 is a workhorse of modern AI. Eleuther has retrained it with a modern tokenizer and for much longer. The result is a greatly improved base model for encoding tasks.
This post explores how to train diffusion models to generate video, how to adapt image models, and even how to generate video from an image model without additional training.
Only 4 days left to grab early-bird tickets [10] to the AI Quality Conference [10] happening this June in San Francisco! Don't miss your chance to join experts from Cruise, NVIDIA, Google, UBER, and more — real practitioners sharing real stories about building rigorous, reliable, and scalable AI. Learn and network where it counts! >> Save $260 when you register by April 20 using _KolenaVIP2024_ -
Addressing the challenges in monocular egocentric 3D human motion capture, especially under difficult lighting and fast movements, researchers introduce a novel method using a fisheye event camera.
Researchers have developed a new approach called Mixture of Prefix Prompt Experts (MPPE) to enhance zero-shot dialogue state tracking, which transfers knowledge to unseen domains without needing new dataset annotations.
Any2Point is a new method that efficiently translates capabilities from vision, language, and audio models into the 3D domain without losing spatial geometries.
Humane's Ai Pin, an AI-powered wearable, has many technical issues, from music streaming to AI assistant malfunctions. Despite its ambition to create an ambient computing experience, the first-generation device suffers from performance lags and a lack of essential features, though future software updates are promised. The Ai Pin is designed with attention to detail but struggles to replace traditional smartphones, positioning itself as a companion device for a more present and less screen-focused lifestyle.
AI differentiation is challenging, but the key lies not in AI models like LLMs, which are becoming commoditized, but in the unique data fed into these models. Effective data engineering is crucial as it directly impacts AI performance, with applications requiring integration of customer-specific data to provide accurate responses. Thus, creating a competitive edge in AI applications hinges on innovative data use rather than the AI technology itself.
Sponsored
Implementing GenAI is easier said than done. Download TrainAI's free roadmap [16] to learn key decisions to make upfront to ensure project success.
A quick utility that dumps GitHub repositories into files that can be used with long context models that cannot browse the web.
Internal tests have shown these AI avatars are not yet as effective as human influencers in driving sales.
Google's Gemini Code Assist, an AI code completion tool for enterprises, features a one million-token context window for more accurate code suggestions and supports on-premises codebases across multiple platforms.
Sponsored
When it comes to AI, you have to move fast. But before you launch, you're going to need NIST AI RMF [3]… and maybe also SOC 2, ISO
xAI has announced that its latest flagship model has vision capabilities on par with (and in some cases exceeding) state-of-the-art models.
Google's new AI chip, Cloud TPU v5p, is now available. It boasts nearly triple the training speed for large language models compared to its predecessor, TPU v4. This release underscores Google's position in the AI hardware race alongside competitors like Nvidia. Google has also introduced the Google Axion CPU, based on Arm's chip infrastructure, promising better performance and energy efficiency.
OpenAI has reportedly fired two researchers who were allegedly linked to the leaking of company secrets following months of leaks and company efforts to crack down on such incidents.
DGMamba is a new framework designed to handle domain generalization challenges using the innovative state space model Mamba.
MindBridge is a singular model capable of decoding brain signals from multiple subjects.
Sponsored
OctoAI just launched OctoStack, the GenAI serving stack that lets you run AI models in your own environment — ensuring data privacy and efficient utilization. Join OctoAI's live event [11] on April 17 to learn how you can run super-efficient inference, customize model usage, and deliver reliable GenAI at scale.
Jina AI reader converts URLS into LLM-friendly markdown to be used as input for a variety of tasks.
Clean implementation of discrete diffusion. This code includes many state-of-the-art pieces and trains quickly and stably.
This project introduces a dual-branch diffusion model, PanFusion, that crafts 360-degree panoramic images directly from text prompts. The method merges the Stable Diffusion approach with a specialized panorama branch, enhanced by a unique cross-attention mechanism to reduce image distortion.
BabyLM is a challenge to see who can train the best text and vision models while only using as much data as a typical human baby sees (~10M tokens).
Meta AI research's embodiment hypothesis suggests that true intelligence requires a physical form for sensory and environmental interaction. AI Habitat 3.0, Meta's updated simulation platform, aims to bridge the sim-to-real gap by training AIs with virtual bodies in simulated environments. It now has human avatars. Real-world testing is underway, with companies like Agility Robotics and Apptronik deploying robots alongside humans in various research and industrial settings.
Sponsored
This demo uses Amazon Bedrock to power the chatbot (in this case using Claude 2 by Anthropic) and Confluent Cloud to make real-time
A small and powerful visual language model trained on LAION and LLaVA data.
AI currently falls into the micromanage category, where motivation is high but the skill level is relatively low, requiring users to define tasks, review work frequently, and guide progress at each step, similar to managing high-school interns.
Dr. Andrew Ng is currently the Managing General Partner of AI Fund and is joining Amazon's Board of Directors.