Search the archive

📍This path starts with the fundamentals of machine learning

using Sklearn and then progresses to advanced concepts [4] in deep learning using PyTorch and Tensorflow.

HUGGING FACE IS SHARING $10 MILLION WORTH OF COMPUTE

Hugging Face is committing $10 million in free shared GPUs to help developers, academics, and startups create new AI technologies, aiming to counteract the centralization of AI advancements dominated by tech giants.

OPENAI PARTNERS WITH REDDIT TO INTEGRATE UNIQUE USER-GENERATED CONTENT INTO CHATGPT

Reddit and OpenAI have announced a strategic partnership that aims to revolutionize user interactions with online communities through AI-powered features. OpenAI will gain access to Reddit's Data API, enhancing ChatGPT with real-time, relevant information from Reddit's vast content.

META IS REPORTEDLY WORKING ON CAMERA-EQUIPPED AI EARPHONES

Meta is reportedly working on AI-powered earphones equipped with cameras. Internally codenamed 'Camerabuds', the earphones will leverage AI capabilities for real-time object identification and foreign language translation. Meta's leadership sees AI-powered earphones as the next logical step in the evolution of wearable technology. It has partnered with Kansas-based electronics company Ear Micro to explore the possibilities of this emerging technology.

CURSOR'S INSTANT FULL FILE EDITS WITH SPECULATIVE EDITING

Researchers used a custom Llama 3 70B model with a speculative prior to edit files at 1,000 tokens per second, giving them near instant rewrites. They did this without diffs and with some clever output formatting.

IMPROVEMENTS TO DATA ANALYSIS

ChatGPT has launched a number of quality improvements to GPT-4o's ability to read and analyze data. It now supports the ability to view tables, view updates, visualize the changes, and perform general data analysis more quickly than was previously possible.

POSE ESTIMATION WITH DENSE CORRESPONDENCE

A new approach to 6DoF pose estimation that uses single RGB-D images has been developed. It focuses on dense correspondence instead of traditional keypoint-based methods.

Generation (RAG) system + benchmarking and evaluation techniques

for mitigating these failure modes. Join the live webinar on May 29.

LLAMA3.NP

A pure NumPy implementation for Llama 3 models.

BLIND IMAGE SUPER-RESOLUTION

CDFormer is a transformative approach to blind image super-resolution that integrates content and degradation understanding through a novel diffusion-based module.

A LIGHTWEIGHT VISION-LANGUAGE MODEL

Xmodel-VLM is a vision language model optimized for consumer GPU servers. Addressing high service costs that limit large-scale multimodal systems' adoption, this 1B-scale model uses the LLaVA paradigm for modal alignment.

CHATGPT-4O VS MATH

This article tests GPT-4o's ability to solve a math problem involving a roll of tape, finding that text-only prompts with zero-shot Chain-of-Thought prompt engineering yield the most consistent and accurate results. Despite its capabilities, GPT-4o still struggles with understanding finer details in images.

ASK HN: DISILLUSIONED AFTER AI? (HACKER NEWS THREAD)

In this thread, developers discuss how they feel about AI technology and its impact on their careers. Advances in AI may leave some developers feeling disillusioned. Some developers feel that whatever they build will just get gobbled away by some big tech company. AI is de-democratizing tech in a big way - a handful of giga-scale companies hold all the cards. The field is still very much in its hype phase, so while the technology is impressive, there is still a lot it cannot do.

OPENAI RULES THE CHANGES BUT META CHANGES THE RULES

Meta's release of the Llama 3 AI models, along with a 405B-parameter version still in training, is transforming the AI landscape by making advanced models publicly accessible, challenging OpenAI's business model based on private, paid access. This strategic move by Meta leverages open-source innovation to potentially redefine industry standards and enhance its position in a market dominated by monetized AI services.

STABLE ARTISAN (PRODUCT)

A multimodal generative AI Discord bot that utilizes the products on the Stability AI Platform API within the Discord ecosystem.

PERSONALIZED IMAGE GENERATION

MasterWeaver is a new method to improve personalized text-to-image generation models.

3D SCENE COMPLETION FOR AUTONOMOUS VEHICLES

Researchers have developed a new technique to fill in the gaps in 3D LiDAR scans, allowing autonomous vehicles to better understand their surroundings.

* Achieving GPT-4 quality at 60x lower costs

Read the blog [4] to dive into the details — then sign up to bring your fine-tuned model to OctoAI today.

OPENAI CHIEF SCIENTIST ILYA SUTSKEVER IS OFFICIALLY LEAVING

Ilya Sutskever, OpenAI's co-founder and chief scientist, is officially leaving the company. Sutskever helped lead the coup against Sam Altman and then later changed his mind. His employment status had been ambiguous since the ouster. Jakub Pachocki, the company's director of research, will be OpenAI's new chief scientist.

PROJECT IDX, GOOGLE'S NEXT-GEN IDE, IS NOW IN OPEN BETA

Project IDX is an AI-centric, browser-based development environment that was launched as an invite-only service in August. Over 100,000 developers have already tried the service, which is now available in open beta. Google has added an integration with the Google Maps Platform into the IDE and integrations with the Chrome Dev Tools and Lighthouse to help debug applications. It will soon enable deploying apps to Cloud Run, Google Cloud's serverless platform for running front- and back-end services, and integrate with Checks, Google's AI-powered compliance platform.

COIN3D

Generating 3D models is challenging. Allowing users to input a coarse mesh first and then run generation allows for finer-grained control and improved output quality from the model.

A NEW BENCHMARK IN TEXT-TO-IMAGE GENERATION FOR ENGLISH AND CHINESE

The newly developed Hunyuan-DiT sets a benchmark in text-to-image diffusion transformers, especially for English and Chinese. It features advanced transformer structures and a refined data pipeline for continual model improvement.

IMPROVING DIFFUSION MODELS WITH PERTURBED-ATTENTION

Perturbed-Attention Guidance (PAG) is an approach to enhancing the quality of images generated by diffusion models without additional training or external modules. By creatively manipulating self-attention mechanisms within the model, PAG significantly improves the structure and fidelity of both unconditional and conditional samples.

CRM data etc.) for your multi-tenant RAG pipeline.

Paragon is an embedded data ingestion engine [10] that engineering teams rely on to extract users' data from 100+ sources, without the plumbing.

IC-RELIGHT

AI Relighting tool that uses foreground separation and text conditioning for extremely strong results.

EFFICIENT MOBILE VIDEO UNDERSTANDING

SqueezeTime is a new lightweight network designed for mobile video understanding that condenses the time axis of videos into the channel dimension to enhance temporal analysis.

SPEEDING UP FOUNDATION MODEL TRAINING

EfficientTrain++ introduces a novel curriculum learning strategy that significantly reduces the training time of major visual models like ResNet and Swin by up to three times on ImageNet databases.

NOUS HERMES FIRST MODEL MERGE (HUGGING FACE HUB)

The new Theta model is a merge between Hermes 2 Pro and Llama 3 8B instruct. It has excellent function calling capabilities and strong performance on a variety of benchmarks.

AI SPENDING GREW 293% LAST YEAR

Ramp's Q1 analysis reveals a 293% year-over-year increase in AI-related spending among its customers, outpacing overall software spend growth, with a significant adoption in non-tech sectors like healthcare and financial services, indicating a broader integration of AI into various industries. Despite a general slowdown in new investments in AI, companies already using AI are doubling down, with average spending on AI tools growing by 138% year-over-year, while finance tools also see increased usage and businesses maintain cautious approaches to travel expenditure.

DETECTING OUT-OF-DISTRIBUTION DATA

Hopfield Boosting is a technique that utilizes modern Hopfield energy to enhance the detection of out-of-distribution (OOD) data in machine learning models.

greatest real-time data. Watch it for free

BRAINTRUST AI RECRUITER (PRODUCT)

Braintrust's AI Recruiter can do things like generate job descriptions, match candidates, async interviews, and grading.

AI COPILOTS ARE CHANGING HOW CODING IS TAUGHT

Educators are shifting away from syntax and emphasizing higher-level skills.

OPENAI'S CUSTOM GPT STORE IS NOW OPEN TO ALL FOR FREE

The ability to browse and use OpenAI's GPT store is now available to all users.

BOOK A FREE CONSULTATION SESSION TODAY. GET IN TOUCH NOW

GOOGLE I/O

Google announced many new features, including Gemini Flash, Veo video generation, Imagen 3, and its newest assistant, Project Astra, at I/O 2024. In all, there are an impressive number of improvements, including 2m token context length, dramatically cheaper models, and improved multimodality.

ANTHROPIC IS EXPANDING TO EUROPE AND RAISING MORE MONEY

Anthropic has expanded its AI assistant, Claude, to Europe. Claude supports multiple languages. Anthropic is offering the service across its website, iOS app, and business plans for teams. The company is beginning the process of raising more money.

MAMBA'S SUITABILITY FOR VISION TASKS

Researchers investigated the Mamba architecture, typically used for tasks with long-sequence and autoregressive characteristics, and its application in vision tasks, and found that while Mamba is not effective for image classification, it shows promise in detection and segmentation tasks that do.

A NEW STATE-FREE SEQUENCE PARALLEL INFERENCE

A new state-space model using a dual transfer function representation has been developed for deep learning. It features a state-free sequence parallel inference algorithm.

API key ↗️

OLLAMA ON GOOGLE FIREBASE

Genkit is a new toolset for Firebase for building and deploying generative products. It can be used to launch servers for open source language models.

IMAGE DERAINING

ESDNet is a Spiking Neural Network (SNN) designed for image deraining tasks. It capitalizes on the unique properties of rain pixel values to enhance spike signal intensity.

WHAT OPENAI DID

GPT-4o multimodal abilities, integrating vision and voice, promise significant advances in how AI interacts with the world, paving the way for AI to become a more ubiquitous presence in daily life.