Search the archive

INSANELY FAST WHISPER

Transcribe 300 minutes of audio in less than 10 minutes with OpenAI’s Whisper Large v2.

APPLE AND AI

Apple is lagging in AI advancements, especially when compared to competitors like Google and Microsoft, and there are doubts among Apple insiders that its AI/ML team can deliver.

QUALCOMM’S NEXT BIG SNAPDRAGON CHIP IS FULL OF AI FEATURES

Qualcomm's upcoming Snapdragon 8 Gen 3 chip, set to debut in Samsung’s Galaxy S24, emphasizes advanced AI camera tools and enhanced processing speeds.

AI GODFATHERS PROPOSE FRAMEWORK TO MITIGATE AI RISKS

AI pioneers Yoshua Bengio and Geoffrey Hinton, along with 22 other AI experts, have proposed a policy framework emphasizing AI safety and have recommended that companies and governments allocate a third of their AI R&D budget to safety efforts.

LABELBOX IS HIRING A FOUNDATION MODELS LEAD ENGINEER (SAN FRANCISCO, $170K-$215K + EQUITY)

Labelbox is a platform for building data-centric AI applications. As the Foundation Models Lead, you will lead research and development for applying foundation models to real world problems for customers like Walmart and Adobe. Learn more [17].

THE HALF-LIFE OF THE AI STACK

The infrastructure layer in AI is rapidly changing due to accelerated hardware development, fierce optimization competition, fast-paced research, long feedback cycles, ambiguity in other layers, and the commoditization of infrastructure components.

MAGNETIC

Easily integrate Large Language Models into your Python code.

FLOWRL (PRODUCT)

Realtime UI personalization, powered by AI.

GOOGLE INVESTS $2B INTO ANTHROPIC

Google has committed to a substantial investment in Anthropic, a competitor to OpenAI, earmarking up to $2 billion for the AI startup.

FIRST DIRECTOR OF THE FRONTIER MODELS FORUM NAMED AND BIG TECH ADDS $10M TO SAFETY INITIATIVE FUNDING

Chris Meserole is the first director of the Frontier Models Forum. Many large tech companies have added funding to the AI safety fund.

AMAZON ADS LAUNCHES AI-POWERED IMAGERY FOR ADVERTISERS

Amazon Ads has introduced an AI-powered image generation feature in beta that creates lifestyle and brand-themed images to improve ad performances. The tool, which requires no technical expertise, enables brands to create more engaging ads, reportedly increasing click-through rates by 40%. It will be rolled out progressively to advertisers based on customer feedback.

EMULATE TRAINING A LARGE MODEL JUST BY TRAINING A SMALL ONE

Emulator Fine Tuning (EFT) comes from the direct preference optimization group and explores what happens if you fine tune a small model and then project that fine tuning onto a large model. This counterintuitive strategy works shockingly well.

IDENTIFYING OUT-OF-DISTRIBUTION DATA

ML models can sometimes wrongly recognize unfamiliar data, thinking they've seen it before. A new method, NNGuide, helps models make better decisions, especially when identifying unknown data.

TOWARDS MONOSEMANTICITY: DECOMPOSING LANGUAGE MODELS WITH DICTIONARY LEARNING

Researchers use sparse autoencoders to tackle neural networks' polysemanticity, extracting clearer features for improved model understanding. These features influence outputs and exhibit complex interactions.

AUDIOFLARE

An all-in-one AI audio playground using Cloudflare AI Workers to transcribe, analyze, summarize, and translate any audio file.

A SYSTEM FOR EVALUATING LARGE LANGUAGE MODELS

This study presents "JudgeLM", a new method to efficiently evaluate Large Language Models in versatile situations. The authors developed a vast dataset and benchmarking system.

MULTI-MODAL PROMPT INJECTION IMAGE ATTACKS AGAINST GPT-4V

The new GPT-4V allows for interactive conversations with image inputs but also introduces new vulnerabilities through visual prompt injection attacks, enabling exfiltration of data and execution of hidden instructions encoded in images.

DEEP LEARNING IN RUST (BOOK)

Rust is a popular systems programming language with strong memory safety. Burn is a framework built on these principles for efficient ML. Importantly, it offers compatibility with the Onyx runtime.

THANKS TO AI, THE FUTURE OF PROGRAMMING MAY INVOLVE YELLING IN ALL CAPS

OpenAI's prompts reveal many quirks about prompt engineering. For example, the prompts use 'please' a lot and important instructions are often all capitalized. Capitalizing instructions likely works as the models were trained on huge numbers of examples where responses clearly paid more attention to capitalized sentences. This could result in a future where people shout at their computers to get them to work better.

IS AI ALIGNMENT ON TRACK? IS IT PROGRESSING... TOO FAST?

AI alignment is a complex issue that lacks concrete benchmarks or numbers for measurement. Predictions of AI-induced doom rely on narratives rather than rigorous models. The absence of hard numbers makes it challenging to assess the state of AI alignment.

MISTRAL POWERED MULTIMODAL LLAVA MODEL (HUGGING FACE HUB)

The Llava fine tuning paradigm combines powerful language models with vision capabilities. It has been applied with Mistral 7B for improved performance.

RAG DEMYSTIFIED

An LLM-powered advanced RAG pipeline.

SPOKE (PRODUCT)

A priority inbox for product teams that reduces information overload, prioritizes work, gives instant context, and augments core workflows using AI.

OPENAI CREATES PREPAREDNESS TEAM

AI EXECUTIVE ORDER EXPECTED MONDAY

LENOVO AND NVIDIA EXPAND PARTNERSHIP FOR HYBRID AI SOLUTIONS

IMAGE DIFFUSION MODEL TRAINED ON OPEN DATA

Most popular diffusion models like Stable Diffusion have been trained on a wide variety of data, but much of that data is subject to different copyright restrictions. This new model from MosaicML is trained purely on Creative Commons data and achieves similar performance to SD 2 while requiring only a fraction of the training data.

4-BIT QUANTIZATION IN LLMS

This research introduces LLM-FP4, a novel method to compress large language models by converting their weights and actions to 4-bit floating-point values after training.

EVALUATING GPT-4V'S OPTICAL CHARACTER RECOGNITION ABILITIES

The paper examines how well GPT-4V, a Large Multimodal Model, performs in various Optical Character Recognition (OCR) tasks, from reading scenes and handwritten texts to understanding complex document structures.

REAL-TIME DEPTH ESTIMATION FOR UNDERWATER VEHICLES

This study presents a model tailored for underwater vehicles that can estimate real-time depth from single images by combining sparse depth details.

GPT PILOT

GPT Pilot is an AI-powered dev tool that writes scalable apps from scratch. It enables developers to create apps using a description of the kind of app they want to build. GPT Pilot creates apps step-by-step while asking questions for clarification or help when it gets stuck. A video demo is available in the repository.

TORA

ToRA is a series of Tool-integrated Reasoning Agents designed to solve challenging mathematical reasoning problems by interacting with tools.

THE CONVERGENCE OF DATA AND SOFTWARE ENGINEERING IN THE AGE OF AI

AI needs data to work, meaning data functions are getting closer to the core of the product-building process and moving away from internal business intelligence.

A VERSATILE APPROACH FOR PIXEL MATCHING IN IMAGES

The study introduces RGM (Robust Generalist Matching), a deep learning model designed for both sparse and dense pixel matching in images. The method leverages a unique cascaded GRU module and a new large-scale dataset.

REDDIT THREATENS TO BLOCK GOOGLE

DISTRIBUTED MAP OVER HUGGINGFACE DATASET (GITHUB GIST)

IVP

GOOGLE AI OFFERS EXPANDED INDEMNITY PROTECTIONS

DRAFTING AI CONSTITUTION THROUGH A DEMOCRATIC PROCESS

ANSWERING QUESTIONS OVER LONG, STRUCTURED DOCUMENTS

Retrieving information from structured text documents is a well-studied problem that gets much harder when the documents are PDFs. Adobe Research has introduced a tool that allows you to condition either structure or text for retrieval. It also released a benchmark dataset for further study.

VIDCHAPTERS-7M: A NEW TOOL FOR EASIER VIDEO NAVIGATION

VidChapters-7M is a big collection of videos with 7 million user-made chapter marks. This new dataset helps in the creation and testing of methods for splitting videos into chapters and naming them, making it simpler for viewers to find what they're looking for.

BOOSTING RECOMMENDER SYSTEMS WITH LLMS

This project introduced RLMRec, a framework that combines LLMs with recommender systems, capturing the deeper meanings of user behaviors and preferences and enhancing the quality of recommendations.

REACTAGENT

ReactAgent is an experimental autonomous agent that uses GPT-4 to generate and compose React components from user stories.

SPELLTEST

AI-to-AI testing for LLM-based applications.

POLYFIRE

An all-in-one managed backend for AI apps.

PEAK DATA

Peak Data explores how an internet increasingly filled with internet data might lead to model collapse, its effect on apps like ChatGPT, and possible solutions that foundation model vendors might adopt to overcome model collapse.

60% FASTER AND 50% SMALLER SDXL MODEL

A distillation of the current leading open source text to image model. The researchers claim a minimal reduction in image quality with a dramatic speed-up at inference time.

FIVE AREAS OF AI OPPORTUNITY ACCORDING TO SNOWFLAKE’S AHMAD KHAN