Search the archive

REALHF BOOSTS TRAINING EFFICIENCY

ReaLHF is a novel system that enhances reinforcement learning from human feedback (RLHF) by dynamically redistributing parameters and optimizing parallelization during training.

LET'S BUILD A STORYTELLER

Andrej Karpathy has a new repository with an outline for a new course he's working on. It involves building an aligned language model that can tell stories. The course includes video lectures, code, and other learning materials.

AUTOMATE SOFTWARE IMPROVEMENTS

AutoCodeRover is a new tool that automates software improvements, including bug fixes and feature additions, by combining large language models with advanced code search techniques.

AI CANDIDATE RUNNING FOR PARLIAMENT IN THE U.K. SAYS AI CAN HUMANIZE POLITICS

"AI Steve," created by Sussex businessman Steve Endacott and backed by Neural Voice technology, is running for office in the UK general election, promising constant connectivity with constituents via AI. Endacott aims to revolutionize democracy by allowing voters to directly influence policy on a weekly basis using technology. While some see this blend of AI and politics as radical, Endacott insists it enhances human involvement in democratic processes.

ANTHROPIC HAS A FAST NEW AI MODEL

Anthropic has launched Claude 3.5 Sonnet, boasting better performance than GPT-4o and Gemini on several benchmarks alongside increased speed and expanded capabilities. The update also introduces the Artifacts feature, enhancing user interaction with the AI's outputs. Claude aims to transition from a chatbot to a central tool in business environments, integrating knowledge and workflow management.

NLUX

NLUX is a conversational AI JavaScript library that provides a UI for large language models. It makes it super simple to integrate powerful large language models into web apps. NLUX features React components and hooks, LLM adapters, streaming LLM output, and custom renderers.

to help get you started. Get the cookbook →

CLAUDETTE FRAMEWORK

Claudette is a higher-level and easier-to-use way to interact with Claude.

THE LONG VIEW ON AI

Historical data suggests the potential for massive increases in growth rates and technological advancements due to AI, with society likely adapting to these rapid changes as it has in the past.

AIS ARE COMING FOR SOCIAL NETWORKS

Butterflies AI, a provocative new social media app where users create AI characters that autonomously generate content and interact with each other, is now live on iOS and Android after raising $4.8 million.

→ Read the full article on the Orca blog

→ See how you can simplify investigations and accelerate remediation

Security is using it to build better defenses. This blog post

shows how AI-powered security enables companies to:

with Orca's built-in generative AI

ANTHROPIC LAUNCHES CLAUDE 3.5 SONNET

Claude 3.5 Sonnet passes Opus in performance at 1/5 the cost. It is also the best vision model available now. This highlights the improved capabilities of the frontier models.

APPLE RESEARCHERS ADD 20 MORE OPEN-SOURCE MODELS TO IMPROVE TEXT AND IMAGE AI

Apple has contributed 20 Core Machine Learning models to the Hugging Face open-source AI repository, enhancing its existing range of public models with advances in image classification and depth segmentation. These contributions follow Apple's release of the Ferret large language model and the four OpenELMs to Hugging Face earlier in the year. The move demonstrates Apple's increasing engagement with the AI research community and its commitment to advancing AI capabilities.

FACTORY'S SERIES A ANNOUNCEMENT

Factory has secured a $15M Series A funding, led by Sequoia Capital, to expand its team and enhance its suite of AI-powered software development tools known as Droids. Its products have achieved a new state-of-the-art on the SWE-bench AI coding benchmark and are driving rapid growth in its customer base. Factory aims to further automate software engineering, reducing tedious tasks and improving development cycles.

OPTIMIZING INFERENCE AT CHARACTER AI

Character AI serves 20,000 queries per second, which is 20% of Google's search volume. It has a number of innovations to run this efficiently.

REMOTE SENSING CHANGE DETECTION

ChangeViT is a framework that utilizes vision transformers (ViTs) for detecting large-scale environmental changes in remote sensing images.

PRUNING WITH LAYERMERGE

LayerMerge is a new method that improves neural network efficiency by jointly pruning convolution layers and activation functions.

CAN WE TRUST VISION-LANGUAGE MODELS?

Vision-enabled language models (VLMs) like GPT-4o and Gemini power autonomous agents capable of tasks such as making purchases or editing code. This work highlights the vulnerability of these agents to malicious attacks.

EVALUATING WEB AGENTS IN REAL-TIME

WebCanvas is a new framework for evaluating autonomous web agents in dynamic, live web environments.

94% ACCURACY ON CIFAR-10 IN 3.29 SECONDS

CIFAR-10 is an image classification benchmark. This code provides a training configuration that achieves good performance in an amazingly short time.

TIME SERIES FORECASTING WITH TIMESIEVE

TimeSieve is a new model designed to tackle common challenges in time series forecasting.

COST OF SELF-HOSTING LLAMA-3 8B-INSTRUCT

Self-hosting an LLM like Llama-3 8B-Instruct can be significantly more expensive than using ChatGPT, costing around $17 per million tokens compared to ChatGPT's $1 per million tokens. The cost of self-hosted hardware can be reduced to less than $0.01 per million tokens, though it would take approximately 5.5 years to break even on the initial investment.

KHROMA (PRODUCT)

Khroma uses AI to learn which colors you like and creates limitless palettes for you to discover, search, and save.

NEW BENCHMARK FOR DEPTH NORMAL ESTIMATION

A new benchmark evaluates state-of-the-art depth and surface normal estimation models.

NVIDIA RELEASES MAMBA 2 (HUGGING FACE HUB)

The hybrid simple Mamba model recently highlighted in an Nvidia paper has been released.

ILYA SUTSKEVER IS STARTING A NEW AI COMPANY

Ilya Sutskever has launched Safe Superintelligence Inc. (SSI), a startup focused solely on developing a safe and powerful AI system free from the commercial pressures faced by companies like OpenAI.

APPLE'S INTELLIGENT STRATEGY

Apple's WWDC revelations highlight its strategic positioning in AI with a focus on privacy and security and employing in-house chips and a zero-trust architecture in its Private Cloud. Apple's AI integrates OpenAI's ChatGPT for tasks beyond its scope, with a business model where AI suppliers may financially contribute to access Apple's user base. Apple's approach to privacy could leverage upcoming regulations as a competitive advantage, aligning with its commitment to user privacy and potentially reinforcing its dominance in tech.

ROBLOX'S ROAD TO 4D GENERATIVE AI

Roblox is building toward 4D generative AI, going beyond single 3D objects to dynamic interactions. Solving the challenge of 4D will require multimodal understanding across appearance, shape, physics, and scripts. Early tools that are foundational for its 4D system are already accelerating creation on the platform.

MICROSOFT'S FLORENCE-2 STATE-OF-THE-ART VISION MODEL (HUGGING FACE HUB)

Microsoft has released an MIT-licensed set of small VLMs that dramatically outperform much larger models on captioning, bounding, and classification.

LOGIT PRISMS: DECOMPOSING TRANSFORMER OUTPUTS FOR MECHANISTIC INTERPRETABILITY

The logit lens method, which helps in understanding transformer models' decision-making, has been enhanced by breaking down logit outputs into contributions from various model components. This approach, which uses "prisms" for residual streams, attention layers, and MLP layers, reveals how these parts influence predictions and provides insights into tasks like factual retrieval and arithmetic performed by the gemma-2b model.

LLMS AS DECISION MAKERS

Decision QA is a new task for LLMs that determines the best decisions based on complex data analysis.

charge. Read more and download

ARGILLA

Argilla is a collaboration platform for AI engineers and domain experts that require high-quality outputs, full data ownership, and overall efficiency.

EFFICIENT LARGE LANGUAGE AND VISION MODELS

TroL is a new family of efficient large language and vision models (LLVMs) with sizes of 1.8B, 3.8B, and 7B parameters.

DOT

A standalone open source app meant for easy use of local LLMs and RAG in particular to interact with documents and files similarly to Nvidia's Chat with RTX.

HOW TO FIX AI'S “ORIGINAL SIN”

This article addresses the copyright challenges posed by AI models trained on copyrighted material without permission. It suggests AI developers respect copyright signals, implement guardrails to prevent generating infringing content, and develop business models that ensure fair compensation for content creators, including techniques like retrieval-augmented generation (RAG) and creating cooperative AI content ecosystems.

NVIDIA PASSES MICROSOFT IN MARKET CAP TO BECOME MOST VALUABLE PUBLIC COMPANY

Nvidia is now the most valuable public company in the world. Its market cap surpassed Microsoft's $3.32 trillion on Tuesday, reaching a high of $3.34 trillion. Nvidia's shares are up more than 170% so far this year. Its market cap hit $3 trillion for the first time earlier this month. Nvidia's rise has been so rapid the company has yet to be added to the Dow Jones Industrial Average, the stock benchmark of the 30 most valuable US companies.

TAKEAWAYS FROM OPENAI AND GOOGLE'S MAY ANNOUNCEMENTS

OpenAI and Google have introduced advanced AI models that enable real-time multimodal understanding and responses and promise improved AI assistants and innovations in voice agents. OpenAI's GPT-4o boasts double the speed and half the cost of its predecessor, while Google's Gemini 1.5 Flash delivers a significant reduction in latency and cost. Both tech giants are integrating AI across their ecosystems, with OpenAI eyeing consumer markets, which could potentially reach up to a billion users, with its products and partnerships.

config! Start now (no signup required)

AI WRAPPER IDEA LIST

This list has a set of AI projects that are reasonable for beginners to implement.

ASANA AI TEAMMATE (PRODUCT)

Asana AI empowers organizations with intelligent capabilities to enhance decision-making and maximize impact.

PARAMOUNT

Agent accuracy measurements for LLMs.

GEOFF HINTON AND MAX WELLING $30M CARBON CAPTURE START UP

Geoff Hinton and Max Welling, preeminent researchers in the field of AI, have assembled a strong team and are working to build AI systems that help with material science for carbon capture.