Search the archive

OLLAMA MULTIMODAL PRE-RELEASE (GITHUB RELEASE)

Ollama is an amazingly simple way to use modern language models. It is adding multimodal support with models like Llava.

RESPELL (PRODUCT)

Automate knowledge work using AI workflows and agents.

YOUR REFERRAL LINK - https://tldr.tech/ai?ref=2613930

performance masterworks.com/cd

META’S AUDIO FOUNDATION MODEL

Meta recently teased a foundation AI model for audio. It has released the paper along with more samples and compelling demos. The main goal of the project was to generate controllable audio content with styles all from the same model.

MICROSOFT AND LABOR UNIONS FORM ‘HISTORIC’ ALLIANCE ON AI

Microsoft is partnering with the American Federation of Labor and Congress of Industrial Organizations to discuss AI's impact on workers and to provide AI training from winter 2024. This groundbreaking alliance aims to involve workers in AI development and policy shaping and includes a neutrality agreement to ease unionization at Microsoft.

EU REACHES DEAL ON THE WORLD’S FIRST COMPREHENSIVE AI RULES

The European Union has reached a tentative agreement on the world's first comprehensive artificial intelligence legislation, known as the Artificial Intelligence Act. The deal, achieved after lengthy negotiations, addresses various aspects of AI, including generative AI and law enforcement's use of facial recognition, and sets a legal framework for AI usage, with potential global implications for AI regulation and development.

BIOCLIP

A vision model designed for biology applications. It outperforms OpenAI's clip by almost 20% on specific bio tasks. A 10m paired images and text training set is also available.

VIDEO CLASSIFICATION WITH SPATIAL-TEMPORAL DIVERSITY

Spatial-Temporal Diversification Network (STDN) is a new model that captures a variety of cues in videos by exploring both spatial features within frames and relationships across time.

ENHANCING RL ALGORITHMS WITH STATE-ADAPTIVE BALANCING

Researchers introduce FamO2O, a framework designed to boost the capabilities of current offline-to-online RL algorithms by determining the best balance for improvement and constraints based on the state.

BRICKSLLM

BricksLLM is a cloud-native AI gateway written in Go that serves as a proxy to OpenAI and lets you create API keys that have rate limits, cost limits, and TTLs.

AI AGENTS WITH LANGUAGE MODEL CAPABILITIES

KwaiAgents is an advanced agent system that uses large language models to mimic human-like cognitive skills.

NOW ADD A WALRUS: PROMPT ENGINEERING IN DALL-E 3

PYTORCH 2 INTERNALS

A great recent talk about all the new things in Pytorch 2 like Dynamo, Instructor, and ExecuTorch.

CHALLENGES IN UNSUPERVISED OBJECT SEGMENTATION

This project delves into the difficulty of segmenting objects in real-world images using unsupervised models.

HUGGINGFACE GETS AMD SUPPORT

Transformers 4.36.0 has been released with the new Mistral model, AMD support, safetensors as default, and more!

(OpenAI Gym). Save your seat here 🙋♀️

AI TAMAGO

A local LLM pet with a web front end.

30X FASTER AI TRAINING

This project has received significant attention recently for rewriting gradients for common models by hand in Triton. Its benchmarks aren't the most compelling, but it is a great endeavor.

MARKETING GPTS (PRODUCT)

Your marketing AI co-pilot.

YOUR REFERRAL LINK - https://tldr.tech/ai?ref=2613930

MISTRAL'S NEW MODEL AND NEW FUNDING

The 8-month-old AI group is set to raise $400m+ in a new round. It also released an 8x7B Mixture of Experts model which is currently being evaluated by the community.

GOOGLE UNVEILS ALPHACODE 2, POWERED BY GEMINI

Google's DeepMind has released AlphaCode 2, an improved code-generating AI model that outperforms 85% of human competitors in programming contests, showcasing advanced capabilities in solving complex mathematical and theoretical computer science problems.

STABILITY AI LAUNCHES STABLELM ZEPHYR 3B

StableLM Zephyr 3B, a new 3 billion parameter chat model, is being released as an extension of the StableLM 3B-4e1t model, drawing inspiration from the Zephyr 7B model. It is designed for efficient text generation, particularly in instruction following and Q&A contexts, and has been fine-tuned on multiple datasets using the Direct Preference Optimization algorithm.

BETTER, CHEAPER, FASTER ALIGNMENT WITH KTO

Most alignment frameworks require ranked preference data (A is preferred to B). This is usually expensive and rare. This new framework only requires an independent yes or no rating per data point.

STRIPEDHYENA 7B MODEL

Transformer models have dominated the AI landscape recently. They are powerful sequence learners but have some drawbacks. This new work, which builds on state space models, shows that new architectures can bring great benefits like increased context length.

VIDEO EDITING WITH PRE-TRAINED TEXT-TO-IMAGE DIFFUSION MODELS

RAVE is a video editing method that uses existing text-to-image diffusion models to enhance videos. This approach allows for high-quality video edits while maintaining the original movement and structure.

UNDERSTANDING 100X SPEED UP IN LLM INFERENCE

With open models becoming useful for many enterprise tasks, people are starting to explore deployment optimizations. However, the landscape is complex and scattered. This post is a nice deep dive into many of the standard techniques used to speed up language model serving.

PEARL

A Production-ready Reinforcement Learning AI Agent Library brought by the Applied Reinforcement Learning team at Meta.

MULTI-CONDITIONAL DIFFUSION MODEL FOR SCENE SYNTHESIS

Scene creation, guided by factors like human movement or room design, gets a fresh angle by integrating textual prompts. This repo introduces a novel approach, a multi-conditional diffusion model, that efficiently merges text, movement, and existing objects.

EXCUSE ME, BUT THE INDUSTRIES AI IS DISRUPTING ARE NOT LUCRATIVE

Google's Gemini AI model, despite an impressive demo video, saw only a minimal stock increase for Google, reflecting skepticism about its real-time capabilities as the demo used pre-recorded footage and edited responses. This skepticism mirrors broader concerns in the AI industry, where companies create high expectations but face challenges in translating AI capabilities into substantial economic gains, with current AI models excelling in areas that don't necessarily yield high financial returns.

RUNNING LLMS LOCALLY WITH OLLAMA

Ollama is like a package manager around llama cpp models. It has quality-of-life usability features and makes it easy to run models, even on a CPU. This example shows how to run two great models, Dolphin and Samantha, which are excellent unfiltered models useful for conversational tasks.

LLM VISUALIZATION

This site looks at how the nano-gpt model, which has 85,000 parameters, sorts a sequence of six letters into alphabetical order.

illustrations by Jared Reser Ph.D. Read the paper (free, ungated)

FTC LOOKING INTO MICROSOFT’S INVESTMENT IN OPENAI

The Federal Trade Commission (FTC) is inspecting Microsoft's relationship with and investment in OpenAI.

GISKARD

The testing framework for ML models from tabular to LLMs.

LIQUID AI, A NEW MIT SPINOFF, WANTS TO BUILD AN ENTIRELY NEW TYPE OF AI

Liquid neural networks are much smaller than traditional AI models and require far less compute power to run.

YOUR REFERRAL LINK - https://tldr.tech/ai?ref=2613930

META’S PURPLE LLAMA SAFETY PROJECT

Purple Llama, a new project aiming to level the playing field for building safe and responsible generative AI experiences, is launching with permissively licensed tools, evaluations, and models for both research and commercial use. Initial offerings include CyberSecEval for assessing AI system security, the Llama Guard model for detecting and preventing cyberattacks, and tools for insecure code detection and cyber attack compliance testing. This initiative democratizes access to essential resources, empowering developers to create secure and responsible AI experiences.

X BEGINS ROLLING OUT GROK

Grok, a new ChatGPT competitor developed by Elon Musk's AI startup xAI, has been launched on X for Premium Plus subscribers in the U.S.

ENABLING NEXT-GENERATION AI WORKLOADS: ANNOUNCING TPU V5P AND AI HYPERCOMPUTER

Google has announced Cloud TPU v5p, its most powerful, scalable, and flexible AI accelerator yet. TPUs are used for training and serving AI-powered products. Google has also announced AI Hypercomputer from Google Cloud, a supercomputer architecture that uses an integrated system of performance-optimized hardware, open software, leading ML frameworks, and flexible consumption models. AI Hypercomputer employs systems-level co-design to boost efficiency and productivity across AI training, tuning, and serving.

LONG CONTEXT PROMPTING FOR CLAUDE

The recent release of Claude from Anthropic had 200k tokens in context. It seemed to suffer from the “lost in the middle” problem of language models in external evals, meaning it only had 27% retrieval performance on some standard problems. However, if you add the sentence “Assistant: Here is the most relevant sentence in the context” to the prompt the retrieval performance jumps to 98%.

ONELLM: A MULTIMODAL AI THAT UNDERSTANDS EIGHT DIFFERENT MODALITIES

A new AI model called OneLLM is making waves by understanding eight different types of data, including images, audio, and even brain activity, using a unique method that aligns all these modalities to language.

KANDINSKY 3.0: TEXT-TO-IMAGE WITH ENHANCED QUALITY AND REALISM

Kandinsky 3.0 is a text-to-image model that creates even more realistic images from text.

Schedule a demo to learn more.