Search the archive

physical performance. Keeping a good balance of water and minerals

is the name of the game — and that's where LMNT come in.

proven to optimize productivity and health with vital electrolytes

(sodium, potassium, and magnesium) — and none of the harmful additives you get in store-bought sports drinks.

🎁 Claim your free sample pack at [4]DrinkLMNT.com

MISTRAL X HARVEY PARTNERSHIP

Harvey, a legal startup, has partnered with Mistral. The announcement is low on details, but it's likely that they will develop a custom legal model together.

H RAISES $220M SEED ROUND

H, which aims to build action models and was started by DeepMind scientists, raised a monster seed round. Another win for French AI talent.

REFLECTIONS ON OUR RESPONSIBLE SCALING POLICY

Anthropic's Responsible Scaling Policy aims to prevent catastrophic AI safety failures by identifying high-risk capabilities, testing models regularly, and implementing strict safety standards, with a focus on continuous improvement and collaboration with industry and government.

COHERE'S AYA MODEL

The Aya project has 3 different models of increasing size and can converse in 101 languages, many of which are extremely low resource. This project is an incredible step for the open and general access research community.

TEXT-VIDEO RETRIEVAL

Researchers have developed a new method, Global-Local Semantic Consistent Learning (GLSCL), to enhance text-video retrieval while significantly reducing computational costs.

TEXT-BASED UNDERSTANDING OF PROTEINS

Researchers present ProtT3, a new framework designed to enhance text-based understanding of proteins by combining Protein Language Models (PLMs) with traditional Language Models (LMs). ProtT3 integrates a PLM for processing amino acid sequences and a language model for generating high-quality textual descriptions using a cross-modal projector called Q-Former.

free.

DISTRIBUTED INFERENCE ON LLAMA CPP

Llama cpp now supports distributed inference across multiple machines. It is limited to FP16 for now, but is a great step for open source deployment.

ENHANCING LONG-TERM MEMORY FOR LANGUAGE MODELS

Streaming Infinite Retentive LLM (SirLLM) is a new approach that helps large language models maintain longer memory during extended dialogues.

REMOTE SENSING IMAGE CAPTIONING

This project introduces a new probabilistic diffusion model for Remote Sensing Image Change Captioning (RSICC), which describes environmental changes over time.

PALIGEMMA: OPEN SOURCE MULTIMODAL MODEL BY GOOGLE

Google has introduced PaliGemma, an open-source vision language model with multimodal capabilities that outperforms its contemporaries in object detection and segmentation. Optimized for fine-tuning on specific tasks, PaliGemma opens possibilities for custom AI applications and provides comprehensive resources for immediate use. PaliGemma demonstrates superior results in OCR and shows promise for various use cases when fine-tuned with custom data.

AI IS THE REASON INTERVIEWS ARE HARDER NOW

This article discusses the rising complexity of technical interview questions, with companies now requiring candidates to solve harder problems in shorter times. It highlights the use of AI tools like Ultracode that can assist even non-technical users in passing such interviews. The piece suggests in-person interviews as a solution to ensure that candidates actually possess the programming skills needed for the job.

WHAT I'VE LEARNED BUILDING INTERACTIVE EMBEDDING VISUALIZATIONS

An individual passionate about interactive embedding visualizations shares their refined workflow for creating such visualizations, which convey the intricate relationships between entities represented as points in high-dimensional spaces. The process involves data collection, building a co-occurrence matrix, sparsifying the matrix, embedding with PyMDE, and projecting to 2D for clear visualization. The creator utilizes web apps for the UI, advocating for their accessibility and GPU-accelerated rendering capabilities.

KREA AI (PRODUCT)

Generate an AI video from keyframe images based on users' unique styles, concepts, or products.

BYT5 FOR TEXT RENDERING

Glyph ByT5 is a model that operates over bytes and has dramatically improved text rendering in synthesized images.

CASPER LABS ANNOUNCES AI GOVERNANCE SOLUTION, PROVE AI

Casper Labs has introduced Prove AI, developed with IBM, to bring transparency and auditability to enterprise AI applications.

OPENAI PARTNERS WITH NEWS CORP

OpenAI and News Corp (which includes WSJ, NYP, The Times, and other publishers) agreed to bring its news content to OpenAI's platform, which they claim will increase the usability and accuracy of generations.

STANFORD HAI RELEASES UPDATED FOUNDATION MODEL TRANSPARENCY INDEX

Stanford HAI released the latest version of its Foundation Model Transparency Index, which evaluates the transparency of 14 major AI developers, including OpenAI and Google. These companies disclosed new information previously unavailable to the public, marking a significant improvement and willingness to open up the discourse about their models. Despite this progress, the average transparency score was just 58 out of 100, highlighting significant gaps in areas like data access, model trustworthiness, and downstream impact.

MICROSOFT RELEASES PHI-SILICA

Microsoft has announced the general availability of its Phi-3 models and introduced Phi-3-Silica, a small language model optimized for Neural Processing Units in Copilot+ PCs. Phi-3-Silica, with 3.3 billion parameters, offers fast local inferencing with low power consumption, marking a significant step in integrating advanced AI directly into Windows devices to enhance productivity and accessibility. It will be available in June.

AURORA ATMOSPHERIC PREDICTION MODEL

Microsoft has trained a foundation model for atmospheric predictions which has set a new state-of-the-art on 5 and 10 day global weather prediction tests.

INTRODUCING MATHBENCH: A COMPREHENSIVE MATH BENCHMARK FOR LLMS

MathBench is a new benchmark designed to provide a thorough assessment of large language models' mathematical abilities.

A FASTER NEURAL NETWORK ARCHITECTURE

Researchers have developed Wav-KAN, a neural network framework that uses wavelet functions to improve interpretability and performance. Unlike traditional models, Wav-KAN captures both high and low-frequency data components, leading to faster training and increased robustness.

DO WE REALLY NEED MAMBA FOR VISION

Mamba is a strong Transformer alternative that has been gaining steam for its ability to use fewer FLOPs while maintaining performance. However, it may not be necessary for some applications. This work shows that a well-tuned CNN baseline outperforms Mamba on a set of vision tasks.

POWERFUL VISION MODELS ON YOUR PHONE

MiniCMP-V has a new version trained on top of Llama 3. This 8B model outperforms many closed proprietary models on a wide variety of tasks. It can handle 30 different languages and excels at OCR and visual question answering.

IMPROVING MEDICAL AI ACCURACY

MedLFQA is a new benchmark dataset designed to improve the factual accuracy of long-form responses from large language models in the medical field. OLAPH is a framework that trains LLMs to reduce inaccuracies by using automatic evaluations and preference optimization.

TARSIER

Reworkd has released Tarsier, a new tool to enhance LLMs for web interaction tasks by visually tagging elements on webpages using brackets and IDs. Tarsier allows an LLM without vision to understand a webpage's structure via OCR-generated text representations outperforming vision-language models in benchmarks.

THE OLD-FASHIONED LIBRARY AT THE HEART OF THE A.I. BOOM

OpenAI's headquarters, a renovated mayonnaise factory with a library-themed design, is emblematic of the company's language-focused success with ChatGPT. The office contrastingly serves as a reminder of the ongoing legal debates over the use of copyrighted material in AI training. Despite these disputes, OpenAI staff view the library as a space for inspiration, reinforcing their belief in the synergy between human and AI-driven creativity.

CHAOS AND TENSION AT OPENAI

Ilya Sutskever has departed from OpenAI amidst concerns over the company's commitment to AI safety, signaling a potentially worrying trend as three other key personnel have also recently resigned. These departures raise questions about the impact on the company's safety-focused mission and its nonprofit status as it pursues commercialization. These events may also reverberate through legal and regulatory landscapes, prompting scrutiny from stakeholders in Washington.

MISTRAL 7B INSTRUCT V3 (HUGGING FACE HUB)

Mistral has released the next iteration of its 7B model, which has extended context length and improved performance.

THE CHATGPT DESKTOP APP IS MORE HELPFUL THAN I EXPECTED

OpenAI has launched a ChatGPT desktop app for macOS, enabling Plus subscribers immediate access with plans for a broader rollout.

SUNO AI RAISES $125M

Suno, a music generation platform, has raised $125 million to continue building a future where anyone can create music.

Skeptical? Try calling it yourself >> Bland.ai

(P.S. TLDR readers can sign up here and access something even crazier... [5])

SCALE AI RAISES $1B

The monstrous series F comes from Accel and previous investors. Demand for the services provided is enormous and Scale is uniquely positioned to continue powering the modern AI data wave.

XAI WORKING TO MAKE GROK MULTIMODAL

Elon Musk's AI company, xAI, is advancing its Grok chatbot to support multimodal inputs, allowing users to upload photos and receive text-based answers.

MICROSOFT WANTS TO MAKE WINDOWS AN AI OPERATING SYSTEM, LAUNCHES COPILOT PLUS PCS

This article rounds up all of the major announcements Microsoft just made, including a new lineup of Copilot Plus PCs and generative AI-powered features like Recall, which helps users find apps, files, and other content they have viewed in the past.

PHI-3 14B (HUGGING FACE HUB)

The powerful small series of models got an upgrade with the 14B version. It performs on par with Command R (a 104B model).

DICTIONARY LEARNING ON CLAUDE SONNET

Anthropic has made a breakthrough in mechanistic interpretability by mapping millions of concepts inside of Sonnet. It even found that you could push on the internal concepts and change Sonnet's sense of self (e.g., researchers made it believe it was the Golden Gate Bridge).

TEXT CLASSIFICATION WITH LLMS

The Smart Expert System is a new approach using Large Language Models (LLMs) for text classification. This system streamlines the process by reducing the need for extensive preprocessing and domain expertise.

blog for all the juicy technical details!

ENHANCED VIDEO SUMMARIZATION

This project introduces a novel CNN-based SpatioTemporal Attention (CSTA) method to improve video summarization. Unlike traditional attention mechanisms, CSTA effectively captures the visual significance of frames using a 2D CNN to understand relationships and key attributes in videos.

DEBIASING VISION-LANGUAGE MODELS

This project identifies a critical bias in Large Vision-Language Models (LVLMs), where outputs lean more towards language model priors than actual visual inputs. The project effectively reduces this bias by introducing "calibration" and "debias sampling" techniques, leading to more accurate and vision-focused responses across various tasks.

A VISION-LANGUAGE MODEL FOR REAL-WORLD APPLICATIONS

DeepSeek-VL is a new open-source Vision-Language Model tailored for real-world uses with an emphasis on diverse data from web screenshots to charts and OCR.