TLDR AIJune 12, 2024

TLDR AI 2024-06-12

6 sections21 entrieslinks blockcomplete parse

Parse notes

Subscription/account footer truncated before parsefooter_truncated

01GENERAL

02HEADLINES & LAUNCHES

MISTRAL RAISES €600M SERIES B

Mistral, a Paris-based AI startup, turns 1 year old today. It announced a monster series B funding raise.

threadreaderapp.com1 min read

ELON DROPS LAWSUIT AGAINST OPENAI

Elon Musk has dropped the lawsuit against OpenAI and Sam Altman that claims they have deviated from their original mission in favor of profit. He has the option to file again in the future.

cnbc.com4 min read

Together has introduced Mixture of Agents (MoA), a novel method that combines multiple LLMs for superior performance that achieves a 65.1% score on AlpacaEval 2.0, surpassing GPT-4o. MoA uses a layered structure where initial responses from various models are refined by aggregators in successive layers, enhancing output quality through collaboration. Despite higher accuracy, MoA faces latency challenges. Future directions include optimizing model architecture and reducing latency.

together.ai7 min read

03RESEARCH & INNOVATION

COMPUTE BETTER SPENT: REPLACE DENSE LAYERS WITH STRUCTURED MATRICES

Most compute for Transformers is spent in the Linear layers. This work builds on muP and Monarch matrices to create a structured representation that uses less compute and has better scaling laws than naive dense layers.

arxiv.org23 min read

TRUSTWORTHINESS IN MEDICAL MODELS

CARES is a comprehensive framework that evaluates the trustworthiness of Medical Large Vision Language Models (Med-LVLMs).

arxiv.org18 min read

ENHANCING ZERO-SHOT LEARNING

PHATGOOSE is a method that significantly improves an AI's ability to generalize and learn new tasks without prior exposure by smartly routing between different specialized language models for each part of a task.

arxiv.org17 min read

04ENGINEERING & RESOURCES

TECH TALK: GENAI FOR ENTERPRISE ANALYTICS

In this webinar: AI thought leaders Steve Nouri and Arun Nandi, and AtScale CTO Dave Mariani, will discuss ways to use generative AI to increase data and analytics accessibility [12] for all enterprise users. Key topics will include NLQ, co-pilot features, and model

bit.lySponsor

suggestions. Join the conversation on June 26

bit.ly

MISTRAL.RS

The Mistral.rs Rust-based inference framework has improvements to its speed, GGUF support, and more in the latest release.

github.comGitHub repository

3D HUMAN MODELING FROM SINGLE IMAGES

The Human Gaussian Model (HGM) is a diffusion-guided framework for creating 3D human models from a single image.

jinnan-chen.github.io2 min read

NIGHTTIME HDR RENDERING WITH REAL-TIME PERFORMANCE

LE3D is a method for real-time HDR view synthesis from RAW images. It is particularly effective for nighttime scenes.

srameo.github.io3 min read

05MISCELLANEOUS

DIABETIC RETINOPATHY DETECTION

Researchers have developed a novel framework that improves the grading of Diabetic Retinopathy (DR), a condition that can lead to visual impairment.

arxiv.org14 min read

EXAMINING LLM PERFORMANCE ON PUBLIC BENCHMARKS

How overfit are popular LLMs on public benchmarks? New research from Scale AI SEAL revealed that Mistral & Phi are overfitting benchmarks, while GPT, Claude, Gemini, and Llama are not. The researchers produced a new eval GSM1k and evaluated public LLMs for overfitting on GSM8k.

threadreaderapp.com10 min read

ANNOUNCING MOZILLA BUILDERS

The inaugural Mozilla Builders Accelerator is embracing Local AI, which brings AI models and applications directly onto personal devices to enhance privacy and accessibility. Key innovation areas include tools for developer productivity, local-based AI agents, dynamic UIs, adaptation of fine-tuning, retrieval-augmented generation, and advanced function calling. The initiative seeks participants to develop a decentralized, open-source AI ecosystem that prioritizes user empowerment.

future.mozilla.org4 min read