AI Roundup 05/02 -> 05/09/2024

DeepMind's AlphaFold 3 enhances drug development with AI, OpenAI rumored to launch GPT-powered search, and a hospital simulation with AI agents that outperforms human doctors

Welcome to the weekly edition of AI Tidbits, where I curate the firehose of AI research papers and tools every week so you won’t have to.

---

Overview

• ✨ Highlights (5 entries)

• Language Models (8 entries)

• Vision (3 entries)

• Open-source Packages (5 entries)

Recent Deep Dives

Top 8 leaderboards to choose the right AI model for your task

12 techniques to reduce your LLM API bill and launch blazingly fast products

Harnessing research-backed prompting techniques for enhanced LLM performance

Highlights

1. DeepMind releases AlphaFold 3, an AI model for predicting the structure and interactions of molecules, enhancing drug development with a 50% improvement in modeling protein-ligand interactions with free access for non-commercial research (Google Blog)

   

2. OpenAI is rumored to release a GPT-powered search product to compete with Google and Perplexity (Bloomberg)

3. Tsinghua University researchers release Agent Hospital, where AI agents like MedAgent-Zero evolve and excel in medical tasks, surpassing human experts in medical diagnosis and treatment capabilities (Paper)

4. IBM open sources for commercial use the Granite Code models - a series of code LLMs optimized for a wide range of coding tasks, from code generation to repository maintenance, achieving state-of-the-art performance in software development tasks (Paper)

5. Scale AI publishes findings from a survey of over 1,800 AI leaders, revealing key challenges and the state of AI optimization and evaluation in organizations (Scale AI)

   

Language Models

1. Researchers open-source Prometheus 2 - a groundbreaking, cost-effective language model evaluator, achieving unmatched performance in aligning with human assessments

2. Researchers release ChuXin - a fully open-source language model that excels in deep retrieval tasks, featuring a context length of 1M tokens

3. CMU showcases MAmmoTH2 - a novel LLM that significantly boosts reasoning performance by using web-extracted instruction data, setting new standards in efficiency and effectiveness

4. Microsoft develops YOCO - a decoder-decoder architecture for large language models that enhances efficiency by caching key-value pairs only once, significantly reducing GPU memory usage while maintaining robust performance

5. Microsoft presents vAttention - an innovative memory management approach for LLMs that simplifies architecture and accelerates inference by up to ~4x over prior methods like vLLM

6. Researchers release FlashBack - a groundbreaking retrieval augmented language modeling approach that boosts inference speeds fourfold by appending rather than prepending documents

   

7. Researchers showcase xLSTM - an evolved LSTM framework with exponential gating and innovative memory structures, scaling effectively to rival top-performing Transformers in language tasks

   

8. UPenn showcases DrEureka - an LLM-guided method that accelerates and automates the sim-to-real transfer, successfully applying it to novel robotic tasks such as balancing on a yoga ball

Vision

1. Google showcases ImageInWords (IIW) - an annotation framework that significantly enhances Vision-Language model training with detailed, accurate image descriptions, leading to major improvements in model performance and description quality

2. Imperial College London introduces EMOPortraits - a new model that significantly improves the realism and emotional accuracy of animated head avatars with intense expressions

3. Meta releases a distillation framework for diffusion models, efficiently producing high-quality samples in up to three steps, setting a new standard for rapid generative processes

Open-source Packages

1. OpenGPTs - an open-source implementation of OpenAI's GPTs from LangChain

2. llamafile - a single file executable to run LLMs like Llama 3 locally

3. Insanely Fast Whisper - a command line to transcribe Audio files w/ Whisper on-device

4. EmotiVoice - a text-to-speech engine with >2000 different voices, allowing emotional synthesis including happy, excited, and sad tonations

5. XAgent - AI agents designed for intricate task-solving, outperforming AutoGPT and GPT-4 across tasks and benchmarks

LangChain’s OpenGPTs allows developers to build tool-using assistants easily

Plus >70 more open-source packages for AI engineers

Last week’s AI Tidbits roundup

AI Roundup 04/25 -> 05/02/2024

Reach AI builders, researchers, and entrepreneurs by partnering with AI Tidbits

If you find AI Tidbits valuable, share it with a friend and consider showing your support.