# Jul 26, 2024

## AI achieves silver-medal standard solving International Mathematical Olympiad problems

- **DeepMind's AI systems, AlphaProof and AlphaGeometry 2, solved four out of six International Mathematical Olympiad (IMO) problems**, achieving a score equivalent to a silver medalist, a first for AI in this prestigious competition.

## Launch HN: Undermind (YC S24) – AI agent for discovering scientific papers

- **Undermind** introduces an **AI-driven search engine** designed to streamline the discovery of scientific papers, leveraging **LLMs** to understand complex research goals through conversational interaction.

## Shared Imagination: LLMs Hallucinate Alike

- **Large Language Models (LLMs) exhibit a "shared imagination space,"** enabling them to generate and respond to questions about entirely fictional concepts with surprising accuracy, suggesting a commonality in how these models process and invent information.

## OpenAI announces SearchGPT

- **OpenAI introduces SearchGPT**, a prototype aimed at enhancing AI search capabilities by delivering **fast, timely answers** with **clear, relevant sources**.

## Google creates self-replicating life from digital 'primordial soup'

- **Google's experiment** with a digital "primordial soup" has led to the **creation of self-replicating artificial life forms**, suggesting a potential pathway for understanding the origins of biological life.

## Paper NAACL 2024: "Reliability Estimation of News Media Sources: Birds of a Feather Flock Together"

- The paper introduces a **novel approach** for estimating the **reliability of news media outlets** using reinforcement learning techniques to analyze their web interactions, aiming to enhance **information verification** efforts like **fact-checking** and **fake news detection**.

## Normalization in transformers

- **Normalization techniques in transformers** have evolved from **Post-Layer Normalization (Post-LN)**, which caused training difficulties, to **Pre-Layer Normalization (Pre-LN)**, improving trainability and restoring the **highway effect** despite theoretical concerns.

## TwoMinutePapers - Blender 4.2 Is Here - A Revolution…For Free!

- **Blender 4.2** introduces revolutionary features like **Eevee's real-time rendering** capabilities, including a form of **ray tracing**, **volumetric effects**, and **subsurface scattering**, enhancing the creation of detailed virtual environments and characters.

## TwoMinutePapers - NVIDIA’s AI Learned From 5,000 Human Moves!

- **NVIDIA's AI** has made significant strides in **text to image synthesis**, achieving **character consistency** and enabling the creation of **full stories** without losing character identity, a leap forward in **AI-driven content creation**.

## Recursive Introspection: Teaching Language Model Agents How to Self-Improve

- **RISE (Recursive IntroSpEction)** fine-tunes large language models (LLMs) to **self-improve by introspecting and correcting errors** over multiple iterations, challenging previous assumptions about the infeasibility of such capability.

## Tracing the path to self-adapting AI agents

- **Microsoft and Stanford University's latest research introduces Trace**, a framework for **end-to-end optimization of AI systems**, treating them as computational graphs and optimizing through a generalized back-propagation approach.

## Gene Regulatory Network Inference from Pre-trained Single-Cell Transcriptomics Transformer with Joint Graph Learning

- The study introduces a **novel joint graph learning approach** that enhances gene regulatory network (GRN) inference by integrating **pre-trained single-cell BERT-based transformer models (scBERT)** with **graph neural networks (GNNs)**, leveraging both scRNA-seq data and existing GRN knowledge.

## RestoreAgent: Autonomous Image Restoration Agent via Multimodal Large Language Models

- **RestoreAgent** introduces a **novel pipeline** for **image restoration**, autonomously assessing and addressing multiple types of degradation such as noise, blur, and low light, which traditional methods struggle with due to their requirement for manual task selection and sequence optimization.

## Financial Statement Analysis with Large Language Models

- **GPT4 outperforms financial analysts** in predicting earnings changes using standardized and anonymous financial statements, showcasing its **superior predictive capabilities** even without narrative or industry-specific information.