# Feb 13, 2025

## Outcome-driven Fine-tuning Enhances LLMs' Forecasting Abilities
- **Outcome-driven fine-tuning** enhances large language models' (LLMs) forecasting abilities by utilizing **self-play** to generate diverse reasoning trajectories, improving prediction accuracy by **7-10%** over base models.

## High-Fidelity Datasets for League of Legends
- **High-fidelity datasets** for League of Legends can be created by reverse engineering the game engine, allowing for precise capture of gameplay events like ability usage and player movements, which are typically inaccessible through public APIs.

## o3 Achieves a Gold Medal at the 2024 IOI
- **o3** achieves a **gold medal** at the **2024 IOI** without relying on hand-crafted strategies, showcasing the power of **reinforcement learning** in large language models (LLMs) for complex tasks.

## Automated Capability Discovery via Foundation Model Self-Exploration
- **Automated Capability Discovery (ACD)** leverages a foundation model as a "scientist" to autonomously propose tasks that probe the abilities of other models, revealing thousands of capabilities and failures that would be difficult to uncover manually.

## Text-to-SQL in Enterprises
- **Fine-tuning open-weight LLMs** on business-specific query-SQL pairs achieved **95% accuracy**, significantly outperforming previous methods that plateaued at **85%** accuracy.

## New Paper on Frontier Models Self-Exploration
- **Automated Capability Discovery (ACD)** enables foundation models to autonomously explore and identify their own capabilities, revealing thousands of abilities that human evaluators might miss.

## SWE-agent: Open-Source SOTA on SWE-bench Lite
- **SWE-agent** is an open-source software engineering agent that supports **massively parallel runs** and **cloud-based deployment**, enhancing its usability across various models, including local LMs like Qwen and Llama.

## Temporally Adaptive Interpolated Distillation for Efficient Knowledge Transfer in Language Models
- **TAID** introduces a **novel knowledge distillation method** that utilizes a time-dependent intermediate distribution to effectively bridge the gap between student and teacher models, addressing challenges like capacity differences and mode collapse.

## Novel Clustering Metric - The Jaccard-Concentration Index
- The **Jaccard-Concentration Index (JCI)** is a novel Python library that evaluates clustering quality by combining the **Jaccard index** with a custom concentration score, offering a more nuanced view of cluster purity. [Learn more about JCI](https://pypi.org/project/jaccard-concentration-index/).

## AlignRec Outperforms SOTA Models in Multimodal Recommendations
- **AlignRec** significantly enhances multimodal recommendation systems by optimizing three alignment tasks: **inter-content (ICA)**, **content-category (CCA)**, and **user-item (UIA)**, effectively bridging semantic gaps between diverse content types.

## LLMs as Few-Shot Data Annotators for Multilingual Text Detoxification
- **LLMs can generate high-quality parallel datasets** for text detoxification through few-shot learning, effectively creating toxic/non-toxic text pairs that retain semantic meaning while significantly reducing toxicity.

## Creating a Causal DAG for Irregular Time-Series Data
- **Dynamic Bayesian networks** can effectively model **causal relationships** in **irregular time-series data**, offering a structured approach to causal inference.

## How Scaling Laws Drive Smarter, More Powerful AI
- **Scaling laws** in AI reveal that performance improves with increased training data, model parameters, and computational resources, leading to the emergence of three distinct laws: **pretraining**, **post-training**, and **test-time scaling**.

## Stable Diffusion 3.5 Large Available on Microsoft Azure AI Foundry
- **Stable Diffusion 3.5 Large (SD3.5 Large)** is now accessible on Microsoft Azure AI Foundry, enabling businesses to leverage **professional-grade image generation** seamlessly within their existing workflows.

## Top-Theta Attention: Sparsifying Transformers by Compensated Thresholding
- **Top-Theta Attention** introduces a method that **prunes less essential attention elements** through calibrated thresholds, enhancing efficiency in transformer models without sacrificing accuracy.