# News Highlights

## NVIDIA GTC 2025: Quantum Day to Illuminate the Future of Quantum Computing

- **NVIDIA's Quantum Day** at GTC 2025 will showcase advancements in **quantum computing**, featuring discussions led by CEO Jensen Huang and industry leaders from companies like **D-Wave** and **IonQ**.

## Homomorphic Encryption in iOS 18

- **Homomorphic encryption** enables Apple to perform operations on encrypted data, allowing users to search for images without exposing sensitive information, thus maintaining privacy while enhancing functionality.

## Reversible Computing Escapes the Lab

- **Reversible computing** is transitioning from theory to practice, with Vaire Computing aiming to commercialize chips that recover energy in arithmetic circuits, potentially achieving a **4,000x energy-efficiency gain** over traditional methods.

## Transformer²: Self-Adaptive LLMs

- **Transformer²** introduces a **self-adaptive LLM** that dynamically adjusts its weights for various tasks, enhancing efficiency and performance across domains like math, coding, and reasoning, while requiring fewer parameters than traditional models like LoRA.

## Getting an All-Optical AI to Handle Non-Linear Math

- **MIT researchers have developed a photonic chip capable of processing both linear and non-linear operations in deep neural networks, achieving a latency of **410 picoseconds**, significantly faster than traditional electronic methods.

## How I Found & Fixed 4 Bugs in Microsoft's Phi-4 Model

- **Microsoft's Phi-4 model**, a 14B open-source competitor to GPT-4-o-mini, had **four critical bugs** affecting output quality, which were identified and fixed by the author, enhancing its performance significantly.

## Kaggle Dataset: One of the Input Features Has a >0.99 Correlation with the Target, Yet Most/All Notebooks Do Not Care

- **One feature exhibits a >0.99 correlation with the target**, yet most models fail to leverage this insight, leading to misleadingly high accuracy without proper normalization of the data.

## Titans: Learning to Memorize at Test Time

- **Titans** introduce a **neural long-term memory module** that enhances attention mechanisms by allowing models to utilize both current and historical context, improving dependency modeling without the quadratic cost of traditional attention.

## Train 400x Faster Static Embedding Models with Sentence Transformers

- **Static embedding models can be trained 100x to 400x faster on CPU** than traditional models while maintaining over **85% performance** on benchmarks, enabling efficient on-device and edge computing applications.

## MathReader: A Text-to-Speech System for Mathematical Documents Using OCR and Fine-tuned T5

- **MathReader** innovatively combines **OCR** and a **fine-tuned T5** model to convert mathematical documents into natural speech, treating mathematical notation as a unique language requiring specialized translation.

## HALoGEN: Fantastic LLM Hallucinations and Where to Find Them

- **HALoGEN** introduces a **comprehensive hallucination benchmark** with **10,923 prompts** across nine domains, enabling precise evaluation of generative LLMs' outputs against high-quality knowledge sources.

## Eliciting In-Context Retrieval and Reasoning for Long-Context Large Language Models

- **In-Context Retrieval and Reasoning (ICR²)** enhances long-context language models (LCLMs) by enabling them to process entire knowledge bases for retrieval and reasoning, thus simplifying the Retrieval-Augmented Generation (RAG) pipeline.

## AutoGen v0.4: Reimagining the Foundation of Agentic AI for Scale, Extensibility, and Robustness

- **AutoGen v0.4** introduces a **redesigned architecture** that enhances **robustness, scalability, and extensibility** for agentic AI applications, addressing previous user feedback on architectural constraints and API inefficiencies.

## Logarithmic Memory Networks (LMNs): Efficient Long-Range Sequence Modeling for Resource-Constrained Environments

- **Logarithmic Memory Networks (LMNs)** introduce a **hierarchical logarithmic tree structure** that reduces memory and computational complexity in long-range sequence modeling from **O(n²) to O(log(n))**, making them ideal for resource-constrained environments.