# Jul 7, 2025

## Functions Are Vectors (2023)
**Conceptualizing functions as infinite-dimensional vectors** allows the application of linear algebra techniques to diverse fields, including **image processing**, **curve fitting**, and **machine learning**.

## Mercury: Ultra-Fast Language Models Based on Diffusion
**Mercury** introduces a new class of **large language models (LLMs)** based on **diffusion**, achieving unprecedented speeds in coding applications with **Mercury Coder Mini** and **Small** models.

## Anthropic cut up millions of used books, and downloaded 7M pirated ones – judge
**Anthropic** allegedly utilized **pirated millions of books** to train its AI model, Claude, raising significant **copyright concerns** in the tech community.

## Launch HN: Morph (YC S23) – Apply AI code edits at 4,500 tokens/sec
**Morph** enables AI-generated code edits at **4,500+ tokens/sec**, eliminating the need for slow full-file rewrites and unreliable search-and-replace methods, thus enhancing developer efficiency.

## [R] Energy-Based Transformers are Scalable Learners and Thinkers
**Energy-Based Transformers (EBTs)** demonstrate the ability to generalize **System 2 Thinking** through unsupervised learning, enabling them to verify input-prediction compatibility and optimize predictions via energy minimization.

## Mirage: First AI-Native UGC Game Engine Powered by Real-Time World Model
**Mirage** is the first real-time generative engine that allows players to create and modify game worlds dynamically using natural language or input devices, marking a significant advancement in user-generated content (UGC) gameplay.

## The Era of Exploration
**Large language models (LLMs) are rapidly consuming data, with projections indicating that high-quality English web text may be exhausted within the decade, necessitating a shift towards self-generated data for meaningful AI progress.** This transition is termed the **“Era of Experience,”** where the focus will be on collecting the _right_ experiences rather than merely increasing model parameters.

## Show HN: From Photos to Positions: Prototyping VLM-Based Indoor Maps
**VLMs can enhance indoor localization** by utilizing semantic maps and image recognition, allowing for accurate positioning within complex environments like shopping malls, as demonstrated in a recent prototype project.

## [R] Using 'carrier functions' to escape local minima in the loss landscape
**Neural Nets' layered structure** increases model complexity while risking entrapment in local minima due to strong parameter coupling, which reduces the **effective** dimensionality of the loss landscape.

## Evaluating the factuality of verifiable claims in long-form text generation
**VERISCORE** is a novel metric that evaluates the **factuality** of both verifiable and unverifiable claims in long-form text generation, outperforming existing metrics like FACTSCORE and SAFE in human evaluations across eight tasks.

## AI Testing and Evaluation: Learnings from pharmaceuticals and medical devices
**Generative AI necessitates a reevaluation of governance practices**, drawing insights from the rigorous testing and evaluation frameworks used in pharmaceuticals and medical devices to ensure responsible AI development and deployment.

## [P] We built this project to increase LLM throughput by 3x. Now it has been adopted by IBM in their LLM serving stack!

## Enabling Fully Sharded Data Parallel (FSDP2) in Opacus
**Opacus** now integrates **Fully Sharded Data Parallel (FSDP2)** with **Fast Gradient Clipping (FGC)** and **Ghost Clipping (GC)**, enhancing memory efficiency and scalability for training large models like **Llama-3 8B**.