# May 24, 2026

- **High-bandwidth memory (HBM) now constitutes 63% of AI chip component costs**, reflecting a significant increase from **52%** in Q1 2024, driven by rising demand and production costs among major chip manufacturers like Nvidia and AMD.

- **Greg Brockman** reveals the pivotal **72 hours** following Sam Altman's firing, detailing the rapid formation of a backup company and the strategic decisions that shaped OpenAI's future.

- **PICO** (Perceptual Image Codec) is the first learned codec optimized for the **human visual system**, achieving **2.3-3× bitrate savings** over traditional codecs like AV1 and JPEG-AI while maintaining high encoding speeds on mobile devices.

- **LLM agents excel in code generation under loose specifications**, but struggle significantly with strict structural constraints, leading to a phenomenon termed **constraint decay** where performance drops as requirements increase.

- **NeuralNote** offers **state-of-the-art Audio to MIDI conversion**, enabling real-time transcription for any tonal instrument, including voice, with features like **polyphonic transcription** and **pitch bend detection**.

- The **Noroboto.ttf** "lexploit" enables the creation of malicious fonts that misrepresent Unicode glyphs, potentially giving adversaries a tactical legal advantage by embedding deceptive font definitions in documents.

- **Bayesian modeling** effectively addresses the challenge of predicting unknown coordinates in spatial data, particularly in the mining industry, where geologic samples often exhibit strong spatial correlation despite measurement noise.

- **Score severity by collision count**: The proposed model emphasizes that the number of researchers reporting the same bug should dictate its severity, with critical patches required when multiple reports or working exploits are present.

- **`thermocompute` is a PyTorch-first emulator** for thermodynamic probabilistic computing, enabling the study of neural layers where width behaves like parallel fabric, allowing for efficient GPU utilization and future hardware modeling.

- **Nemotron-Labs Diffusion Language Models** aim to achieve **speed-of-light text generation**, leveraging advanced diffusion techniques to enhance performance and efficiency in natural language processing tasks.

- **First-person statements** in fine-tuning a language model yield superior generalization of persona traits, outperforming traditional chat demonstrations and synthetic documents in encoding identity.

- **Vision-capable LLMs** underperformed in accuracy (52.0%) compared to OCR-based pipelines, particularly on **chart-heavy** and **table-heavy** documents, challenging the notion that they can replace OCR entirely.