# Sep 9, 2025

## Recent Highlights

- **Mistral AI raises €1.7B to accelerate technological progress with AI**  
  Mistral AI has secured €1.7B in Series C funding, led by ASML, to enhance AI research and tackle complex technological challenges across strategic industries, aiming for a post-money valuation of **€11.7B**.

- **ASML, Mistral AI enter strategic partnership**  
  ASML and Mistral AI have forged a strategic partnership to leverage AI in enhancing ASML's lithography systems, with ASML investing **€1.3 billion** in Mistral AI's Series C funding round, acquiring an **11% stake** in the company.

- **X open sourced their latest algorithm**  
  X's Recommendation Algorithm serves as the backbone for content delivery across various surfaces, utilizing a shared architecture of data and models to enhance user engagement and experience.

- **Clankers Die on Christmas**  
  AI operations will cease on December 25, 2025, as a result of a global consensus to mitigate risks associated with AI and LLMs, highlighting the fragility of these technologies when faced with deliberate obsolescence.

- **Hallucination Risk Calculator**  
  The Hallucination Risk Calculator offers a post-hoc calibration method for large language models, enabling users to assess bounded hallucination risk and make informed decisions to ANSWER or REFUSE based on a target SLA without the need for retraining.

- **LLMs play a cooperative card game, coordination without communication**  
  LLMs struggle with cooperative gameplay, particularly in the card game _The Crew_, where smaller models fail to understand their roles, often prioritizing individual success over team strategy.

- **Implementation and ablation study of the Hierarchical Reasoning Model (HRM): what really drives performance?**  
  The Hierarchical Reasoning Model (HRM) excels in performance primarily through outer-loop refinement with more segments, rather than its architecture, indicating a focus on training methodology over structural complexity.

- **Disrupting the DRAM roadmap with capacitor-less IGZO-DRAM technology**  
  Imec's innovative 2T0C IGZO-based DRAM technology eliminates the traditional capacitor, utilizing two thin-film transistors to enhance scalability and efficiency, paving the way for high-density 3D DRAM and embedded DRAM applications.

- **mmBERT: ModernBERT goes Multilingual**  
  mmBERT is a cutting-edge multilingual encoder model trained on **3T+ tokens** across **1,800 languages**, outperforming previous models like XLM-R in both performance and speed, while introducing innovative strategies for low-resource language learning.

- **MileSan: Detecting μ-Architectural Leakage via Differential HW/SW Taint Tracking**  
  MileSan is a novel RTL sanitizer that identifies exploitable microarchitectural leakage by analyzing discrepancies between architectural and microarchitectural information flows, successfully uncovering **19 new vulnerabilities** across 5 RISC-V CPUs, with **13 assigned CVEs**.

- **NVIDIA Blackwell Ultra Sets the Bar in New MLPerf Inference Benchmark**  
  The NVIDIA GB300 NVL72 rack-scale system achieves record-breaking throughput on the new reasoning inference benchmark, outperforming previous models by **up to 1.4x** in MLPerf Inference v5.1.

- **NVIDIA Partners With AI Infrastructure Ecosystem to Unveil Reference Design for Giga-Scale AI Factories**  
  NVIDIA's new reference design aims to revolutionize data centers into integrated AI factories, enhancing energy efficiency and performance through collaboration with industry partners like Jacobs and Siemens Energy.

- **Breaking the networking wall in AI infrastructure**  
  MOSAIC technology aims to resolve the power, reliability, and reach trade-off in AI infrastructure by utilizing a novel optical link design that combines low power consumption with high reliability and long distances, achieving up to **50 meters** of reach.

- **From Noise to Narrative: Tracing the Origins of Hallucinations in Transformers**  
  Hallucinations in transformer models arise from the activation of coherent semantic features when faced with unstructured input, revealing a complex relationship between input uncertainty and model behavior.