Today’s top stories include Microsoft Azure’s recovery after Red Sea cable disruptions, the launch of AI-enhanced gravitational wave detection by LIGO and DeepMind, and Tessolve’s $150 million funding to expand semiconductor testing amid industry growth. Additionally, Tesla redefined FSD as an advanced driving system, ending its promise of full autonomy.
▶️ Internet Infrastructure
Microsoft Azure Restores Stability After Red Sea Cable Disruptions
Microsoft stated its Azure service is now unaffected after disruptions caused by severed Red Sea cables, which initially increased latency for Middle Eastern traffic involving Asia and Europe.
- Microsoft reported no ongoing issues with its Azure cloud platform after multiple Red Sea cables were cut.
- Initial advisories indicated increased latency affecting traffic through the Middle East, originating or ending in Asia or Europe.
- The cables’ severance cause was not disclosed; Microsoft’s engineering teams worked to mitigate the impact.
LIGO and DeepMind Launch AI Tool to Boost Gravitational Wave Detection
LIGO and Google DeepMind introduced Deep Loop Shaping, an AI-based noise reduction system that significantly improves gravitational wave detection sensitivity and enables lower-frequency observations.
- LIGO has partnered with Google DeepMind to develop Deep Loop Shaping, an AI tool that enhances gravitational wave detection sensitivity.
- The technique reduces mirror motion noise by 30 to 100 times compared to traditional methods, based on a proof-of-concept study in Science.
- Deep Loop Shaping employs reinforcement learning trained on simulated LIGO data to improve noise suppression and expand low-frequency gravitational wave observation capabilities.
Tessolve Secures $150M from TPG Growth to Expand Semiconductor Testing Facilities
Tessolve raised $150 million from TPG Growth to fund acquisitions and expansion, leveraging its end-to-end semiconductor engineering capabilities amid industry growth projected to reach $2 trillion by 2039.
- Tessolve secured $150 million from TPG Growth for acquisitions, expanding test labs, and strengthening global delivery centers
- Founded in 2004, Tessolve has built 11 semiconductor test and embedded labs worldwide and employs over 3,000 engineers
- Major backer Hero Electronix, part of Hero Group, owns Tessolve; revenue grew from $25 million in 2015-16 to over $150 million in FY25
▶️ Management and Leadership
Tesla Redefines FSD as Advanced Driving System, Dropping Unsupervised Autonomy
Tesla reclassified FSD as an “advanced driving system” capable of autonomous transportation under specific conditions, abandoning its original promise of full unsupervised self-driving.
- Tesla redefined “Full Self-Driving” (FSD) in September 2025, abandoning its original promise of unsupervised autonomy.
- Tesla’s current FSD, sold for up to $15,000, now only offers “supervised” driver-assist without autonomous capabilities.
- The company confirmed vehicles from 2016-2023 lack hardware for true unsupervised self-driving; no concrete upgrade plans exist.
Armin Ronacher Criticizes 996 Work Culture and Urges Sustainable Productivity
Armin Ronacher condemns the 996 work schedule, highlighting its risks and advocating for sustainable productivity based on output rather than excessive hours.
- Armin Ronacher criticizes the 996 work culture (9 am to 9 pm, six days a week), emphasizing its unsustainability and negative impact on personal life and company health
- Cites examples from China and Silicon Valley, where such schedules are promoted to build billion-dollar companies
- Advocates for measuring productivity by output, not hours, and warns against promoting burnout as a company culture
Noon Plans to Go Public Within Two Years to Boost Regional E-Commerce
Middle East’s e-commerce platform Noon aims to go public within two years, signaling expansion and increased regional competitiveness in online retail markets.
- Middle East’s Amazon competitor Noon plans a public offering within two years
- No specific financial or valuation details provided
- The article emphasizes Noon’s strategic growth plans and market positioning
From Tech Layoffs to Teaching: Sarah Henschel’s Career Transition for Stability
Sarah Henschel left tech after three layoffs in two years to pursue a one-year master’s in education, seeking stability, long-term benefits, and a fulfilling career in teaching.
- Sarah Henschel experienced three layoffs in two years in the tech industry, prompting her to leave tech sales for a teaching career.
- She is pursuing a master’s in education in NYC, starting within one year, with a salary starting around $80,000 and potential to reach $140,000.
- Transitioning from tech sales to teaching involves a salary reduction but offers long-term stability, pension, health insurance, and job security.
Yondr CEO Embraces Simplicity and Leads Shift to Phone-Free Schools
Graham Dugoni, Yondr CEO, uses a flip phone and opposes social media; his company provides phone-locking pouches for schools and events, supporting the shift toward phone-free environments with infrastructure and cultural change.
- Graham Dugoni, CEO of Yondr, uses only a flip phone and avoids social media, citing simplicity and fewer inputs.
- Yondr produces phone pouches for schools and concerts; the company is profitable with around 150 employees and charges approximately $30 per student.
- The company emphasizes infrastructure support, cultural change, and ongoing assistance for implementing phone-free environments in schools amid increasing bans.
▶️ Technology
OpenAI Research Links Language Model Hallucinations to Guessing Over Uncertainty
OpenAI’s research attributes hallucinations to training and evaluation practices that reward guessing over uncertainty, highlighting the need to update scoring to reduce confident errors.
- OpenAI’s research explains that language model hallucinations occur because training and evaluation reward guessing over uncertainty acknowledgment.
- Hallucinations are plausible but false statements, persisting despite model improvements; GPT-5 has fewer hallucinations, especially in reasoning.
- Current evaluation methods incentivize models to guess rather than admit uncertainty, with accuracy-based metrics encouraging confident errors and hallucinations.
Research Goblin Enhances Search with GPT-5 for Accurate, Context-Aware Results
Simon Willison’s GPT-5-powered “Research Goblin” in ChatGPT significantly improves internet search capabilities, offering detailed, accurate, and context-aware results across various complex queries.
- Simon Willison’s “Research Goblin” uses GPT-5 in ChatGPT with the “Thinking” model for internet search, providing comprehensive results.
- Demonstrates high accuracy and depth in diverse queries, including historical research, building identification, and comparative analysis.
- GPT-5 search integrates tool calling and chain-of-thought reasoning, supporting interleaved thinking and effective mobile use.
Understanding High School Math for LLM Inference
The article outlines high-school math essentials—vectors, matrices, softmax, embeddings, projections—necessary to understand LLM inference, focusing on likelihood vectors, high-dimensional spaces, and neural network projections.
- The article explains foundational math concepts—vectors, matrices, softmax, embeddings, and projections—needed to understand LLM inference, assuming high-school level knowledge.
- It details how logits vectors (e.g., 50,257-dimensional for GPT-2) represent next-token likelihoods in vocab space, and how softmax converts these into probability distributions.
- It describes how neural network layers function as matrix multiplications projecting between high-dimensional spaces, with matrices like 50,257×768 projecting from vocab to embedding space.
Windows 11 Insider Builds Add Keyboard Shortcuts for Dashes
Insider builds of Windows 11 (Dev and Beta channels) now support keyboard shortcuts for em and en dashes via ViveTool activation; other methods include ALT codes and PowerToys.
- Windows Insider builds in Dev (build 26200.5761+) and Beta (build 26120.5770+) channels now include a feature for typing em (—) and en (–) dashes via keyboard shortcuts
- Activation requires using ViveTool to enable shortcuts: Windows key + dash (-) for en dash, Windows key + Shift + dash (-) for em dash
- Alternative methods include ALT codes (0150 for en dash, 0151 for em dash) or using Microsoft PowerToys’ Quick Accent feature