AI
LiteLLM Hack Shows AI Infrastructure Is Fragile
LiteLLM's supply chain attack shows why modern AI infrastructure is still dangerously fragile and far too trusted.
LLM
A collection of posts
AI
Wikipedia Bans AI-Generated Content
Wikipedia officially banned its 260,000 human editors from using AI to write articles. Only typo fixes and formatting tweaks are still allowed. Here is what it means and why it matters.
AI
Wikipedia Bans AI-Generated Articles: New Policy Targets LLM Content
Wikipedia bans AI-generated articles with new policy targeting LLM content. Editors can still use AI for copy editing and translation, but not for writing articles.
AI
GPT-5.4 Ties Gemini 3.1 for the Top Spot
GPT-5.4 matched Gemini 3.1 within 0.01 points. Nine models shipped in March, seven open-weight. The leaderboard race just became irrelevant.
AI
DeepSeek V4 Is Still Missing: What We Know in March 2026
DeepSeek V4 has missed every predicted launch window. As of March 23, 2026, no model exists, no API ID, no announcement. Here is what we know and what the silence means.
AI
Terence Tao on LLMs: Jumping vs Climbing the Walls of Mathematics
Terence Tao frames AI's role in mathematics as jumping versus climbing. LLMs make probabilistic leaps, but the walls of mathematics still require human judgment and sustained reasoning.
AI Security
LLM Jailbreak Research Hits 97% Success Rate Against Frontier Models
New research reveals 97-99% success rates for LLM jailbreak attacks, with large reasoning models now capable of autonomously planning attacks against other AI systems.
LLM
20 Uncensored Local LLMs Released in March 2026: The Complete Guide
20 new uncensored LLMs hit the market in March 2026. From 3GB edge models to 42B MoE giants, here's your complete guide.
AI
Hunter Alpha: The Mystery 1T AI Model Nobody Can Identify
A trillion-parameter AI model appeared on OpenRouter with zero attribution. Early evidence points to DeepSeek V4, but the truth remains elusive.
AI
MIT Ontdekt Slimmere Weg voor LLM Denkkracht
MIT onderzoekers ontwikkelden instance-adaptive scaling: LLM's die zelf bepalen hoeveel rekenkracht ze nodig hebben. Tot 50% efficiënter met dezelfde nauwkeurigheid.
AI
Anthropic Overtakes OpenAI as #1 Enterprise LLM Provider
Anthropic overtook OpenAI as the #1 enterprise LLM provider with 40% market share. A dramatic shift in just 2 years.
AI
Google's Gemini 3.1 Flash-Lite: Faster, Cheaper AI for Scale
Google's Gemini 3.1 Flash-Lite offers API access at just $0.025 per million tokens with 2.5x faster inference, making it the most cost-effective option for scalable AI applications.