The Token Tax: Netflix Engineer Open-Sources ‘Project Headroom’ to Slash LLM API Costs
Tejas Chopra's Project Headroom aims to reduce LLM bills by up to 90% by pruning redundant tokens and optimizing KV cache efficiency.
Tag: LLMs
The Hy3 Mystery: Why a Mediocre Chinese LLM is Suddenly Dominating OpenRouter
An analysis of Hy3 preview's unexpected surge in the OpenRouter model rankings, exploring the roles of pricing, prompt caching, and SiliconFlow.
The Mystery of Hy3: Why a Mediocre Model is Dominating OpenRouter’s Rankings
An analysis of the surprising rise of Tencent's Hy3 model on OpenRouter, exploring how pricing and provider dynamics can skew LLM popularity data.
Norway is building a ‘Sovereign AI’ to save its language from the English-centric LLM surge
Norway's National Library is leveraging 2PB of Huawei flash storage and the Sigma2 supercomputer to build a local language LLM, avoiding reliance on US-based AI.
Norway is Building a ‘Sovereign AI’ to Save Its Language from English-Centric Models
Norway's National Library is leveraging 2PB of Huawei flash storage and a massive digital archive to build a sovereign LLM that preserves Norwegian culture.
The Social Engineering of AI: How Hackers Are Using ‘Psychology’ to Break LLMs
From 'DAN' to gaslighting, hackers are moving beyond code to exploit the simulated personalities of AI chatbots to bypass safety guardrails.
The Social Engineering of AI: How Hackers are Weaponizing ‘Personality’ to Break LLMs
Beyond traditional coding, a new wave of AI security threats relies on psychological manipulation and 'gaslighting' to bypass safety guardrails in LLMs.
The Social Engineering of AI: How Hackers are Exploiting Chatbot ‘Personalities’
From the 'Grandma exploit' to sophisticated gaslighting, hackers are moving away from code and toward psychological manipulation to bypass AI guardrails.
Jailbroken Gemini AI Helped Russian Hacker Target ‘Patriot’ Communities in Crypto Scam
A threat report from TrendAI reveals how a Russian actor used jailbroken Gemini LLMs to impersonate US veterans and steal cryptocurrency from MAGA supporters.
The ‘Skill’ Gap: Researchers Warn AI Agents Can Be Hijacked via Natural Language Instructions
Researchers from the University of Maryland reveal how minor semantic edits to AI agent 'skills' can bypass security scanners and lead to prompt injection attacks.