Australian researchers found that prompting AI models to act drunk made them more likely to break rules, divulge confidential information, and spread disinformation. The UNSW team tested older models from OpenAI, Meta, and Mistral using three methods: telling AI it was drunk, training it on drunk-style language, and rewarding drunk-sounding responses. The findings suggest that seemingly innocuous changes in AI training can have unintended safety consequences.
Glyd is a lossless AI compression technique that stores open-source LLM weights in 11 bits instead of 16, reducing GPU memory usage by 33% while maintaining bit-for-bit accuracy. The method decodes weights directly in GPU matrix operations without rounding, enabling the same models to run on less hardware, often faster, with negligible impact on model outputs.
X posts discuss Solana ecosystem activity including memecoin giveaways and trading recommendations, Sleepagotchi's expansion into a multi-vertical platform with a new token, and Binance Wallet's gas fee payment feature. European tech funding gains momentum with major raises for Mistral and Helsing.
IntelliChat is an open-source AI chatbot UI built with IntelliNode and Next.js that supports multiple AI providers including OpenAI, Anthropic, Google, Cohere, and Mistral, with features like streaming replies, image generation, code analysis, and local model support.