• Tiny, but mighty – the French Mistral AI released Mistral 7B, a small model that matches and even exceeds the performance of similar-size SOTA models, and even bigger ones like Llama 1 34B. - Mistral
• GitHub Copilot and Amazon CodeWhisperer leaked hard-coded credentials (some operational) and secrets from their users’ code, signaling that code copilots might have trouble with data privacy. - arXiv
• “I’d love to build a generative chat app but I can’t code!“ – you can build AI apps without any coding thanks to Hugging Face, here’s how to build a Llama 2-based chat app by clicking through their tools. - Hugging Face
• Building complex LLM workflows just got easy – AutoGen is a framework that simplifies building complex multi-agent conversations, reduces the necessary coding effort by 4x, and will empower a new generation of powerful AI apps. - Microsoft
• Turns out AMD wasn’t sleeping on genAI – enterprise genAI platform, Lamini, unveiled a secret that they’ve been using exclusively AMD Instinct GPUs for their genAI superstations for a year now, and the performance is comparable to Nvidia. - Lamini
• “AI is 10 times, 100 times, more than [PCs, the internet, mobile] in terms of how it’s impacting everything that we do,” said AMD CEO Lisa Su – their answer to Nvidia’s H100s is coming, codename MI300, and will be competitive when it comes to training workloads. - The Verge
• The Meta AI assistant is coming to WhatsApp, Messenger, Instagram, Ray-Ban Meta smart glasses and Quest 3 along with 28 different AIs to talk to – they also introduced AI studio, a platform to build AIs, and several other features including AI-generated stickers. - Meta
