Greetings,

Would you say AI is evil? Some artists seem to think so, and an assistant bot that provides a recipe for chlorine gas probably doesn’t help AI’s reputation. In the end, it’s a tool, and whether it helps or not depends on how it’s built and how it’s used.

• Build, manage, and deploy full-stack web and multiplatform apps in the browser with Project IDX – Google’s AI-powered development platform with features like smart code completion, an assistive chatbot, and contextual code actions. - Project IDX
  
  
• Explore the top 10 most critical vulnerabilities of LLM applications with potential impact, ease of exploitation, and prevalence in real-world applications, researched and summarized by the OWASP foundation. - OWASP
  
  
• Nobody fully understands why Large Language Models work as well as they do – this article explores the phenomenon of ‘grokking’ (model generalization) and introduces the field of mechanistic interpretability as a way to understand models better. - PAIR
  
  
• Nvidia shouldn’t be the only hardware option for AI, and now it isn’t – MLC-LLM lets you compile and deploy models on consumer-grade AMD GPUs with competitive performance. - MLC 
  
  
• Interesting case of human-AI interaction – a supermarket chain released an AI companion to help people use up leftovers, and it provided poisonous recipes like the ‘aromatic water mix’ that was just deadly chlorine gas. - Guardian 
  
  
• Explore three detailed examples of fine-tuning Llama 2 for niche tasks – one of the most promising solutions to extract business value from LLMs, and an area where open-source models win with closed-source like GPT-4. - Anyscale
 
• Artists and data science – outraged authors forced a data scientist to take down an interesting project that produced detailed analyses of books, showing that for some people AI has become synonymous with ‘evil’. - TechDirt
  

----

Found this helpful? Forward this email to a colleague.

Was this email forwarded to you? Sign up here to stay on top of the AI news.