Greetings,

Things are moving really quickly. AI is being tested by both the US military and US hospitals with very positive results, but one key limitation is still holding this tech back – you just can’t trust these models 100% yet, both in terms of security and output accuracy. Oh, and did I mention that AI can reproduce images from your brain with almost perfect accuracy? 

• Strict regulation of general purpose AI models could take away societal benefits of AI and lead to centralization, with the most powerful AIs being available only to a handful of people in a few companies. - fast.ai
  
  
• Open source development is one of the antidotes to AI centralization, but it has risks too – it’s easy to publish a ‘poisoned’ model that produces misinformation, and it’s currently impossible to trace if a model was trained on a high-quality dataset. - Mithril Security
  
  
• If you need a simple API that lets you use an LLM to organize and analyze data with a deterministic output, this tool might be for you. - Thiggle
  
  
• To gain an advantage in a world dominated by AI, leaders and C-levels should master soft skills like communication, empathy, and emotional intelligence that will help rally employees while also improving collaboration with AI. - HR Brew 
  
  
• Inspired by mice, researchers enabled an AI agent to self-reflect and learn from past experiences with the ‘curious replay’ method, achieving a significant improvement in how AI agents navigate unknown environments. - Stanford
  
  
• The US Defense Department is using OpenAI’s LLMs for military tasks – the only barrier left before broad deployment is to ensure LLMs are fully secure and produce trustworthy outputs. - Bloomberg
  
  
• American hospitals are testing Google’s Med-PaLM 2 chatbot to reply to medical questions, summarize documents and manage data – physicians prefer its opinions over other doctors in 8 out of 9 categories. - WSJ
  
  
• Many studies like this have been done, but these are the best results ever – Stability’s MindEye almost perfectly reconstructs images from fMRI images of the brain, essentially achieving AI mind-reading capability. - Stability

----

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.