Hi Synapticians!

Discussions about the personality and behavior of large language models (LLMs) have been gaining traction. OpenAI has just released a major update to its Model Spec, now publicly available for use and modification. This document defines the principles guiding ChatGPT’s responses, with a stronger emphasis on customizability, transparency, and intellectual freedom.

Notable updates include a more structured approach to handling controversial topics, clearer moral stances on misinformation and harmful content, and efforts to reduce "AI sycophancy"—where models are overly agreeable. OpenAI has also made the document freely accessible under a Creative Commons Zero license, inviting public input to help refine AI behavior.

We expect that all major LLM providers will follow this move, publishing similar documents and maintaining them carefully over time. Standardizing how AI models handle complex interactions will be critical as these systems become more embedded in daily life.

Here’s an example of what’s in the OpenAI model spec on how OpenAI models should behave “without imposing a subjective moral judgment”

Top AI news

1. OpenAI’s ChatGPT Embraces Intellectual Freedom, Ends Strict Content Limits
OpenAI is changing ChatGPT’s behavior by prioritizing intellectual freedom, allowing discussions on controversial topics while maintaining neutrality. The AI will no longer refuse to answer certain political or moral questions but will instead provide multiple perspectives. A key principle in OpenAI’s updated Model Spec is truthfulness—ChatGPT must not lie or omit crucial context. Some speculate this shift is an effort to align with a possible Trump administration, but OpenAI denies political motives, emphasizing user control instead. While these changes address conservative criticism, they also raise concerns about how AI should handle sensitive issues and misinformation.

2. AI Helps Zillow Create Software With Zero Developers
Replit's partnership with Anthropic and Google Cloud enabled Zillow's non-technical employees to develop software applications without coding experience. These applications have successfully routed over 100,000 homebuyers to agents. The collaboration integrates Anthropic's Claude AI model with Google Cloud's Vertex AI platform, allowing anyone with an idea to create custom software solutions. This initiative addresses the global developer shortage, projected to reach 4 million by 2025, by empowering non-technical teams to build their own solutions. Replit ensures security and reliability through Google Cloud's enterprise infrastructure, making AI development tools accessible while maintaining stringent security standards.

3. Mistral Saba: AI Tailored for Middle Eastern Languages
Mistral AI has launched Mistral Saba, a 24-billion-parameter language model specifically designed for Middle Eastern and South Asian languages. Trained on region-specific datasets, Mistral Saba delivers more accurate and relevant responses than models over five times its size, while being significantly faster and more cost-effective. The model supports languages such as Arabic, Tamil, and Malayalam, and is ideal for applications like conversational support, domain-specific expertise in sectors like energy and finance, and culturally relevant content creation. Available via API and capable of local deployment on single-GPU systems, Mistral Saba responds at speeds exceeding 150 tokens per second.

Bonus. How Sougwen Chung merges AI, robotics, and live art
Sougwen Chung, a Canadian-Chinese artist, integrates AI and robotics into their creative process, transforming art into a live performance. Their work, including Spectral, uses robotic arms guided by AI and brainwave data to create unique, real-time pieces. Rather than replacing human creativity, Chung sees AI as a collaborator, embracing errors as part of the artistic journey. Their approach challenges the traditional narrative of AI as a competitor, instead positioning it as a tool for expanding artistic expression. This perspective reshapes how we think about the intersection of technology and creativity.

Tweet of the Day

And the shepherd's response (Sam Altman, CEO of OpenAI) to the shepherdess (Aravin Srinivas, CEO of Perplexity) 😂 

Theme of the Week

AI Automation with no codes - Concept presentation
No-code AI automation allows people to build and use AI-powered tools without writing any code. Imagine you want to make a robot that sorts your emails. Normally, you’d need to learn programming. But with no-code platforms, you can just drag and drop blocks, like playing with LEGO, to tell the AI what to do. These platforms use pre-built AI models and simple visual interfaces to automate tasks.

Stay Connected

Feel free to contact us with any feedback or suggestions—we’d love to hear from you !