Wednesday, 10 May, 2023

Anthropic developing safe, scalable Generative AI

Anthropic, the exciting new AI startup, wants to raise $5 billion over the next four years for ChatGPT-like text generation technology. They invented a new technique called constitutional AI, where text-generated content is checked or filtered before publishing.

How does it work
The AI system trains two models, one to critique or revise its response using certain pre-defined principles. It then trains another model using AI-generated feedback. This layered approach traps issues before publishing and unlike ChatGPT, it won’t need human contractors to compare outputs manually.

Anthropic has cleverly opted for an improved machine-based auditing approach, allowing the technology to be scaled easily with increased consistency.

The guiding principles or ‘constitution’ is based on the UN Declaration of Human Rights. It is designed to align AI systems to rigorous human-centric values and goals.

Anthropic have said that this is not a complete solution to all generative AI problems, but it is a major step forward in the fight against Generative governance issues.

This AI approach is a masterstroke by Anthropic that may attract more investments in the future. Given the rise in AI concerns, we will see more safeguarding when it comes to governance issues ahead of any potential legislation in the future.

The companies that address these critical governance issues will be in a leading position for years and influence others to change too.

Feel free to contact me today for help with this area

I invest many hours researching the latest AI, technology, and business trends for all my clients. The knowledge is refined, distilled, and combined with many other forms of analysis and expertise to give the best advice possible. 

I can get the best out of AI by:

  • Spotting the latest AI opportunities for your company
  • Pinpointing the best AI experts for your project or company
  • Giving the best advice on AI strategy and approach
  • Helping extract the most value out of  AI
  • …and lots more, please feel free to contact me today

Reference read more

Thank you to these excellent references for inspiring our thinking:

Kyle Wiggers Tech Crunch – Anthropic thinks ‘constitutional AI’ is the best way to train models

Please feel free to contact me today