?? What Anthropic’s ‘hybrid reasoning’ model means for comms
Each week we'll gather headlines and tips to keep you current with how generative AI affects PR and the world at large. If you have ideas on how to improve the newsletter, let us know!
What You Should Know?
New Claude Model Integrates Reasoning, Thinks Longer
Hybrid cars have been on the market for decades, allowing drivers to use electricity at lower speeds or shorter distances and gas when they need more power. Anthropic is applying the same principle to AI with its latest model, Claude 3.7 Sonnet.
“Hybrid reasoning” allows users to control the depth of the AI’s problem-solving process. The model is available for all users, and shows improvements over Claude 3.5 Sonnet in every benchmark Anthropic measured, but paid users can select “Extended” in the model dropdown list to ask Claude to think longer and improve its answers through self-reflection. Extended thinking features a “scratchpad” that displays Claude’s reasoning steps, providing valuable insights into its decision-making process.?
This extended reasoning could be invaluable when crafting crisis response statements, preparing for media interviews, or analyzing potential reputational risks. Instead of relying on one AI model for quick drafts and another for reasoning, users can get the best of both worlds in one model that anticipates tough follow-up questions, identifies messaging gaps, or stress-tests a new strategy.?
Elsewhere …
Tips and Tricks
?? How to prompt reasoning models
What’s happening: AI models designed for complex reasoning, like Claude 3.7 Sonnet or OpenAI o1, require a different prompting approach than typical large language models. While those models have improved at generating outputs based on short prompts, clear and deliberate instructions yield better results in reasoning models.??
He said it: Greg Brockman, President and Co-Founder of OpenAI, posted on X, commenting on how an AI engineer spelled out the anatomy of an o1 prompt.?
“o1 is a different kind of model. great performance requires using it in a new way relative to standard chat models.”
Try this: The elements listed — the goal, return format, warnings, and a context dump — are a great starting point, but there’s no single right way to prompt. You’ll develop your own style based on what works best for you. Experimentation is key. Tweak your approach, test different formats, and refine as you go.
Quote of the Week
“AI is not here to do yesterday’s work better tomorrow. It’s a Ctrl + Alt + Delete moment and you need to decide how you want to reboot.”
— Brian Solis, Futurist, Author and Head of Global Innovation at ServiceNow, at Ragan Communications’ AI Horizons event
How Was This Newsletter?