No, OpenAI’s new model is not ‘reasoning.’

Happy 10 Year Anniversary, Data & Society!

Hi, it’s Charley and this is Untangled, a newsletter about our sociotechnical world, and how to change it. I launched Untangled Live this month, and my first live, online, course is Sociotechnical Systems Change. That’s a mouthful, so let’s break it down.

Sociotechnical Systems Change is an approach for how we can — as individuals and diverse collectives — transform power dynamics in the systems we’re in, and chart new sociotechnical futures. I’ve developed an 80-page workbook (which you’ll get as part of the course) that integrates key concepts from three areas of study — science and technology studies, complex adaptive systems, and futures — into a coherent approach. If that sounds a li’l wonky, don’t worry, the course is interactive and practical! It offers mindset shifts, tools, and resources to develop your organizational and programmatic strategies and shape your individual actions — and a lab-like environment to help you practice them. Sound interesting? Fill out this li’l survey and become eligible for early-bird pricing.

I was in NYC this week to celebrate the 10th anniversary of Data & Society Research Institute (D&S). I’ve worked at D&S for nearly seven years. I have a deep respect for the organization's rigor, integrity, and creativity, and I’m thrilled to have played a small part in its successes. (This is also a good moment to remind everyone that Untangled is a personal project — all thoughts and opinions are my own.)

Here’s me at the D&S 10th anniversary standing next to my book.

On to the show!

---

No, OpenAI’s new model is not ‘reasoning.’

Over the next few months, we’ll hear all about ‘AI reasoning.’ See OpenAI previewed its new, much-hyped o1 or ‘Strawberry’ model. OpenAI is promoting the model’s ‘reasoning abilities’ — and for that reason, sees it as a massive step toward ‘artificial general intelligence.’ It’s not. Large language models (LLMs) are not ‘reasoning machines,’ they’re prediction machines.