Foundation Models Need a Foundation Tool
LLMs don’t—and can’t—do everything. What they do is very impressive—and useful. It’s broad. And in many ways it’s human-like. But it’s not precise. And in the end it’s not about deep computation.
So how can we supplement LLM foundation models? We need a foundation tool: a tool that’s broad and general and does what LLMs themselves don’t: provides deep computation and precise knowledge.
And, conveniently enough, that’s exactly what I’ve been building for the past 40 years! My goal with Wolfram Language has always been to make everything we can about the world computable. To bring together in a coherent and unified way the algorithms, the methods and the data to do precise computation whenever it’s possible. It’s been a huge undertaking, but I think it’s fair to say it’s been a hugely successful one—that’s fueled countless discoveries and inventions (including my own) across a remarkable range of areas of science, technology and beyond.
But now it’s not just humans who can take advantage of this technology; it’s AIs—and in particular LLMs—as well. LLM foundation models are powerful. But LLM foundation models with our foundation tool are even more so. And with the maturing of LLMs we’re finally now in a position to provide to LLMs access to Wolfram tech in a standard, general way.
It is, I believe, an important moment of convergence. My concept over the decades has been to build very broad and general technology—which is now a perfect fit for the breadth of LLM foundation models. LLMs can call specific specialized tools, and that will be useful for plenty of specific specialized purposes. But what Wolfram Language uniquely represents is a general tool—with general access to the great power that precise computation and knowledge bring.
But there’s actually also much more. I designed Wolfram Language from the beginning to be a powerful medium not only for doing computation but also for representing and thinking about things computationally. I’d always assumed I was doing this for humans. But it now turns out that AIs need the same things—and that Wolfram Language provides the perfect medium for AIs to “think” and “reason” computationally.
There’s another point as well. In its effort to make as much as possible computable, Wolfram Language not only has an immense amount inside, but also provides a uniquely unified hub for connecting to other systems and services. And that’s part of why it’s now possible to make such an effective connection between LLM foundation models and the foundation tool that is the Wolfram Language.
The Tech to Use Our Foundation Tool Is Here
On January 9, 2023, just weeks after ChatGPT burst onto the scene, I posted a piece entitled “Wolfram|Alpha as the Way to Bring Computational Knowledge Superpowers to ChatGPT”. Two months later we released the first Wolfram plugin for ChatGPT (and in between I wrote what quickly became a rather popular little book entitled What Is ChatGPT Doing … and Why Does It Work?). The plugin was a modest but good start. But at the time LLMs and the ecosystem around them weren’t really ready for the bigger story.
... continue reading