In a thrilling breakthrough announced today, Microsoft and Ben Cochran unveiled Statewright, a system that introduces visual state machines to AI agents, potentially alleviating the pesky issue of AI unpredictability. After two exhausting decades in the tech mines, Cochran revealed his epiphany: why not make the labyrinth of AI decisions smaller, instead of just hurling more data at it?
Statewright, crafted with a Rust engine, deftly evaluates state machine definitions—ensuring pesky models can no longer wander aimlessly through infinite solution spaces. Models in the lumbering 13-20 billion parameter range are delicately shepherded through task-solving sessions like wayward toddlers in a playpen, according to a protocol enforced with none of that bothersome prompting.
“We’ve seen consistent improvements with models navigating these state machines,” said Robby Miffles, Microsoft’s newly appointed Statewright Evangelist. “It’s like giving them a smaller toy box and saying, 'You can play here, but not with the expensive breakable stuff.' They don’t seem to mind the constraints, which is cute.”
In an astonishing turn of events, even frontier models are seen outperforming expectations within Statewright’s watchful embrace. This approach quietly suggests that utilizing a limited yet orderly context at each step might just trump the chaos of unlimited wandering in AI decision-making. Skeptics may call this just another spin of the wheel, but this time there’s a state machine chart to prove its seriousness.
With a free tier available for those brave enough to try it, Statewright welcomes feedback on what problems to solve next, as long as they can be neatly nestled in a plushy state machine-defined cradle. Indeed, the breakthrough message is clear: if you can’t teach an old AI new tricks, maybe you can at least show it who’s boss.