With an OpenAI leak, the race for "agentic AI" is on
Google, Meta, Microsoft, Amazon and now OpenAI are all known to be working on AI that can "think," make decisions and change course in pursuit of goals.
OpenAI, the company behind ChatGPT, is working on an AI system code-named “Strawberry,” according to Reuters.
(It’s likely the same initiative previously code-named Q* — pronounced “cue-star” — which I told you about in a Computerworld column.)
“Strawberry” is designed to perform what OpenAI internally calls “deep research.” A chatbot or assistant or tool based on this technology would be able to not only answer questions but also plan ahead and navigate the internet autonomously. It's a step towards creating AI that can appear to think and reason more like humans do, moving beyond simple information retrieval and text generation.
“Strawberry” should be able to handle “long-horizon tasks” — complex problems that require planning and executing a series of actions over time. To achieve this, OpenAI is using specialized training techniques, including what they call “post-training” on a “deep-research dataset.” This process aims to refine the model's performance in specific ways after its initial training on general data. In other words, the AI agent goes out intending to achieve a specific goal, and updates its model based on what it finds out and also on changing circumstances and new information.
The details of “Strawberry” are still secret, and the project is still in development. Its public deployment is not likely to happen soon.
“Strawberry” is probably best classified as “agentic AI.” I described and outlined the concepts around “agentic AI” in my Computerworld column published this morning.
And while Google, Meta, Microsoft and, of course, OpenAI, are working on “agentic AI” — and it’s unknown how close any of these companies are to shipping something — Amazon just made a licensing deal with a company that is already marketing “agentic AI” tools and services, called Adept.
“Agentic AI” is theoretically capable of doing the powerful and terrifying things often feared over genAI chatbots. Because it has agency, and permission to make decisions, it could make decisions never intended or even imagined by the people who design it.
Regardless — the biggest companies in AI are now in a full-out race to develop and ship “agentic AI.” This is clearly the Next Big Thing in AI.
Shameless Self-Promotion
The promise and peril of ‘agentic AI’
AI washing: Silicon Valley’s big new lie
Digital nomads just got huge screens and fast internet
The rise of AI-powered killer robot drones
A decade of global cyberattacks, and where they left us
More from Elgan Media!
My Location: Silicon Valley, California
(Why Mike is always traveling.)
