Back to feed
Planning Notes·제품에 대한 소고

Gato, the General-Purpose AI Agent

NS
normalstory
cover image


Most AI used in corporations or public institutions is machine learning. Because in the process of securing and approving budget and running a project, you need explainable validity, causal relationships, and predictability about the mission you're trying to push.
On the other hand, so-called AI in startups or some experimental TFs (excluding innovation-clumps like Tesla) runs on deep learning. Explainability drops, but trust in the data infrastructure and in the results rises dramatically.
But deep learning and machine learning are, at the end of the day, ai (artificial intelligence). In short, they are no more than high-end computers or calculators, and that's their limit.
The opposite of this kind of ai (sometimes called applied / weak ai) is agi (sometimes called strong / complete / general-purpose / general ai). If ai is a computer or robot with intelligence, agi can be read as an entity with intelligence that isn't confined to a specific category — just, broadly, a person-like being. The former is closer to capabilities (analysis, processing, prediction) optimized for a specific role (use-case); the latter, imprecisely, has the capacity to do what's more human-like, person-like (autonomy) in many ways.

In that context, a news about Gato (a 'cat') being released in the agi family caught my eye, so I'm scrapbooking it..




about agi
https://www.irsglobal.com/bbs/rwdboard/14752

What is general-purpose AI (AGI)? The difference from AI, and where the technology stands today.

Source: https://getnews.jp/archives/2529931. Let's look at AGI more closely. Up to now we've compared weak AI and strong AI, AI and AGI — not sure if the difference came through. But still...

www.irsglobal.com




about gato
https://www.deepmind.com/publications/a-generalist-agent

A Generalist Agent

Inspired by progress in large-scale language modelling, we apply a similar approach towards building a single generalist agent beyond the realm of text outputs. The agent, which we refer to as Gato, works as a multi-modal, multi-task, multi-embodiment gene

www.deepmind.com





Related coverage
https://zdnet.co.kr/view/?no=20220515115413#_enliple

DeepMind unveils new AI 'Gato' that performs 604 tasks

The ultimate goal of people in the AI industry is to implement general-purpose AI (AGI). AGI is AI that behaves like an intelligent person — in theory, everything a person can do...

zdnet.co.kr


http://www.aitimes.com/news/articleView.html?idxno=144568

DeepMind takes a step closer to general AI (AGI)… releases Gato, a generalist agent — AITimes

DeepMind, using a single neural network model, can process text, images, and video — generating text, describing images, directing game play, chatting, or controlling a robot's actions.

www.aitimes.com






This English version was translated by Claude.

친절한 찰쓰씨
Written by
친절한 찰쓰씨

Pleasant Charles — UI/UX researcher at AIT. Keeping notes on design, planning, and slow days here since 2010.

More on the author's page

Keep reading

Planning Notes

May 26, 2026·1 min
Planning Notes

Turning AI’s Decisions into Real-World Action

May 24, 2026·2 min
Planning Notes

The two unchanging principles of vibe coding

Apr 12, 2026·3 min