Agree, it's confusing. While the model of a monothread which can create forked tasks situationally is interesting, their current information architecture makes both task initiation and resumption quite confusing. E.g.: You might assume you should always start an agentic task with your Dot. This is certainly how they position it. But no. Dots have a weirdly different safety bar (refuse things Codex is fine with), for one thing. Dots also can't control your local Chrome via the Chrome extension yet, even with "your computer" authorized and connected, AFAICT.
So users now need to choose between ChatGPT Work, Codex, and Dots ๐ซ
Excited for folks to get to try Gemini 4 Argon. As a "knowledge worker" myself, I've been thrilled by the step-change in the complexity of tasks it can handle end-to-end.
Lots of discussion out there about our next model(!), so I wanted to give an early look as soon as possible. Introducing Gemini 4 Argon!
It shows frontier performance in complex workflows, cyber defense and software engineering. Teams are using it extensively at Google, from coding to quantum computing, great feedback.
Hereโs a look at the benchmarks:
That seems like a roundabout way of saying "user need and motivation."
IMO there's more traction in knowledge work because the motivation to "work smarter" at one's job is clearly incentivized, perhaps even existential for some.
Maybe eventually that translates to consumer. Needs to be proactive--i.e., not starting with a prompt. It will be interesting to see.
I decided to see if Muse would buy me a "mystery gift" without disclosing anything to me about what it was buying.
A primary source about what it was like tending sheep in the Sierra in the 1860s? Honestly, spot on.
@pdrmnvd Makes sense for personal use. App review seems like a pain in the ass if you want to share it easily, though (especially with folks who don't use this stuff)
@alymoursy@AsideAI Would love to understand what capabilities you'd like to see in Gemini in Chrome (Ask Gemini) to help you do things quicker. More agentic capabilities?
(1) Most people don't know how to prompt these tools to do these things
(2) Even if it's easy/preset, there is a huge trust barrier to granting required access and asymmetric loss of trust when it fails
(3) If 1 and 2 are overcome, the systems you're describing will likely adapt; e.g., by blocking access or pivoting to extract rent in other ways.