Whoa. This is what I have wanted for the past three years, and they've got it working well enough to actually make my life easier. Some of y'all may have cleaner inboxes, but for me this is a killer app. 🤯
Replies
What are the technical limitations on it being prompt injected by email content and exfiltrating other data via markdown images or similar, I wonder.
sep 2, 2025, 7:54 am • 1 0 • view
Ted will find out , I guess…
sep 2, 2025, 3:45 pm • 0 0 • view
I don't want Chat GPT going through my mailbox.
sep 2, 2025, 7:59 pm • 0 0 • view
*Ted Underwood 6 months from now
sep 1, 2025, 11:17 pm • 9 0 • view
sep 1, 2025, 11:20 pm • 5 0 • view
that was going to be my reply — six months? six fricking hours!
sep 1, 2025, 11:21 pm • 3 0 • view
Love this for you 😂🤣😂
sep 1, 2025, 11:30 pm • 3 0 • view
Ted trying to disable his email auto response next fall
sep 1, 2025, 11:31 pm • 3 0 • view
And all you had to do was give it full access to your Gmail account? I don’t see how this is any different than getting hacked, except they do something slightly helpful and you requested it.
sep 2, 2025, 12:54 am • 1 0 • view
Gemini is OK at this too in my experience, faster and (having only tried to ChatGPT version once without any prompt engineering), maybe better
sep 1, 2025, 11:25 pm • 2 0 • view
is this chatgpt straight up or something else, asking for a friend
sep 1, 2025, 11:19 pm • 1 0 • view
Straight up ChatGPT, the plus plan. I got a pop-up window today asking me if I wanted to make "connections" to apps that included Gmail, Dropbox, &c
sep 1, 2025, 11:23 pm • 2 0 • view
Bluesky
sep 1, 2025, 11:23 pm • 1 0 • view
cool. am curious what the safeguards are, have been thinking about wiring something up but I don't want it to be able to send email directly, only summarize / compose then ask permission to send
sep 1, 2025, 11:28 pm • 1 0 • view
I know. I had to swallow hard. Requires more trust in Sam Altman than I really possess. But I'm afraid this is going to be too seductive for me; I'm quite bad at email triage.
sep 1, 2025, 11:31 pm • 5 0 • view
Right. Like prompt engineering is not a safeguard for this. It has to be API level. But I'm worried they would just expose a bunch of the Gmail API to chatgpt and uh, I dunno about that. Will have to look at it more...
sep 1, 2025, 11:34 pm • 0 0 • view
I've been playing around with having Gemini write scripts to do stuff through the Gmail API, and like that I can read/edit the scripts before running them. Hard to argue that that approach is the one that will catch on though.
sep 1, 2025, 11:40 pm • 3 0 • view
It's under "settings" -> "connectors"
sep 1, 2025, 11:28 pm • 0 0 • view
I never want an "agent" to shop for me. I never want it to make a plane reservation. I never want it to write fan letters to my favorite athlete. I have always wanted it to help me manage this chain of catastrophes, this "inbox," this pile of debris before me growing skyward
sep 1, 2025, 11:20 pm • 21 0 • view
To me this points to the deep disconnect between the management class and others when it comes to these tools: most people have very limited problems that they want solved and would be happy with discrete tools to perform those tasks while management is looking for magic generalised tools
sep 1, 2025, 11:22 pm • 1 0 • view
In a sense yes, but notice also that part where it thought for 1 m 51 s The Gmail integration is nice, but they had to develop generalized reasoning models to make it useful
sep 1, 2025, 11:25 pm • 3 0 • view
I’ve actually found the thinking time to be a real negative for the model—I’ve often found that the longest thinks are for the simplest questions and it’s made me very sceptical of what that element signals
sep 1, 2025, 11:28 pm • 0 0 • view
Understood. But in this case I can see the nature of the challenge and I know it's truly not easy.
sep 1, 2025, 11:34 pm • 1 0 • view
Totally fair—I definitely don’t understand this tech as well as you do :)
sep 1, 2025, 11:38 pm • 0 0 • view
It’s mostly just, I know what a wreck my inbox is, and how many tough judgment calls are required to extract genuine inquiries I need to address from the spam or chatter
sep 1, 2025, 11:43 pm • 1 0 • view
I find this really interesting because when I was interviewed for a web dev position back in 2020 I was asked what I thought the next big change in web would be and they seemed very disappointed with my answer, which was that we’ll need to design for voice interfaces. I feel partially vindicated 😅
sep 1, 2025, 11:46 pm • 1 0 • view
Also my red-team impulses immediately kicked in at your reply and I thought to myself “I wonder how spam and marketing emails will change to get picked up as authentic as this continues to develop ?” 😅
sep 1, 2025, 11:47 pm • 3 0 • view
You’re much better educated on this than I, so I am likely getting things wrong, but I feel like a lot of what it’s needed most is integration with other, traditional technologies where the LLM is more of a human/computer interface than doing the heavy lifting
sep 1, 2025, 11:34 pm • 0 0 • view
The main things LLMs and other models do best at are things I could run off a model on my phone that just does a few things really well and doesn’t need to send all my data back to the company to be monetised
sep 1, 2025, 11:23 pm • 2 0 • view
I was at a UX camp a few months back and had the revelation that most things people use something like ChatGPT for aren’t actually problems that something like that can solve: they’re UX problems that can be fixed with better design.
sep 1, 2025, 11:25 pm • 1 0 • view
I don’t need an LLM to parse a website to extract a business’s address and hours: I just need a data standard for this info included on a site that can be accessed by other tools without needing to “read” the page for that info included
sep 1, 2025, 11:26 pm • 2 0 • view
is this Gemini?
sep 1, 2025, 11:24 pm • 0 0 • view
chat gpt
sep 1, 2025, 11:25 pm • 3 0 • view