please someone disrupt the email space with an app that uses a LLM to give me options like:
“block every email that looks like an ad person reaching out to me”
“suggest the solution i am typing out now every time you see a support email that asks this question”
“automatically unsubscribe from everything that is a newsletter"
“block every pr request i get”
"rewrite every angry email i receive into a neutral tone”
@jakintosh @zaratustra lol. to be fair, i think it's maybe more nuanced than just being opaque/imprecise.
im also not entirely convinced that the raw cli style interface we have for this stuff right now is going to be the end-all interface for most of it
@jakintosh @zaratustra i dont think that’s it's primary feature. the primary feature of transform models is that they can identify patterns that we cannot. they're like an electron microscope for patterns.
it also happens that we cannot look through this particular electron microscope
@jakintosh it’s been pretty amazing to knock together prototypes with i will admit
no i think we don’t look through it in the specific way that transform models too.
for example Google has a way to identify diabetes through a photo of someone’s eye. That is not something we could previously do.
@jakintosh you’re talking about an abstract process by which people learn that you’re nervous we will stop doing. I'm talking about a literal actual process that exposes things that we previously did not have access to.
I think I am not so worried that people will stop doing the process. You’re worried about losing it just because we have this new tool.
@jakintosh But what has instead happened is people are learning and writing of variety of wild and interesting programming languages, and they're building and working with microcontrollers on a scale way beyond anything anyone was doing along time ago because we have raspberry pis and FPGAs etc.
Tools that make the stuff more accessible may mean a smaller percentage of users in a given field are working with raw technology, but they almost always mean more overall are still doing the hard work
@jakintosh to take your very narrow example, you're worried about friends of yours who already know how to program who are instead programming with this machine.
but instead of looking at them, imagine a world where millions more people can program.
it's the same argument the people creating lua and processing were making.
@jakintosh i think we may disagree fundamentally on the causes and requirements here.
so much of our world is tech based we have a requirement of so many tech workers, this leads to abstractions required so we can allow for lower-skill labor, there just isn't enough high skill labor out there to support the system.
also, afaik most security and efficiency problems dont come from bad software construction, but collisions between environments and workflows
@jakintosh for example, one company subsumes another company, or even one workgroup merges with another. different procedures, structures, software, need to get glued together. everyone can't just pause the business and rebuild and learn a new system from scratch
(this is from my security worker friend, we were just talking about this last week)
@jakintosh im going to skip responding to the energy use thing because that seems like a late stage addition that isn't really what we were talking about before.
i do think it is a full-scale embrace of messiness. I guess I think this idea that problems can be solved elegantly is just not really true in the real world. I can't think of a single real-world problem that has an elegant solution.
even insurance and permitting, which should be straightforward is messy all the way down.
@jakintosh i think part of moving programming into a space where it is more accessible, more functional, and more practical for use in the world, means making it more world-like, and less abstraction-math like.
and i do think you're right that it is a kind of anti-engineering sentiment.
i think where we maybe have the disagreement is that I think the engineering idea that problems can be solved elegantly just does not scale at all
@jakintosh and all it takes is 5 minutes talking to a security researcher to realize how much it doesn't scale (like a major proximity attack vector is ipv6, for example, a system that is beautifully elegant, but was just never really used).
or considering that we don't have a single elegant functional secure software system for an entire business
@helvetica i built a developer demo that does a basic version of this. I’d love to hear your feedback on how you’d want it extended to be useful for a real-world user. Your list here is an awesome start.
My write up on the repo (repo link at the top of the article):
https://www.ashryan.io/nylas-openai-triage-email/
very cool! my initial response which isn't reasonable in context is that i am bummed by how many tools only support gmail which i do not use.
past that though, i think the real trick to a lot of truly great professional ai software is going to be building an interface to the ai that is configurable and trainable on the fly.
the magic of this stuff is that it's kind of a personal assistant. but nobody wants a generic personal assistant, the whole value is personal
@ashryan you can see this part of my ask in the phrasing of many of my requests in my post (learn this specific thing that im writing and also what the ask is from input so that you can reply in my stead to this kind of thing).
i think all of the features i asked for are much more powerful if they can be designed by the user rather than a checkbox that says ‘erase mean emails’
because you never know what someones habits and comforts are going to be
@helvetica +1000 to personalization being critical to the value. I grow tired of re-explaining to ChatGPT who I am and what’s important to me in every chat 😊
fwiw, the demo repo only connects to Gmail since that’s what I use at work, but the Nylas APIs will connect to most email providers out there.
@matti tbh i think we're going to see that last one and stuff like it not just for social media but for the entire web experience.
you can already see some toes in the water on this stuff with artifact (app)
@helvetica Never tried that one, but I agree. Maybe I’m dead inside and negative content should still be part of my world, but my god would I love a toggle to turn off at least some of the time.
(Big fan of your games btw. Can't believe you answered. 🥳)
@matti thanks! ah glad you're enjoying them!
yeah I think it goes way beyond tone policing (although that will be a huge thing).
artifact is doing AI clickbait-title-rewrites and automatic article summaries if you ask for them
Zach Gage
Martin Du4
zaratustra
jak
Ash Ryan Arnwine
matti has moved
PeoriaBummer
Alex Sikora
Ryan Morey
Kristopher Browne
Chris Hokanson
Veith
Rt (ir)Rev Hendel 🤦♂️
Jonathan L.