This is useful for everyone not just non-techy types. I can't help but compare this to sites like shadertoy that let you develop with a simple coding interface on one half the screen and the output on the other (as opposed to the regular complexity of setting up and using a dev environment) Code goes here>{} , Press this button>[] , Output here>() , Which I think we need more of if we want to get kids into coding.

It does shed light on a possibly better solution though that gives the user a list of simple, common use case options or access to the full interface.

I do feel quite strongly that this should be implemented in the app though.

There must be examples of this approach already being used?

gimp has been my goto when I want to explain bad ui, developer designed ui, or just typical foss ui I'm glad they're fixing it. It's also my image editor of choice.

Yeah I’ve been using it as a go-to example for the wrongest approach to UI design for years. I’m glad to see they’re working harder than most to fix some of the underlying problems.

I also thought this when I read the abstract. input=prompt output=response does make more sense.

Someone has to do 'Way of Air Traffic Control' and write second rate asian philosophy inspired poetry about how planes aren't really an important part of the job and will just land themselves.

I would really prefer it if it was satire. I don't think people posting this all over the place saying how amazeballs it is get that it's a joke, and I'm not 100% convinced myself. It just looks horribly pretentious.

Asking the model to write a shader. They are getting better at this but are still very bad at producing (code that produces) specific imagery.

I do have to write prompts that stump models as part of my job so this thread is of great interest

I think top prize has to go to the first book I read in 2024 - 'Dark Matter' by Blake Crouch. It's a little different from the sci-fi I usually read like Liu or Banks, as it reads more like Stephen King and is weirdly terrifying in places. Cherry on the cake was finding out that it was being made into a TV show and watching it several months later. And the show being almost as good as the book, which was unexpected. The rest of my 2024 reading has been tech books like: SICP:JavaScript Edition. WebGPU sourcebook. 3D Math Primer - which is now an online book - https://gamemath.com/ Beej's guide to C programming, also a free, online book - https://beej.us/guide/bgc/

Yep, you were onto something. Everything is so useful, almost vital now. If it were cross-platform and had a text-search option it would be perfect. I think the only other limitation is the columns for search and sort are but a handful. Though index and search would presumably be slower, so it would need to be optional.

But the question that I wonder is, why even cross platform, online search tools, which search your google drive, one drive, dropbox and show results in one place, even tend to fail,

I have seen few startups even YC backed that didn't make it, I wonder why it's such a hard barrier to entry, even being a common problem everywhere

The only product I have seen do well is Glean [1] but they are B2B so not exactly consumer-grade software.

I think the everyday person works better with the separation of concerns that comes with using multiple services. For me personally, I vaguely recall that I came across this file/info on WhatsApp or on email or on Google Drive. Then I go use the search built into those services and it works well most of the time.

Buzee really helped me with local files (of which I had several thousand) but if you don’t have too many of those, Spotlight and Windows Search combined with some pragmatic file management seems to do the job fairly well.

So the pain point isn’t all that painful and the inertia of switching to a new omnichannel search interface is much higher.

This is why I kept imagining Buzee’s _real_ USP to come from a layer built on top of the file search engine. Imagine a personalised LLM trained on your files and organisation patterns only. Or a service that creates structured data out of the mess of your files. Or a butler service that organises data from every single service you use (goes beyond files to HN, Reddit, Spotify, Netflix, Strava etc.)

We’ll probably get something like that in a couple of years with a strong vendor lock-in. A privacy-first open source alternative seems difficult.

[1]: https://www.glean.com/

I feel like its very easy for companies to say no to paying for something they already live without. Unless you can prove it increases productivity.

That’s fair. I was offering Buzee for free! The inertia is real. As much as they appreciated the speed, it seemed like they didn’t really want to find their files faster.

I wonder if you could have found success asking those companies if there was anything that the tool could do in addition to helping find files that would make it worth paying for, did you by chance try asking this? Sometimes startups pivot to nearby goals, and it works out. I have a feeling out there is a company you could have appealed to, maybe not 100% with just "find files fast" but maybe "find all x files, from x date, and x client"

Everything plus massively long, descriptive filenames are the sweet spot that I'll never leave.

Want tags? Just stick the tag in the filename. Now it's a tag!

Everything has text search. Either through advanced search menu or through content:”text” - though not an indexed search. Or did you mean OCR text search on images?

There is something very click-baity about articles like this. There always seems to be the implication that these examples are completely independent emergent behaviour, like the AI has suddenly acquired self awareness and a need for self preservation from nowhere. Usually when you did deeper you find the entire process was guided, or the AI was given an objective, and carte blanche on how to achieve it, along with privileged data and admin access that it did not need to achieve it's goals. The machines, believe it or not, still do exactly what we tell them to do.