I did this, perhaps thirty years ago (rocking a flatbed in the 90s #ROFL)... for about two years. Then decided that OCR was terrible. I revisited on a multifunction copier, mid-00s — to the same conclusion.
Once this can be run entirely offline, with simple github installer [0]... I'll be scanning again. This definitely "reminesced a nerve" that "took me back..."
Unfortunately not looking good for accountants, among others...
[0] I'd recon the majority would use a cloud-based, off-device processing — they just selfie each receipt
----
Ten years ago I still used a smartphone; when the banks started allowing mobile deposit, was a very Trekkie day for me...
Me: dropped out of grad school, eventually becoming an electrician (IBEW). Decades as handyman doing various sideworks (my own "startup"?). Retired my own residential electrical contractor license (during Covid), good riddance [1]. Forty-something "you're still young!" #yeahOK
Also me: have worked part-time, as-needed, for three family startups (one as lowly eng.tech, other two in hardware manufacture/assembly [3d-manufacturing & EV energy management].
----
I incurred severe student loan (&c) debt, wasting years both in college (IMHO: don't go, unless it's for an accredited engineering degree[0]) ...and wrecklessly pissing away my twenties drunk-and-stupid (anxiety from being -$235k in-the-hole, then).
When most of my electrician brothers were getting their first $80k pickuptrucks, I was trudging myself out of debt. A decade ago, I became worth $0.00.
[0] Seriously, if you're in college right now: read this again. Whether you want to be a PE, or doctor/lawyer, a B.E. will become an ultimate fallback (and incredible methods of viewing worldly interactions of fundamentals problem solving). To a certain clientele ($$$), that undergraduate in engineering will justify increased billingrates (not as much as MD/JD/MBA, but would still enhance even these).
----
And my body has paid the price of blue collar drudgeries, despite other extremely-fortunate (&unexpected!) windfalls. I've had a handful of weeks in my life where serious consideration has been given to will I ever run/walk again...
Just as I've begun a quest to transition into something less physical (i.e. I dream of desk/office of my own, outside my messy home "office"), this brilliant genAI stuff comes along... and I'm just so glad past blue collar work has allowed me goodénuf savings, even perhaps a few more years of wandering around lost (like most-everybody else increasingly is).
[1] Last advice: you need to find niche tradework — just being a "residential electrician" is increasingly impossible to maintain, with competition from both legal, not, and tech workers. Be the guy (e.g.) that installs (just) meters or lighting or hottubs — or whatever — but don't be the oneguy that does everything (==bankrupt, sooner than not).
----
Life is good, even on a Monday morning. Who the hell knows anymore...
The edit was perhaps personal... actuarily, three decades is what I'd be given =D
Now that I'm revisiting these comments, thanks for pointing out that 30 - 25 == five years into the future [honestly, I hadn't even given this any thought...]
I worked at a Dell repair facility as Apple adopted x86 architecture — and watched as all my bosses gawked at my plastic 13" MacBook Core2Duo [what the Neo most-closely reminds me of, physically and metaphorically]. They'd transition themselves onto larger desktop iMacs within just a few months.
IIRC, 2006's 13" MacBooks started at ~$1000... with a conservative inflation metric this new Neo's pricepoint is absolutely incredible. The last time I remember something like this was 2023 M2Pro Mac Mini, but the above 2006 example was the previous... and so many new Mac users are minted with such brilliant hardware introduction.
>>So I told Codex “we have unlimited tokens, let’s use them all,” and we pivoted to sending every receipt through Codex for structured extraction. From that one sentence, Codex came back with a parallel worker architecture - sharding, health management, checkpointing, retry logic. The whole thing. When I ran out of tokens on Codex mid-run, it auto-switched to Claude and kept going. I didn’t ask it to do that. I didn’t know it had happened until I read the logs.
----
For anybody still thinking my goodness, how wasteful is this SINGLE EXAMPLE: remember that all of the receipts from the article have helped better-train whichever GPT is deciphering all this thermalprinting.
For a small business owner (like my former self), paying $1500 to have an AI decipher all my receipts is still a heck of a lot cheaper than my accountant's rate. It would also motivate me to actually keep receipts (instead of throw-away/guessing), simply to undaunt the monumental task of recordskeeping.
----
>>But the runs kept crashing. Long CLI jobs died when sessions timed out. The script committed results at end-of-run, so early deaths lost everything. I watched it happen three times. On the fourth attempt I said “I would have expected we start a new process per batch.” That was the fix ... Codex patched it, launched it in a tmux session, and the ETA dropped from 12 hours to 3. Not a hard fix. Just the kind of thing you know after you’ve watched enough overnight jobs die at 3 AM.
>>11,345 receipts processed. The thing that was supposed to take all night finished before I went to bed.
Cormack McCarthy proved to me that when structured correctly conversive language does not need any quotation marks (while remaining entirely comprehensible).
But " " still exist (for us/mere mortals) because few can write so clearly.
I know nothing about coding (beyond changing others' variables to fit my installation), but would imagine this parallels many coding environments (e.g. yours).
Does anybody know a good solution of bringing "file labels" (color coding files) back to being more than just adjacent circular dabs — i.e. the previous behavior where the selected-color would illuminate behind the entirety of filename.text?
>Everyone needs a rewarding hobby. I’ve been scanning all of my receipts since 2001. I never typed in a single price - just kept the images. I figured someday the technology to read them would catch up, and the data would be interesting.
This is perhaps among the best openers I've ever read.
[spoiler: the tech caught up, the data is interesting]
It'll be less than a year and an .app will exist solely for this (to then be bought out by Accountoglomerate Co, INC, mostly for all the consumerdata). I can imagine (as a former) business owner support for this is worth $100/month, a personalplan much less but still interested [introfree tier?!].
IMHO the worst part of running any business/project is the paperwork...
My taxesforms still get struck typewritten. For 2022, I thanked ChatGPT (lol). Audit me, I don't care (you will).
Once this can be run entirely offline, with simple github installer [0]... I'll be scanning again. This definitely "reminesced a nerve" that "took me back..."
Unfortunately not looking good for accountants, among others...
[0] I'd recon the majority would use a cloud-based, off-device processing — they just selfie each receipt
----
Ten years ago I still used a smartphone; when the banks started allowing mobile deposit, was a very Trekkie day for me...
reply