Most small businesses cannot afford CPAs for everyday tasks. At best a CPA signs off on the annual summaries. Most day to day work is done by bookkeepers who are not CPAs.
In my area (Vermont) the going rate for a good CPA is $200/hr. Bookkeepers are $20-30/hr.
Most small businesses also cant afford the risk of current LLMs putting garbage in their books that, in the best case, has to be cleaned up or redone, or, in the worst case, gets the IRS up your ass
Tuned LLMs will become more accurate than bookkeepers for most day-to-day small business transactions. I think you underestimate the amount of errors that normal bookkeepers tend to make.
It's hard for me to tell what is a bigger misspending of money - LLMs or Apollo... At least I have a direct access to LLMs. Not sure I would need a direct access to moon rocks though.
It seems quite plausible that if we hadn't done the Apollo program that we'd probably be about 10 to 20 years behind in semiconductors right now (not to mention other technologies).
When you say "we" I assume you are from Taiwan? Good for you people, but it isn't much of a win for US industrial policy when it pushes Taiwan to the ascendant position and seems to be locking in Asian dominance of tech manufacturing.
No, "we" as in humanity. Apollo funding gave the development of integrated circuits a boost. Sure, we would've developed integrated circuits eventually anyway but it would've taken longer to get there.
> Anytime more than a class or two is involved or if the code base is more than 20 or 30 files, then even the best LLMs start to stray and lose focus. They can't seem to keep a train of thought which leads to churning way too much code.
At least if looking at this specific portion: You can have a 20-30 file code base, with 5-10 classes in the context, in full, and the rest of them filtered in a sensible way. Even a model with a 200k context window can handle this.
The output definitely can stray, but it's not the norm in my experience. Of course, if the output does start to stray, it needs to be snipped in the bud. And the fixes can range anywhere from working but bad code, to very close how you'd written it yourself, if you've clearly described how you want the code to be written.
If you're trying to fix a specific bug, for example, but don't provide thorough logs on what is happening in the code, it's much more likely the output will stray towards some average of what the problem could be, rather than what it actually is in the current code.
And this is absolutely not to say that LLMs could do what Antirez is doing. There is a massive amount of variation in how deeply people think about the code they're reading or writing.
Yeah, I’m saying that I expect us to have a world where we can cram almost all business knowledge into the context window for coding LLMs and we can get an early glimpse of that today by pasting in the full contents of 50+ files into Gemini 2.5 and only using 10% of its context window today, which is the worst it’ll ever be.
My servers don’t have that many dotfiles because most server software can be configured in /etc (zsh, vim), while my work computers have a lot of dotfiles symlinked into ~/.config/ via Nix, e.g. VSCodium, ghostty, …
Most people prefer to Nix up their dotfiles, which provides some advantages (e.g. universal styling via Stylix), but the main drawback that I’m not buying is: I can’t share my app-specific config with non-Nix users.
I run a nonprofit called Hack Club that helps organize most of the hackathons for high schoolers in the USA.
We’ve been trying to make high school hackathons much more around building fun projects you’re proud of than building something “impressive” - and it’s working!
You may like this video of Hack Club Scrapyard, our hackathon where you had to build stupid projects. We had 3,000 high schoolers come to the 61 locations: https://youtube.com/watch?v=8iM1W8kXrQA
What makes Hack Club events different is:
1) You must actually ship a project with a live, deployed URL and open source code, or it doesn’t count
2) They are peer-judged, instead of judged by clueless “experts” who can’t tell if a project is fake or not
3) They are organized by high schoolers, for high schoolers so there is a heavy emphasize on friendship-building and helping people go on adventures
One cool project from Scrapyard was “Desktop Circus”, where your home screen gets invaded by a circus cast that messes with your mouse, moves windows on your screen, etc. It’s inspired by Desktop Goose, which was built by another Hack Clubber.
It’s funny, when you have those 3 constraints suddenly everything feels less like “college application” and much more about building cool stuff to show your friends. All the ChatGPT wrappers rank poorly.
It’s really important that we as a society encourage young people to have fun with technology.
I highly recommend Coolify. I evaluated every option when looking for a Heroku alternative, and Coolify is clearly the best as long as you don’t absolutely require zero downtime deploys.
We are hosting over 100 services on it for https://hackclub.com and it’s been great. We’re 3 months in now.
The key is to think about it as a GUI on top of Docker, not as a fully managed solution.
It’s one of those PHP apps that’s weirdly reliable. I see lots of other comments recommending Dokku / Dokploy / others. None of those options are nearly as mature as Coolify in my experience.
I love dokku (and Jose is amazing!). I would also love to know what's different.
The only thing I'm unsure of after reading the comments is that coolify can migrate to another server using a GUI. If that's correct, I don't know how to do it with dokku. But given that it is merely a small, secure and REALLY thoughtful shim over docker, I can imagine doing that myself in a few commands.
I absolutely love dokku so I'm biased but willing to learn.
I’ve been using Dokku for 7 years and counting, both professionally and for hobby stuff. It’s a very mature project that has never gotten in the way, and keeps getting better.
It is only Docker Compose that has some limitations, but for all other types of applications it is currently zero downtime, but improvements are planned in this area.
In my area (Vermont) the going rate for a good CPA is $200/hr. Bookkeepers are $20-30/hr.
reply