Jamie Gaskins

I build distributed systems and try to make life just a little bit more bearable for the people around me.

pronouns: he/him or they/them
location: Cleveland
github: https://github.com/jgaskins
twitter: https://twitter.com/jamie_gaskins

Public Key

npub14ulh2dm4fd7p0essp2kxlfplpht5asvyc5klpe9es0vdacrkcvys602fe0

NIP-05 Address

[email protected]

Profile Code

nprofile1qqs270m4xa65klqhucgq4tr05slsm46wcxzv2t0sujuc8kx7upmvxzgprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmntqyxhwumn8ghj7mn0wvhxcmmvqyt8wumn8ghj7un9d3shjtnswf5k6ctv9ehx2aqpz3mhxue69uhhyetvv9ujuerpd46hxtnfducyqn0z

Publishing to

Author Public Key

npub14ulh2dm4fd7p0essp2kxlfplpht5asvyc5klpe9es0vdacrkcvys602fe0

Show more details

Last Notes

2026-08-24 01:03:23 UTC

- reply

Indeed, it's one of the autistic traits I have to mask a lot

2026-08-24 00:59:34 UTC

"the mid-2000s", as a timeframe
What they mean: 2005 ± a couple years
How I read it: 2500 ± a couple hundred years

2026-08-19 23:53:21 UTC

- reply

Imagine how happy these two would be together

2026-07-19 23:41:06 UTC

An LLM is conscious in the same way that the sex worker you hired really likes you.
That is, it's not, they don't, and the whole thing is just a fantasy.

2026-05-04 21:31:05 UTC

- reply

Interesting, I didn't know about the newsletter. I've got the book, but I haven't read it yet, which I guess is a bit of meta-commentary.

2026-05-04 12:10:44 UTC

- reply

Atomic Habits?

2026-04-03 06:20:27 UTC

- reply

In GitHub's defense, git is such a pain in the ass to learn and GitHub gives you useful abstractions if you want to defer learning about it. That can be useful.
I've been noticing a trend where engineers want to ignore details about a lot of things, though: git, infrastructure, deployment, observability, scale, latency, customers, stakeholders, the consequences of the systems they're building, even the code they're writing.

2026-04-03 06:12:14 UTC

- reply

The second thing that bothers me about GitHub squashed PR commits is that someone is gonna cherry-pick them at some point because "it's so much cleaner than merging the original branch in".
But then when the branches need to be merged, you get conflicts because some commits are applied out of order and multiple times because a cherry-pick is a whole different commit. If you use merge commits, this doesn't happen because each commit is only applied once.

2026-04-03 06:04:59 UTC

- reply

Squashed PR commits don't make your git history clean. It just gives you the illusion of clean. I get that merge commits are ugly, but so is the reality of your repo's history. It didn't arrive at its current state linearly, so don't sacrifice the history's usefulness in an effort to try to make it pretty.

2026-04-03 05:56:56 UTC

- reply

And then, because so many people like to squash-merge GitHub PRs because it "keeps the history clean", you run into the problem where `git blame` points at the squashed commit with the frankenstein message rather than the specific self-contained commit I made.
And since, as I mentioned above, nobody looks at the commits inside the PR, the context is effectively lost.

2026-04-03 05:29:14 UTC

- reply

I say this because I write detailed commit messages. Then in code review, someone inevitably asks why I'm doing something a certain way. The answer to their question is invariably in one of the commit messages.

2026-04-03 02:16:19 UTC

- reply

I also don't like that everything became all about pull requests instead of commits. Nobody reads the commit messages these days. They might as well not be there.
If I wrote all my commit messages at work by rolling my face across the keyboard, not one person would notice.
#OldManYellsAtCloud #GetOffMyLawn

2026-04-03 02:15:09 UTC

The more you understand the inner workings of git, the more you realize all the people who pointed and laughed at GitHub for being a centralized place for everyone's git repos were right.
This is especially true now that GitHub has reached 0 nines of uptime.

2026-03-25 15:18:51 UTC

What's a work-appropriate way to call something a "circle jerk"?
I need to respond to someone at work posting something in Slack without HR getting involved.

2026-02-26 02:13:01 UTC

An important part of world building in a lot of sci-fi shows is establishing swear words

2026-02-20 04:06:56 UTC

- reply

That article covers a lot of my thoughts on AI and is a great read. It even covers the topic of a thread I posted the other day (copyright implications with AI) which, coincidentally, Cory participated in.

2026-02-20 03:35:50 UTC

- reply

You'd think if it was really as helpful as its boosters claim, they wouldn't have to impose it on us.
I honestly don't think it's about AI as a technology, but the wealthy investors who have a stake in AI companies pushing it on the companies whose boards they sit on because it's an easy sell to the CEO ("increase capacity without headcount" is music to a typical startup exec's ears). That is, I suspect they're juicing adoption numbers to inflate their portfolio.

2026-02-13 03:57:51 UTC

- reply

The argument has been that the model doesn't contain the copyrighted works directly. Like, you can't grep the model file on disk for a passage from a book it can still somehow reproduce.
It's a ridiculous argument, though, because the models deal in numbers, not text. Those numbers are converted to text for human consumption only, so of course it won't contain the raw text anywhere in the model.

2026-02-12 21:49:35 UTC

- reply

Buddy, you're the only one that's been whining this whole time. Whining about what I said, whining about "get a Claude subscription".
I was literally talking about "I'm gonna have popcorn ready". I don't know how you read fear from that.
It seems more like you feel attacked because someone criticized AI. You've been the only one alarmed in this whole thread.

2026-02-12 21:34:38 UTC

- reply

You came into my mentions with guns blazing, so you'll have to forgive me if I don't consider that to be an honest question.

2026-02-12 21:32:18 UTC

- reply

The guarantee is that the parts of the codebase that aren't written by humans are not called out. You can read into the rest however you like.

2026-02-12 21:24:45 UTC

- reply

I don't have a JD, that's correct. I do, however, have two siblings that do and this entire thread came out of a conversation with one of them.
And no, I will do you no favors. You came into *my* mentions. I had no idea you existed 40 minutes ago. So I owe you nothing. You'll just have to find a way to accept that I will continue posting my thoughts on my own account on my own server.

2026-02-12 21:11:10 UTC

- reply

My company pays for both Claude and Cursor. I use both and my team discusses these tools daily.
I'm also very familiar with how LLMs work down to a mechanical level. I've trained models from scratch with code I wrote by hand and used those models to do some pretty neat things.
I assure you, you don't have a leg up on me in this discussion.

2026-02-12 20:57:50 UTC

- reply

You find that alarming? I meant it like "Holy shit, this is an interesting thing I hadn't considered."
And it's not an absurd claim. It's pretty much exactly what someone will say in court.

2026-02-12 20:52:51 UTC

- reply

There's no alarm in what I said.

2026-02-12 20:41:01 UTC

- reply

FWIW I'm not a lawyer and I'm not recommending that you do this. 😄 Even if companies have no legal standing on copyright, their legal team will try it. It *will* cost you money.
But man, oh man, I'm gonna have popcorn ready for when someone inevitably pulls this move.

2026-02-12 20:37:07 UTC

- reply

It'll be interesting to see what happens when a company pisses off an employee to the point where that person creates a public repo containing all the company's AI-generated code. I guarantee what's AI-generated and what's human-written isn't called out anywhere in the code, meaning the entire codebase becomes public domain.
While the company may have recourse based on the employment agreement (which varies in enforceability by state), I doubt there'd be any on the basis of copyright.

2026-02-12 20:23:14 UTC

If you use AI-generated code, you currently cannot claim copyright on it in the US. If you fail to disclose/disclaim exactly which parts were not written by a human, you forfeit your copyright claim on *the entire codebase*.
This means copyright notices and even licenses folks are putting on their vibe-coded GitHub repos are unenforceable. The AI-generated code, and possibly the whole project, becomes public domain.
Source: https://www.congress.gov/crs_external_products/LSB/PDF/LSB10922/LSB10922.8.pdf
https://uploads.zomglol.wtf/media_attachments/files/116/059/407/618/511/575/original/bb3695240426d40d.png
https://uploads.zomglol.wtf/media_attachments/files/116/059/412/520/242/557/original/7177c68f88a00d30.png

2026-02-11 18:19:56 UTC

- reply

I really like the idea of comparing it to a placebo. It's easy to *feel* productive when using AI coding tools. It's a very similar dopamine hit to when you're really in the zone and I think a lot of folks are having trouble telling the difference.
But if they were more productive with it, January would've dominated previous months.

2026-02-11 17:57:11 UTC

A few devs on my team are convinced that their recent usage of AI has made them more productive than ever.
See if you can guess when the company began paying for AI subscriptions for the team.
---
February is low mainly because it's in progress.
https://uploads.zomglol.wtf/media_attachments/files/116/053/260/978/797/756/original/f257481d72125fe0.png

2026-01-24 23:02:48 UTC

- reply

At this point, they probably *should* be afraid for their lives. We outnumber them on the order of 1000:1 or more. Their gear will not save them if the people decide to hit back.

2026-01-16 18:12:40 UTC

This post by Bruce Schneier contains so many thoughtful soundbites:
> The question is not simply whether copyright law applies to AI. It is why the law appears to operate so differently depending on who is doing the extracting and for what purpose.
> Like the early internet, AI is often described as a democratizing force. But also like the internet, AI’s current trajectory suggests something closer to consolidation.
https://www.schneier.com/blog/archives/2026/01/ai-and-the-corporate-capture-of-knowledge.html

2026-01-13 00:14:07 UTC

- reply

I think it sends a request to the other user to approve the quote. Similar to how an ActivityPub `Follow` message requires an `Approve` message.

2025-12-15 17:11:20 UTC

- reply

I think about this video a lot because something LeVar says in it flipped a switch in my mind: "if they could get away with it, they'd pay you nothing".
It never occurred to me before then because if I were a CEO I wouldn't be like that. But then I remember a company I worked at where they replaced the CEO who thought more like I did with someone who thinks more like Benioff does and the drastic difference it made in the culture. I left a month later.
https://www.youtube.com/watch?v=MAoWo7yFRQk

2025-12-08 05:44:16 UTC

- reply

My favorite part is that the organization governing a sport where you can't touch a ball with your hands gives a guy that cheats at everything an award that looks like a bunch of people trying to touch a ball with their hands.
https://uploads.zomglol.wtf/media_attachments/files/115/682/337/011/756/889/original/7a9a570f661ebb71.png

2025-06-27 15:13:22 UTC

- reply

I only found out about that movie on Wednesday and saw Lance Reddick, one of my favorite actors and a Baltimore guy, on the movie poster. The movie must've been in production for a loooong time because he died 2 years ago.

2025-06-04 05:00:49 UTC

- reply

I think I've got a different read on it. IMO, mediocre code is code that gets the job done to specification, just in a boring way.
For example, plain-old CRUD. It's mediocre, boring, and bland. It works and you can trust it to work, but it doesn't go the extra mile.

2025-06-04 01:08:44 UTC

- reply

What's the objection on this one? Asking because I'm actually not sure I disagree with it and I'm wondering if we're just reading it differently.

2025-05-05 18:55:39 UTC

- reply

And then that reminds me of Commander Adama’s line in Battlestar Galactica, “There's a reason you separate military and the police. One fights the enemies of the state, the other serves and protects the people. When the military becomes both, then the enemies of the state tend to become the people.”
The military aren’t the cops (with caveats, I have stories) but we do have cops armed like soldiers (and often better) with effectively no accountability.

2025-05-05 18:51:08 UTC

- reply

I was literally thinking yesterday about how Robocop feels more prescient today than it should have.
Cops have always served the whims of the capitalist class but these days that seems more overt, direct, and robotic and they are much more heavily armed than they should be.

2025-05-03 06:09:13 UTC

- reply

This is where I'm at with it, too.
I asked because I keep wondering if I'm missing something. I see a lot of people saying some of the small self-hosted models are as good as the SOTA models. They're a huge improvement from older self-hosted models, certainly, but I'm just not seeing the kinds of results they're reporting.

2025-05-03 00:04:14 UTC

- reply

What are you using it for?

2025-04-16 16:24:56 UTC

- reply

I'm thoroughly interested in this. Have any of these presentations been recorded in a place we can watch online? Happy to pay to access it if needed.
My interest comes from also often having to respond to overenthusiastic people who don't understand the tech, both at work and outside of it. I'm often unsuccessful in convincing folks it's not the panacea they desperately want it to be, so I would love to see and hear how you present it.

2025-01-23 19:15:03 UTC

- reply

I respect so many of the things you say, even when I disagree with your position. That's been true for over a decade now. Please keep that in mind as you read the next part.
The comparison of LLMs to guns and meth sounds every bit as ridiculous as you see Simon's position. I have my own issues with "AI" (including how the term is applied), but LLMs are not inherently destructive like either of those things are.
@npub1028…0xd3 @npub1gv2…tlwl

2024-12-23 21:49:19 UTC

- reply

This definitely explains a lot of what I feel when I see people in the AI field claiming an LLM has any level of consciousness. They've bought into the delusion.
Reminds me of the quote by Upton Sinclair: "It is difficult to get a man to understand something when his salary depends on his not understanding it."

2024-12-01 19:09:05 UTC

- reply

Usually when things go wrong on my computer, it's one of these things. They're all safe to restart, so I literally just write `plzwork` in a terminal and things generally start working again.

2024-12-01 19:05:47 UTC

Just in case folks need it, here is the shell function I use when basically anything stop working in macOS. It restarts all of these things:
- Dock
- Spotlight
- Universal Control
- Bluetooth
- Audio
… and flushes the DNS cache.
```
plzwork () {
killall Dock corespotlightd "Universal Control"
sudo killall -9 -e bluetoothd coreaudiod
sudo dscacheutil -flushcache
sudo killall -HUP mDNSResponder
}
```

2024-09-14 19:30:52 UTC

- reply

I hadn't heard of that before, but it looks like it. I was doing this 2-3 years ago, before their first release, but I may give this a shot.

2024-09-14 19:02:28 UTC

- reply

I tried implementing this a few years ago. I still fronted it with Postfix because I wasn't prepared to deal with the public internet.
I had Postfix route the emails to an LMTP server which I wrote in Crystal and was not exposed to the internet. It stored the messages in a processing queue (a SQL DB table). Processing involved splitting out text/HTML parts, attachments, etc, and storing everything where it needed to go: message content in the SQL DB, attachments via an S3-compatible API.

2024-09-14 18:41:56 UTC

- reply

Postfix and Dovecot aren't even good software for hosting your own email server. They're just the least terrible thing.
I want a unified SMTP/IMAP server that stores emails in a SQL DB or Elasticsearch or S3 or literally anything other than RFC2822-encoded files on disk somewhere.

2024-09-14 18:31:57 UTC

I really wish Gmail filters weren't horrible. Honestly, I wish there were easy ways to run your own email server.
Everyone who says it's easy is like "All you have to do is set up postfix and Dovecot and DNS PTR and DKIM and DMARC and Return-Path and …" no fuck you.
I know why all these things are needed. I know why it's complicated. It still sucks that there isn't a good, simple way to host your own.