Remote
Kumar McMillan
@kumarvibe@mastodon.social
mastodon 4.8.0-nightly.2026-10-06🇨🇦 staff [object Object] at Shopify 🇨🇦
🦊 open source, open web ally 🦊
🌳 hill walker 🌳
🎶 takes periodic dance breaks 🎶
.
.
.
📸 header photo: Glencoe, Scotland
.
All photos by me unless credited.
872 Followers
1541 Following
50 Posts
Joined August 20, 2018
Music:
Artwork:
Software blog:
Location:
Edinburgh 🏴
My eleven year old has been reading hitchhiker’s guide to the galaxy and he just came downstairs, dying with laughter, “Dad, you have to hear this.” He reads me a section he’s just finished:
“The answer to the great question, the meaning of life, the universe, and everything is … 42”
(Such a joy to see it through his eyes.)
Band photo of me and the fam taken by my wife a while back. Sophomore album out in June.
My friend just finished the John Muir trail, Cottonwood Pass to Yosemite.
His pics are amazing 🤩 I live near Dunbar in Scotland where John Muir was born. In his time, John saw America as an unspoiled version of Scotland. I find this ironic because, today, that wild version of America that John saw has mostly been turned into concrete while his version of Scotland has not.
That said, America is huge and still has vast wilderness. So much beauty has been preserved, thanks to John Muir of course.
Replying to
If anyone ever replies to you with an AI generated, unreadable wall of text, you can call this a slop grenade and send them to https://noslopgrenade.com
Replying to
@anttipeltola@mastodon.world according to the linked article they only ran a benchmark *after* the LLM rewrote SQLite. Yeeeaaah, that’s not gonna work.
Here’s what happens when you tell the LLM to run a benchmark continuously after each change and discard any patch that makes things slower: https://awesomeagents.ai/news/shopify-ceo-ai-agent-liquid-engine-53-faster/
Feedback loops matter. They’ve always mattered for hand written software but they matter more for AI because LLMs are good at reacting to corrective instructions.
Replying to
@jeffmcneill @brainsmatter the point is the user error, though. If real human people are asking open ended questions without feeding in their medical context and getting garbage results then the results are still garbage.
Replying to
@hikingdude it’s looking nice but also very instagrammable so lots of people out for selfies or with tripods 😄
I am living in the right country 😄 I hate voice notes. I have one friend who always sends them and I feel like it’s disrespectful of my time. Text is faster to read. Voice notes are disruptive — they disturb people around me. I don’t feel comfortable playing them out loud on the bus, etc.
I like hearing people’s voices but I just call them directly. This requires pre-planning so I’m in a quiet space.
@zekjur git never ceases to amaze. What a terrible footgun.
Replying to
@landel @jeffmcneill @brainsmatter I’ve had plenty of real human doctors give me the wrong information — things I found out later when speaking to the right doctor or after doing my own research. The difference is you can hold the real human doctors accountable through lawsuits, revoking licenses, etc.
I don’t think we really *need* AI to do the job of doctors [1] but if we do it then it should have that same layer of accountability.
[1] then again, I dunno, some wait times are pretty bad
My taxi driver played me all the music he’d generated with AI, narrating each song. He was very proud of how “original” it was because of his elaborate prompting like “epic goth guitar battle” and so on. He stopped short, though, because he didn’t want to give away his “secret recipe.” It was a long, excruciating ride.
Replying to
@Kathmandu oh god yes, there is an entire alternate universe of fairies living in Edinburgh and all over Scotland (and Wales, England, Ireland…) 🧚♀️✨
Replying to
@hikingdude mostly like someone taking their friend’s photo for an Instagram drop (or whatever)
U.S. presidents come and go but the current Supreme Court is there for life, slowly destroying the country and its people — effects that will last for many generations 😞 https://www.theguardian.com/us-news/2026/apr/29/supreme-court-louisiana-congressional-map-case-ruling
OnlyFans content creators are about to dominate agentic engineering. They have more experience than anyone else right now at chatting with many people in a loop, hurrying them up to completion.
Replying to
@se_davidobrien@mastodon.scot @JackTheCat@mastodon.scot @AbramKedge@beige.party yep, walk highlands in general will allow you to search by hike distance and proximity to you
Replying to
@anttipeltola@mastodon.world if you think tech is bad now, try reading Hackers. There’s only one mention of Margaret Hamilton: she was annoyed she couldn’t use the main frame after Steven Levy broke it (cuz he’s a dick).
Open source collaboration made software fun for me in the 2000s. I’m still excited about the new era. It’s different. For the first time we can build software at light speed and, ultimately, I find *building* software the most fun. I don’t represent Shopify. It’s a big co doing big co things.
Replying to
@emma@orbital.horse @anttipeltola@mastodon.world correct, it’s not programming, it’s building software that people can use. Personally I like using software and it brings me joy to build useful software for others.
Opus 4.6 was a turning point. Prompts just work the first time, no need for “prompt engineering”
Does it wreck the environment? Sure, I guess, but crossing the Atlantic in a boat would be a ridiculous choice. I’d take a jet.
Fascism, really? OK, maybe leave the US like I did? Fascists are everywhere. Good luck
Replying to
@vilain@piaille.fr very cool. Is it referencing the turtle island creation myth? https://deadlystory.com/page/culture/articles/World_s_Indigenous_Peoples_Day/Turtle_Island
Replying to
@johanvos@mastodon.social @n1k0@mamot.fr filter all your messages through an LLM with a prompt to detect this pattern? I’m kidding💀(I’ll see myself out)
Replying to
@shaknais@mastodon.social what was the no code craze of the 90s? Generating soap WSDL APIs from some windows GUI? @glyph@mastodon.social
Replying to
@dabeaz@mastodon.social ha 😂 “Compilers? Where we’re going, we don’t need compilers.” To the future… 🚀
(Although, to be fair, someone will have to write the inevitable compiler for LLM token optimization when everyone admits prompting is stupid.)
Replying to
@moira@mastodon.murkworks.net @cwebber@social.coop nah, they’re just high on Pied Piper VC money. I would be shocked if AI ever achieves utility status. It’s not a conduit like the internet. It’s not essential for survival in any way.
After the dust settles and everyone gives up on their stupid chatbots, AI will just exist as another tool for niche endeavors like software development, bioinformatics, medical research, and a few others that don’t require much creativity.
Replying to
@pluralistic @wendynather weird, it’s still broken for me. I tried loading it for the first time just a few minutes ago.
Replying to
@gws@social.chinwag.org The creative part of coding is in building software that humans actually need, designing the UIs for it, architecting the code for maintenance / scalability, and designing ways to verify quality.
AI won’t replace the humans doing that work. Anyone who blindly wrote code (before AI) without any input into those tasks above also won’t get replaced by AI, they’ll just shift to more creative roles, I think.
The speed-up from not having to type code by hand is just too good to pass up.
Replying to
@gws@social.chinwag.org it’s easy to define success for most coding tasks. The code either works or it doesn’t (and this can be tested). The code might not be elegant or fully optimized but that doesn’t always matter. Example: “write a script to sort these images by aspect ratio.” It’s easy to define success for that. Not much creativity is required to write the code.
This sets coding apart from many other things like art or music where success is not so easy to define. A good song is soulful in a very human way.
Replying to
@mattblaze@federate.social I learned recently that every country has their own AQI — the numbers don’t match if you’re trying to compare since the scales are different. It’s so confusing.
Replying to
@violetmadder@kolektiva.social @moira@mastodon.murkworks.net @cwebber@social.coop we’ll have to see how that one plays out. We can already run open source LLM models like Qwen on a local machine and they’re pretty good for coding.
Replying to
@Snoro@mastodon.social @bitprophet@social.coop after living in Chicago for 25 years where the weather is constantly miserable (aside from ~10 days in September), I’m here from the future to say this can be a good thing: I got so much indoor work done 😄 music, writing, reading, coding, researching, movie watching, whatever.
Replying to
@_thegeoff @jimfrost glories aren’t particularly rare but you can only see them from up high. I see them from airplanes all the time.
Replying to
@sungo running open weight LLM models locally already works pretty well for coding (like Qwen) and it will only get better. We need folks like HuggingFace and Mozilla to keep pushing this ecosystem forward, though. We’ve been here before with Linux vs Windows Server.
Replying to
@andymckay@mastodon.social timeless classic ❤️ he wants to watch the movie next, which I haven’t seen. I’m sure it will ruin everything 😂
Replying to
@hdv@front-end.social it’s all bonkers right now. At Shopify they will call people out who waste tokens. My theory: token maxing schemes are also a way to set up a future system for *restricting* token usage when that day comes.
And, yes, token usage is currently rewarded at Shopify but the problem is they don’t have a better way (yet) to figure out who *isn’t* adopting the actual AI accelerators. Some accelerations you get are undeniable — no one knows what they’re doing but everyone has to at least try 🤷