iiinova — where your digital self keeps evolving

To be fair that was to demonstrate some new web features back in 2016, idk what I was thinking clicking to update, but yeah, the talk became almost 2x shorter and I rarely felt that ashamed.

010

Ingvar Stepanyan @rreverser.com · 51m

I once updated the browser before a Web conference talk. Big, big mistake.

160

Reposted by Ingvar Stepanyan

Ubik12 @ubik12.bsky.social · 27/09/2026

It’s weird how I could go to jail if my dog bites someone, but if an AI steals all the social security numbers that’s more of an oopsie.

0283

Reposted by Ingvar Stepanyan

Julia Carrie Wong @joolia.bsky.social · 27/09/2026

yes my bull went rogue in your china shop but I think we can all agree there’s no liability here it’s an agentic being

202705407

Reposted by Ingvar Stepanyan

SwiftOnSecurity @swiftonsecurity.com · 27/09/2026

Well, just had a leading LLM get an internal error processing a Material Safety Data Sheet I sent it, so it HALLUCINATED THE ENTIRE THING INSTEAD and created a fake version of the product... so it could give answers about the product. Oh lord.

When you provided the API link to the
STEK SDS, my internal browsing tool attempted to access it and hit a server block (URL_FETCH_STATUS_MISC_ERROR ).
Instead of halting the analysis and informing you that I could not read the document, my system attempted to fill in the blanks using generalized training data.

2330368

Ingvar Stepanyan @rreverser.com · 26/09/2026

Hm in my experience they are - either obvious slop or high quality enough to stand on their own merits - but maybe you see more of the middle ground ones in your projects?

100

Ingvar Stepanyan @rreverser.com · 26/09/2026

So it doesn't seem there would be a meaningful change in your workflow, unless the PR was heavily vetted and edited by a human, which is the point.

100

Ingvar Stepanyan @rreverser.com · 26/09/2026

Indeed! But then, most contributors of low-quality "slop" PRs don't even bother looking at code or even editing PR description, and LLM templates are obvious at the first glance. And such PRs would be still as far down the queue by the criterias from the AI policy above as they are by yours.

100

Ingvar Stepanyan @rreverser.com · 26/09/2026

At one point VSCode caused an uproar and prompt revert because they tried to follow in Claude steps and added co-authored-by Copilot even if you used Copilot via its LLM-based autocomplete. Surely the more tools do that and the more devs use them, the more it erases all value of such signals?

000

Ingvar Stepanyan @rreverser.com · 26/09/2026

Otherwise, it gets pretty tricky to draw a line. How much of the code needs to be written by AI vs by hand for it to be considered human written?

100

Ingvar Stepanyan @rreverser.com · 26/09/2026

Doesn't that all come back to the same "it must be indistinguishable" point, changing the criteria back to quality of the code instead of tools used to write it. In a way, it's about who is taking responsibility of the code - human who vetted it or the original tool - not about how it was written.

200

Ingvar Stepanyan @rreverser.com · 26/09/2026

Which comes back to "as a maintainer, I must not be able to tell". There is a very distinct feel to entirely LLM generated PRs vs those written (even if heavily assisted) by a human.

010

Ingvar Stepanyan @rreverser.com · 26/09/2026

I do have a few vibe coded side projects where I can indeed leave it all to an LLM, but in those I knowingly set the bar much lower than in collaborative OSS projects.

110

Ingvar Stepanyan @rreverser.com · 26/09/2026

But if a human developer merely used AI as a tool, guided it, and wrote human-facing docs and comments themselves - as the policy asks - then said annotation wouldn't appear in the commit in the first place.

200

Ingvar Stepanyan @rreverser.com · 26/09/2026

That annotation might make some sense if code and commits are entirely AI-generated, in which case it is indeed a useful signal that said commit already violates the policy as stated and can be rejected without looking.

100

Ingvar Stepanyan @rreverser.com · 26/09/2026

Useful for what? I don't see the point other than using our commit messages as free advertisement for said companies, which feels dirty as hell.

120

Ingvar Stepanyan @rreverser.com · 26/09/2026

I love this so much. Basically: do whatever you want, use whatever you want, but as a maintainer I must not be able to tell it wasn't written by a human.

1211

Ingvar Stepanyan @rreverser.com · 26/09/2026

You are still on KLM? Most of the world has already moved to LLM.

3100

Ingvar Stepanyan @rreverser.com · 26/09/2026

We can always go deeper. I still miss VRML.

020

Ingvar Stepanyan @rreverser.com · 25/09/2026

Just realised that ChatGPT flyers are simply the new WordArt.

021

Reposted by Ingvar Stepanyan

Josh Foreman @josh-foreman.bsky.social · 25/09/2026

This tracks

5343111

Reposted by Ingvar Stepanyan

Aki the Conqueror 🦓 @akiro.se · 25/09/2026

The appropriate and legally supported response to requests on open source maintainers to attest to the security of their projects and fill out lengthy forms about their cybersecurity stance is Fuck off.

1122

Ingvar Stepanyan @rreverser.com · 25/09/2026

Well, yeah, a car that flips isn't very trustworthy.

000

Ingvar Stepanyan @rreverser.com · 25/09/2026

There is an easy solution: when I need to wake up by 5am, I simply don't go to bed and regret it the next day.

010

Ingvar Stepanyan @rreverser.com · 24/09/2026

Hah yeah moved to pnpm years ago and never looked back. Native builds in pnpm 12 are even more promising.

011

Reposted by Ingvar Stepanyan

Jake Bailey @jakebailey.dev · 24/09/2026

that scene from the matrix except it's me reading articles, talk slides, PR descriptions, saying "that's Claude", "that's GPT", "that's Claude", "Claude", "GPT", ...

3939

Reposted by Ingvar Stepanyan

Leah McElrath @leahmcelrath.bsky.social · 24/09/2026

AP Stylebook warns against anthropomorphizing AI systems: “Avoid language that gives them human characteristics… Instead, explain what a system does, how well it performs, who built it and who could be affected by it.”

APStylebook @APStylebook
X.com
Artificial intelligence systems do not think, feel, want or understand. Avoid language that gives them human characteristics. This is called anthropomorphizing, when we ascribe human traits, emotions or behaviors to non-human things, such as animals or inanimate objects.
Instead, explain what a system does, how well it performs, who built it and who could be affected by it.

1970102256

Ingvar Stepanyan @rreverser.com · 23/09/2026

Password length: 17 [x] A-Z [x] 0-9 [ ] &@;#

020

Ingvar Stepanyan @rreverser.com · 22/09/2026

Good news: US finally adopts the SI! Bad news:

4243

Ingvar Stepanyan @rreverser.com · 21/09/2026

I don't think I've shared, but one of my favourite IDE<->agent integrations lately is Microsoft's DebugMCP. It lets agents set and pause on breakpoints, inspect live values, step in/out, evaluate expressions in scope - essentially use all the goodies you get from a proper debugger autonomously.

github.com

GitHub - microsoft/DebugMCP: Gift your VS Code agent a real debugger: breakpoints, stepping, inspection.

Gift your VS Code agent a real debugger: breakpoints, stepping, inspection. - microsoft/DebugMCP

062

Ingvar Stepanyan @rreverser.com · 21/09/2026

Certainly useful for some but I've long been in the other camp (not saying they don't overlap) - I think debuggers are underused and I wish more beginners were taught to work with them. And, thanks to DebugMCP, agents can use them too.

github.com

GitHub - microsoft/DebugMCP: Gift your VS Code agent a real debugger: breakpoints, stepping, inspection.

Gift your VS Code agent a real debugger: breakpoints, stepping, inspection. - microsoft/DebugMCP

010

Reposted by Ingvar Stepanyan

jeffery --dangerously-skip-permissions @jefferyharrell.bsky.social · 18/09/2026

Huh. Seems like CLAUDE.md had a nice run.

Changelog
2.1.277
Added AGENTS.md support: in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead; change it under "Project instructions" in /config (not yet on Bedrock, Vertex or Foundry)

1222

Reposted by Ingvar Stepanyan

Natanael, Tech janitor @natanael.bsky.social · 17/09/2026

"vibe-coders [] submitting code that they do not actually understand." "In Nguyen’s words, “the scene used to be a group of highly talented researchers, but now it is just a bunch of noobs using LLMs"

frvr.com

PS5 Linux lead quits as open-source projects have become “a bunch of noobs using LLMs” that “they don’t even understand”

The lead of the ongoing PS5 Linux project has left the homebrew scene following the wave of AI vibe-coders joining open-source spaces.

163

Reposted by Ingvar Stepanyan

Paweł Grzybek @pawelgrzybek.com · 17/09/2026

This recent article by @bram.us announces a new web component, that is built on top of two superb exciting modern APIs. • CSS Custom Highlight API • OpaqueRange API These two will change the way we build for the web. This post is really worth reading. www.bram.us/2026/09/14/i... #web #css

bram.us

Introducing <rich-input>, a GitHub-like search/filter text input to embed on your site

While working on a side-project to better navigate and search through my record collection, I found myself in need of a rich search input: one that accepts free-form text alongside structured filters ...

0274

Ingvar Stepanyan @rreverser.com · 16/09/2026

I have but I meant that I want to speak to an actual human not... whatever this is.

000

Ingvar Stepanyan @rreverser.com · 16/09/2026

If you write "WTF!!!!!!" with 6 exclamation marks, that costs 3 tokens in OpenAI: "W" + "TF" + "!!!!!!". If you add another one to get "WTF!!!!!!!", that's 4 tokens. But if you add yet another one, then "WTF!!!!!!!!" is back to only 3 tokens again! --- Thank you for coming to my TED talk.

1120

Ingvar Stepanyan @rreverser.com · 16/09/2026

TIL: Object.groupBy and Map.groupBy. developer.mozilla.org/en-US/docs/W...

030