Hacker News: dreis_sw

New comment by dreis_sw in "How many of the 170k English words do you know?"

dreis_sw — Sat, 20 Jun 2026 11:03:37 +0000

I found a big problem with this - I noticed that the longest answer is very often the correct one, which kinda ruined the game. Even though I didn't want it to, it started affecting my decision-making. Luckily, I only noticed this around question 85, though those are really the tricky ones.

Good news for the project is that I think you can easily tweak the LLM to generate better alternatives.

I got 89/100, which extrapolates to 72,700. As a non-native speaker, I'm quite happy with that.

New comment by dreis_sw in "Claude Fable is relentlessly proactive"

dreis_sw — Sat, 13 Jun 2026 10:36:27 +0000

It's well known that companies with an abundance of raw technical skills but poor judgement tend to fail. On the technical side technical debt accumulates, while on the business side the wrong choices are made. I think it's valid to generalize this to AI.

New comment by dreis_sw in "Claude Fable is relentlessly proactive"

dreis_sw — Fri, 12 Jun 2026 14:45:29 +0000

I think the new high effort settings are so strong that selecting them when the task doesn't require it actually impacts the output negatively.

New comment by dreis_sw in "Claude Fable is relentlessly proactive"

dreis_sw — Fri, 12 Jun 2026 14:42:42 +0000

Seems like this model delivers on what has already been scaling quite nicely, which is the length and complexity of the requested tasks, but isn't such a big improvement on what hasn't been scaling so far - common sense, discernment, good judgement.

Show HN: GitHub Copilot port of Anthropic's AI vulnerability discovery harness

dreis_sw — Mon, 08 Jun 2026 14:05:33 +0000

Last week, Anthropic released https://github.com/anthropics/defending-code-reference-harne..., a reference harness for autonomous vulnerability discovery that uses Claude Code agents to find, verify, and patch memory-safety bugs. I wanted to use it but I only have access to GitHub Copilot.

This is a port of that harness to the GitHub Copilot CLI. PORTING-PLAN.md covers the decisions made to map the handful of features that work differently between Claude Code and the Copilot CLI.

The result is a working reference for anyone who wants to build autonomous security agents on Copilot, tracking Anthropic's approach as closely as possible.

Feedback welcome!

Comments URL: https://news.ycombinator.com/item?id=48445559

Points: 2

# Comments: 0

New comment by dreis_sw in "Spain blocks prediction markets Polymarket, Kalshi over lack of gambling licence"

dreis_sw — Tue, 26 May 2026 20:00:17 +0000

They are certainly critically flawed at the moment, but I still feel that with proper traceability, KYC, etc. it would be a much fairer system than the current betting companies. It’s one of those use-cases where blockchain technology actually makes sense.

New comment by dreis_sw in "Computer Use in Codex [video]"

dreis_sw — Tue, 12 May 2026 22:09:35 +0000

Wonder if that's also the case for web-based UI like Spotify, or if the accessibility APIs are only useful for native applications.

New comment by dreis_sw in "Where the goblins came from"

dreis_sw — Thu, 30 Apr 2026 12:35:12 +0000

History shows continuous evolution, there won't be a "final AGI thing". The definition of AGI is so vague anyways that any conversation around it is hardly useful. 5 years ago, what we have today would have been considered AGI.

New comment by dreis_sw in "Why AI Sucks at Front End"

dreis_sw — Mon, 13 Apr 2026 10:29:40 +0000

If the economy wasn't completely broken you'd have a job right now.

New comment by dreis_sw in "Project Glasswing: Securing critical software for the AI era"

dreis_sw — Tue, 07 Apr 2026 18:47:47 +0000

It also took many years to put capable computers in the hands of the general public, but it eventually happened. I believe the same will happen here, we're just in the Mainframe era of AI.

AI Coding Tools Are Stuck in 2007's Dev UX

dreis_sw — Tue, 24 Mar 2026 00:36:21 +0000

Article URL: https://www.davidreis.me/2026/ai-coding-tools-stuck-in-2007s-dev-ux

Comments URL: https://news.ycombinator.com/item?id=47497209

Points: 5

# Comments: 1

New comment by dreis_sw in "No, it doesn't cost Anthropic $5k per Claude Code user"

dreis_sw — Tue, 10 Mar 2026 16:55:54 +0000

We'll only know for sure when either of these companies goes public. Google serves inference and Google Cloud is profitable but we don't know how much inference is costing.

If they never go public, there's our answer as well.

Why Your AI Coding Agent Gets Worse over Time (and How to Fix It)

dreis_sw — Tue, 10 Mar 2026 09:52:20 +0000

Article URL: https://www.davidreis.me/2026/why-your-ai-coding-agent-gets-worse-over-time

Comments URL: https://news.ycombinator.com/item?id=47321092

Points: 3

# Comments: 0

New comment by dreis_sw in "Ladybird adopts Rust, with help from AI"

dreis_sw — Mon, 23 Feb 2026 21:27:06 +0000

Works even better if you have a good test suite, which is surely the case here with Ladybird

New comment by dreis_sw in "Instagram chief orders staff back to the office five days a week in 2026"

dreis_sw — Wed, 03 Dec 2025 22:53:33 +0000

Happy to hear that it's working out well for you, but there is always a cost. The less common successful remote work becomes across the industry, the more companies like yours stand out. And remote work goes back to being a premium, highly competitive perk, like it used to be in the past.

And I agree that remote work is perfectly possible with a good culture and responsible employees, but most companies don't really have the culture or hiring practices that optimize for that.

New comment by dreis_sw in "Instagram chief orders staff back to the office five days a week in 2026"

dreis_sw — Wed, 03 Dec 2025 16:22:42 +0000

(2) At the office, it's harder to pretend that work took longer than it actually took. If you're done with the design doc in 2h, with colleagues sitting all around you, what can you do to pretend that it took you longer? Mess around with Confluence, pretending to work? You might as well move on to the next design doc, or find something useful to do meanwhile.

At home, you can just pick up the Xbox controller and say it took you 4 hours.

I like the comfort of WFH, and there are plenty of people (myself included) that are responsible enough to handle it, but the problem is that the average person isn't. And we all pay for it.

New comment by dreis_sw in "Claude Opus 4.5"

dreis_sw — Mon, 24 Nov 2025 21:14:33 +0000

I agree with your sentiment, this incremental evolution is getting difficult to feel when working with code, especially with large enterprise codebases. I would say that for the vast majority of tasks there is a much bigger gap on tooling than on foundational model capability.

New comment by dreis_sw in "Nano Banana can be prompt engineered for nuanced AI image generation"

dreis_sw — Fri, 14 Nov 2025 14:50:20 +0000

So the watermark is being added to the image on the client-side? That's pretty bad

New comment by dreis_sw in "Thieves steal crown jewels in 4 minutes from Louvre Museum"

dreis_sw — Mon, 20 Oct 2025 11:35:09 +0000

How do you break the window glass, how do you get into the display cases, how and when do you do it so that you're not stopped by the security, how do you carry the items away, how do you get away from the crime scene without being caught... And I'm sure I am forgetting other details. So I can see how it would take some work to make everything look so simple.

New comment by dreis_sw in "Show HN: Unregistry – “docker push” directly to servers without a registry"

dreis_sw — Thu, 19 Jun 2025 16:03:24 +0000

I recommend using GitHub's registry, ghcr.io, with GitHub Actions.

I invested just 20 minutes to setup a .yaml workflow that builds and pushes an image to my private registry on ghcr.io, and 5 minutes to allow my server to pull images from it.

It's a very practical setup.