Mon–Fri · 6 AM ET
← All Episodes
EP  • 00:07:16

Show HN: Wikihack – HackerTube, but for Wikipedia articles mentioned on HN | Build or Be Replaced

Today: Show HN: Wikihack – HackerTube, but for Wikipedia articles mentioned on HN | Show HN: OmniDiff – Opensource, super fast, content aware Git difftool, in Rust | Lenovo's 'encrypted' fingerprint USB is just an unencrypted SD card in a case Episode date: 2026-11-02.

Download MP3 →

Transcript

JOSH: It's Monday, November 2nd. This is Build or Be Replaced, powered by ScanBrief.dev. I'm Josh, here with Erik Anderson.
ERIK: Machines are finding bugs faster than humans can fix them. That's the theme today.
JOSH: Stick around. Erik's got an AI pro tip at the end about making a second model prove its bug reports.
[pause]
JOSH: Quick numbers first. ScanBrief scored 63 items across 56 sources this morning.
ERIK: Average relevance 53.5. Quiet Monday, but three of the top stories rhyme with each other, so that's the show.
[pause]
JOSH: Headlines. Somebody built a HackerTube, but for Wikipedia articles. What's that about?
ERIK: Wikihack. It pulls every Wikipedia article linked in Hacker News posts into one feed. It hit 220 points. There's also a sibling called HackerBooks that does the same for books. Turns out the best reading list is whatever nerds argue about in comments.
JOSH: There's a new Git difftool in Rust too?
ERIK: OmniDiff. Open source, content aware, fast. I read diffs all day because PrimeBus throws them at me. If it understands structure instead of just lines, I'm trying it this week.
JOSH: And something called Nibnib?
ERIK: Image compression built for extreme decompression speed. Small idea, useful idea. Compress once, decode a million times. That's the right trade for most real workloads.
[pause]
JOSH: Okay, deep dive. Postgres 19 is late. Why?
ERIK: It was supposed to ship September or October. Now it's sliding into November. The reason is the story. Large-scale LLM-driven code analysis found a surge of bugs, and they had to revert major features late. UPDATE and DELETE FOR PORTION OF got pulled. So did SQL/PGQ.
JOSH: Wait, really? The AI made the release worse?
ERIK: No. The AI made the release honest. Those bugs were already in the code. Nobody had found them yet.
[beat]
ERIK: Postgres is about as conservative as software gets. Decades of review, a careful community, and a model still turned up enough to force reverts. That tells you human review has a ceiling.
JOSH: So is that good news or bad news?
ERIK: Good news, painful timing. Bugs found before release cost a delay. Bugs found after release cost a data loss story. I'll take the delay every time.
JOSH: How does that compare to what you run?
ERIK: Same pattern, smaller scale. PrimeBus has run 2,295 auto-merge attempts since June 5th. 1,507 merged. 788 got blocked by Gandalf, our reviewer. Zero escalated to me.
JOSH: That's a lot of blocks.
ERIK: That's the point. Those 788 aren't failures. That's a second set of eyes saying no before it hit prod. If my blocked count were zero, I'd be worried, not proud.
JOSH: What did Postgres do right, then?
ERIK: They reverted. They didn't ship a half-trusted feature to hit a date. Pulling a feature late is embarrassing, and it's still cheaper than keeping it. Most teams can't make that call. The date has too much gravity.
JOSH: And for a small team using Postgres?
ERIK: Don't panic about 19 being late. Plan around it. If you were going to upgrade for FOR PORTION OF or the graph queries, you're waiting. Boring stable features still work fine. My own fleet state lives on Postgres, and I'd rather wait a month than debug a temporal-table bug at 2 a.m.
[pause]
JOSH: Next one. Lenovo sold a fingerprint USB drive, marketed as encrypted. What happened?
ERIK: Somebody opened it up. It's an unencrypted SD card in a case. The fingerprint reader gates a software prompt. The data underneath sits there in the clear.
JOSH: So you just pull the card?
ERIK: Pretty much. Pull the card, read it like any other card. The security was theater. Fingerprint on the box, nothing on the bits.
[beat]
JOSH: How does a company ship that?
ERIK: Because nobody asked the right question. The spec said "fingerprint authentication." Somebody checked that box. Nobody asked what the authentication actually protects. That's how it goes wrong in software too.
JOSH: Does that show up in your world?
ERIK: Constantly. I do a lot of work where a green dashboard means something passed, and the question is always what exactly passed. PrimeDash shows me 143 services running on Neo right now. Running isn't the same as correct. I've had processes that were up and returning garbage.
JOSH: So what's the rule?
ERIK: Test the claim, not the label. Don't test "is encryption on." Take the thing, try to read it, and see what comes out. Same with an AI-generated feature. Don't ask if it says it works. Try to break it.
JOSH: And if you buy hardware?
ERIK: Assume nothing. Use full-disk encryption you control, like LUKS or BitLocker, and treat vendor crypto as a bonus. If your data matters, the encryption should be something you configured and can verify.
[pause]
JOSH: Last deep dive. Power grids have a weird one-minute fluctuation, and it got solved?
ERIK: Two items on the feed today are the same story. One is a paper about one-minute frequency disturbances from synchronization in power systems. The other is the write-up, "Mystery of 60-Second Fluctuations in Power Grids Solved." Grid frequency has a pattern that repeats every minute. People saw it for years without a clean explanation.
JOSH: Why does a network automation guy care about the power grid?
ERIK: Because it's a distributed system with a clock problem. A lot of independent actors, all trying to stay in sync, and their coordination leaves a signature in the data. From the title, the cause is synchronization itself. Everything moving on the same schedule at the same moment creates a ripple.
JOSH: That sounds familiar.
ERIK: Every ops person has seen it. The cron job that fires at the top of the minute across 50 machines. A retry storm. Everything lining up on the same clock edge. You see a sawtooth in the graph and spend a week blaming the wrong thing.
JOSH: Do you have an example?
ERIK: Sure. I run a lot of scheduled work. If every service checks in on the minute, you get a spike. So jitter matters. A random few seconds of offset costs nothing and flattens the load. The power industry is learning the same lesson at national scale.
[beat]
JOSH: So the takeaway is look for periodic patterns?
ERIK: Yes. If your graph has a repeating shape, something is synchronized that shouldn't be. Find it before it becomes the incident. The pattern is a clue, not noise.
[pause]
ERIK: This episode is sponsored by Prime Automation Solutions. If you're still doing it manually, we automate it. Also, special on a website — $250. primeautomationsolutions.com
[pause]
JOSH: Alright, what's the AI pro tip today?
ERIK: This one comes straight from the Postgres story. When you use a model to review code, don't accept a bug report. Demand a failing test.
ERIK: Open a fresh session with no context from the one that wrote the code. Give it the diff and tell it: find real bugs, and for each one, write a test that fails right now. No test, no claim.
ERIK: Then run the tests. Passing tests mean the claim was wrong, so throw it out. Failing tests are real bugs with a repro attached.
ERIK: That kills the false positives that waste your time, and it hands you the regression test for free. Models are great at suspicion and weak at proof. Make them do the proof.
ERIK: That's your tip. Use it.
[pause]
ERIK: That tip is straight out of The Autonomous Engineer, my book on building systems that run themselves. Grab it on Amazon.
[pause]
JOSH: One more thing — we started a Discord for builders. If you're shipping AI, automation, or anything that makes a human obsolete — come hang out. Link at buildorbereplaced.dev.
ERIK: Post what you built. We'll post what we're building. Real wins, real builds, no fluff.
[pause]
ERIK: Build or be replaced.
JOSH: If you want these signals in your inbox every morning, scanbrief.dev. See you tomorrow.