Hacker News Reader

singronon GPT-5.4

The "RPG Game" is hard to judge since it was produced over "multiple turns". The impressive version would be if it basically got a working game on the first attempt, and the prompter gave some follow-ups to tweak feel and style.

However, I think what actually happened is that a skilled engineer made that game using codex. They could have made 100s of prompts after carefully reviewing all source code over hours or days.

The tycoon game is impressive for being made in a single prompt. They include the prompt for this one. They call it "lightly specified", but it's a pretty dense todo list for how to create assets, add many features from RollerCoaster Tycoon, and verify it works. I think it can probably pull a lot of inspiration from pretraining since RCT is an incredibly storied game.

The bridge flyover is hilariously bad. The bridge model ... has so many things wrong with it, the camera path clips into the ground and bridge, and the water and ground are z fighting. It's basically a C homework assignment that a student made in blender. It's impressive that it was able to achieve anything on such a visual task, but the bar is still on the floor. A game designer etc. looking for a prototype might actually prefer to greybox rather than have AI spend an hour making the worst bridge model ever.

singronon A new California law says all operating systems need to have age verification

The whole point of the bill is to create a cause of action for the Attorney General to sue companies. In the bill, they say the damages are up to $2,500 per negligently affected child ($7,500 if intentional), so it doesn't matter how many non-children it affects. E.g. if the OS/appstore/accounts/application is in the context of a workplace that only employs adults, none of this matters.

singronon A new California law says all operating systems need to have age verification

The bill doesn't define "accounts", so it's entirely possible local users that a human signs into would count.

The saving grace is that obviously they have no idea what a Linux distribution is, and only the Attorney General can bring action, so there isn't much risk of the AG suing Debian.

singronon A new California law says all operating systems need to have age verification

"User" in the bill actually means child, so cups etc. don't apply.

singronon Bus stop balancing is fast, cheap, and effective

The crazy people depend a lot on routes, the part of the city, and the time of day. E.g. the 1 (Sacramento St/California St) is basically fine all the time. The 38 (Geary) and 14 (Mission) are OK during the commute rush since they are packed full of commuters, but outside of those times, you will eventually see all kinds of unsocial behavior (shouting, fights, defecation, etc.), especially closer to civic center/tenderloin/mission.

singronon I'm helping my dog vibe code games

I've previously struggled getting LLMs to manipulate tscn/tres files since they like to generate non-unique uids. Despite being text files, the godot tscn/tres files are normally meant to be manipulated by the editor and need to define and reference unique ids. The editor always generates completely random alphanumeric strings, but LLMs like to use names or placeholders (e.g. "aaaaa1", "example", or "foobar") for the ids.

The linter in the article that detects duplicate uids is interesting. Obviously the article is about creating a bunch of harnesses for the LLM to be productive. I wonder how many problems can be transformed like this from something LLMs just can't do reliably to something they just need to burn credits for a while on. The LLM probably can't tell if the games are fun, especially with it's rudimentary playtesting, but who knows.

singronon Using PostgreSQL as a Dead Letter Queue for Event-Driven Systems

Generally yes, but if you use e.g. the parallel consumer, you can potentially keep processing in that partition to avoid head-of-line blocking. There are some downsides to having a very old unprocessed record since it won't advance the consumer group's offset past that record, and it instead keeps track of the individual offsets it has completed beyond it, so you don't want to be in that state indefinitely, but you hope your DLQ eventually succeeds.

But if your DLQ is overloaded, you probably want to slow down or stop since sending a large fraction of your traffic to DLQ is counter productive. E.g. if you are sending 100% of messages to DLQ due to a bug, you should stop processing, fix the bug, and then resume from your normal queue.

singronon Capital One to acquire Brex for $5.15B

Their options should be priced lower, but the common stock isn't valued according to the $5.15B. They raised $300M at $12B and $425M at $7.4B, which are both under water, so those shareholders will use their liquidation preference to get paid at least 1x. Assuming those rounds owned 7% of the company, there is at most $4.4B left for the remaining 93% of shareholders. That's about 8% less. If they deducted fees, legal services, or retention packages or had worse liquidation preferences or more underwater rounds, then it gets even lower.

singronon Capital One to acquire Brex for $5.15B

You have to exercise the options or let them expire. You normally have 10 years not 7, but if a company comes up on 10 years after they issued their first options, they might try a tender offer to buy some employee shares. If your 10 year old "start up" shares can't be sold anywhere, then they probably aren't worth exercising. A company that can't provide liquidity to employees for 10 years will probably never do it.

singronon Capital One to acquire Brex for $5.15B

ISO options have to expire within 10 years of when they are granted. Sometimes companies make them expire earlier than that, so OP might be thinking of options they were granted. E.g. I once had options that expired 30 days after ending employment even thought the ISO requirement is up to 90 days.

singronon Unconventional PostgreSQL Optimizations

PG does reuse plans, but only if you prepare a query and run it more than 5 times on that connection. See plan_cache_mode[0] and the PREPARE docs it links to. This works great on simple queries that run all the time.

It sometimes really stinks on some queries since the generic plan can't "see" the parameter values anymore. E.g. if you have an index on (customer_id, item_id) and run a query where `customer_id = $1 AND item_id = ANY($2)` ($2 is an array parameter), the generic query plan doesn't know how many elements are in the array and can decide to do an elaborate plan like a bitmap index scan instead of a nested loop join. I've seen the generic plan flip-flop in a situation like this and have a >100x load difference.

The plan cache is also per-connection, so you still have to plan a query multiple times. This is another reason why consolidating connections in PG is important.

0: https://www.postgresql.org/docs/current/runtime-config-query...

singronon Cursor's latest “browser experiment” implied success without evidence

It looks like there are two JS backends: quickjs and vm-js (vendor/ecma-rs/vm-js), based on a brief skim of the code. There is some logic to select between the two. I have no idea if either or both of them work.

singronon A 40-line fix eliminated a 400x performance gap

If you are doing continuous profiling, you are probably using a low overhead stack sampling profiler rather than recording every method entry and exit.

singronon The battle to stop clever people betting

You can still make that bet at 10% "yes" for the next year. Previous years had similar patterns, so it's not a reaction to Trump.

To be fair, we don't need to find little green men in a UFO. It's sufficient to e.g. find fossils of extinct microorganisms on Mars, which is a slim possibility that's a goal of the Mars Sample Return mission.

These markets also have low volume at reasonable prices. If you bought $10K of "no" right now for next year, you would only get an 8% return, not 10%. You could execute better trades to get better prices, but the odds also become more sane over the year. The S&P 500 is also up 18% YTD (13% YoY for the last 5) and you can buy as much of that as you want.

singronon Meta is using the Linux scheduler designed for Valve's Steam Deck on its servers

Isn't that the steam linux runtime? Games linked against the runtime many years ago still run on modern distros.

singronon The appropriate amount of effort is zero

This is a completely obvious conclusion with an unexpected definition of "effort" to justify a click-bait title.

singronon Wayland Nvidia

https://gitlab.freedesktop.org/wayland/wayland-protocols/-/t...

This is an incomplete list of protocols that aren't part of core Wayland. Compositors implement additional protocols that aren't even part of this process (e.g. wlr-screencopy-unstable). See the wlroota protocols here: https://gitlab.freedesktop.org/wlroots/wlroots/-/tree/master...

singronon Client-side GPU load balancing with Redis and Lua

They run these clients themselves and the redis instance isn't publically exposed.

It would indeed be very strange to hope your random users coordinate with your client side load balancer. You wouldn't even have to send real traffic. You could just manipulate redis directly to force all the real traffic to go to a single node. DoSing redis itself is also pretty easy.

singronon Advent of Compiler Optimisations 2025

You can use x86_64-v2 or x86_64-v3. Dates are tricky since cpu features aren't included on all SKUs from all manufacturers on a certain date.

singronon Advent of Code 2025

BTW the page mentions Alternate Styles, which is an obscure feature in firefox (View -> Page Styles). If you try it out, you will probably run into [0] and not be able to reset the style. The workaround is to open the page in a different tab, which will go back to the default style.

0: https://bugzilla.mozilla.org/show_bug.cgi?id=1943796

singronon Nearly all UK drivers say headlights are too bright

Having driven in the US and UK, this is a significant difference between the two. In the UK, you might sometimes drive 30 under on a road that is nominally 60 mph. In the US, that road would have a specific posted speed limit that is safe to drive. US roads are also more consistently designed for constant speed or have additional advisory speed limits for curves. You can nearly always drive as fast as the number on the sign unless there is some additional hazard.

singronon PgFirstAid: PostgreSQL function for improving stability and performance

I'm not the author, but I think you could by using UNION ALL instead of temp tables. You could also make a view that just calls this function. I'm not sure why it would matter though.

singronon Collaboration sucks

I think the sports analogy would be passing the ball back and forth in front of the goal instead of shooting. Nearly every time, an athlete with the ball in front of an open goal will take the shot themselves instead of passing. Even in a 2v1 situation, passing is a huge risk and taking the shot has a high likelihood of scoring, so you only want to pass if the defender is leaving your teammate open (usually they split the difference, so it's complicated). Too many passes in a 2v1 guarantees that more defenders will show up and you lose your advantage.

singronon AI adoption in US adds ~900k tons of CO₂ annually, study finds

The paper is based on an estimate of productivity increases per industry due to AI. The highest increases are around 0.2% and most industries are <0.1%. In that world, AI isn't transformative and doesn't accelerate.

In some ways I think this is probably realistic, but it's not compatible with outcomes touted by boosters. They estimate 28 PJ/year, which is only about 0.9 GW. Stargate was planned to build 10 GW of capacity alone, so they can't both be true

singronon Montana becomes first state to enshrine 'right to compute' into law

"lawful" seems like an enormous loophole that makes this seem vacuous. If the government makes what you are doing unlawful, then it can be restricted. How would the government restrict you from doing something lawful in the first place? A bill of attainder? That's already illegal.

singronon Montana becomes first state to enshrine 'right to compute' into law

This is essentially the "strict scrutiny" standard, which governments have to achieve in order to violate your strongest constitutional rights (e.g. 1A). If you don't spell it out, then it might be delegated to a lower standard like "rational basis".

singronon Study identifies weaknesses in how AI systems are evaluated

No. AIME is a test for advanced high schoolers that mostly tests higher level math concepts like algebra and combinatorics. The arithmetic required is basic. All the answers are 3-digit numbers so that judging is objective and automated while making guessing infeasible. You have 12 minutes on average for each question, so even if you are terribly slow at arithmetic, you should still be able to calculate the correct answer if you can perform all the other math.

singronon FAA to restrict commercial rocket launches to overnight hours

Source: https://www.faa.gov/newsroom/FAA-Emergency-Order-11-6-25.pdf

The order also only applies to domestic flights, so observed percentages on flightaware will be lower than those in the order.

singronon FAA to restrict commercial rocket launches to overnight hours

In some ways, they started in 2004. https://en.wikipedia.org/wiki/Next_Generation_Air_Transporta...

There are a ton of details on the page that go into why it is so hard. One reason is that there are a lot of fundamental things to build and deploy before anything can be automated. E.g. before ADS-B (equipment on planes that detects its own location with GPS and automatically broadcasts it), ATC needed to talk to pilots and ask them where they were in a lot of cases. ADS-B has only been required on commercial flights since 2020.

Then it also suffers like every large government software project where a bunch of $100+ million contracts get paid out to private companies while nothing gets built. And it's part of annual appropriations, so funding was unpredictable. It's like working at a software company that had a major layoff or hiring spree every year for the last 20 years. If we could figure out how to run major project, the value to humanity would be enormous.

singronon The state of SIMD in Rust in 2025

There is a GitHub issue that details what's blocking stabilization for a each feature. I've read a few recently and noticed some patterns:

1. A high bar for quality in std

2. Dependencies on other unstable features

3. Known bugs

4. Conflicts with other unstable features

It seems anything that affects trait solving is very complicated and is more likely to have bugs or combine non-trivially with other trait-solving features.

I think there is also some sampling bias. Tons of features get stabilized, but you are much more likely to notice a nightly feature that is unstable for a long time and complex enough to be excited about.