HNHacker News
TopNewBestAskShowJobs

pbronez

2,117 karma · joined May 10, 2019

[ my public key: https://keybase.io/peterbronez; my proof: https://keybase.io/peterbronez/sigs/aK3o2l59iuXf4ZE61-arwlc3CsPE1jYuB6pu_vjfi0c ]

<a rel="me" href="https://hachyderm.io/@PeterBronez">Mastodon</a>

submissionscomments
pbronez··on Ask HN: What are you working on? (October 2025)
I think this concept has legs to be much bigger than just foods. There are lots of influencer types who focus on testing.

For example, there are two individuals who own the same $100k machine for testing the performance of loudspeakers.

https://www.audiosciencereview.com/forum/index.php

https://www.erinsaudiocorner.com/

Both of them do measurements and YouTube videos. Neither one has a particularly good index of their completed reviews, let alone tools to compare the data.

I wish I could subscribe to support a domain like “loud speaker spin tests” and then have my donation paid out to these reviewers based on them publishing new high quality reviews with good data that is published to a common store.

pbronez··on Macro Splats 2025
When was that made? The YouTube video is 14 years old but it feels at least a decade older than that.
pbronez··on Macro Splats 2025
The linked paper describes a pipeline that starts with “point cloud from SfM” so they’re assuming away this problem at the moment.

Is it possible to handle SfM out of band? For example, by precisely measuring the location and orientation of the camera?

The paper’s pipeline includes a stage that identifies the in-focus area of an image. Perhaps you could use that to partition the input images. Exclusively use the in-focus areas for SfM, perhaps supplemented by out of band POV information, then leverage the whole image for training the splat.

Overall this seems like a slow journey to building end-to-end model pipelines. We’ve seen that in a few other domains, such as translation. It’s interesting to see when specialized algorithms are appropriate and when a unified neural pipeline works better. I think the main determinant is how much benefit there is to sharing information between stages.

pbronez··on Does our “need for speed” make our wi-fi suck?
It’s configurable
pbronez··on Does our “need for speed” make our wi-fi suck?
If you’re the kind of person who wants better than gigabit Ethernet, it’s very available. 2.5Gbe is just a USB adapter away. Mac Studio comes with 10GbE. Unifi networking gives you managed multi-gig and plenty of others do unmanaged multigig at affordable prices. Piles of consumer NAS support multigig.

I think this market is driven by content creators. Lots of prosumers shoot terabytes of video on a weekly basis. Local NAS are essential and multi-gig local networks dramatically improve the editing experience.

pbronez··on Synology reverses policy banning third-party HDDs
Yeah that’s a really interesting gap in the feature set. Ubiquiti’s new offerings seem like pretty good network storage appliances, but not home servers.

I’ve gone through a couple iterations of home server. First I upcycled an obsolete Dell Power edge. Then a N100 mini PC. Neither was as reliable as I wanted.

Now I’m running persistent apps on Railway and compute hungry stuff on my MacStudio. Pretty good so far.

pbronez··on Synology reverses policy banning third-party HDDs
Ubiquiti smelled the blood in the water and released a whole new NAS product line. They don’t run arbitrary apps but for basic storage on the network they look pretty solid.
pbronez··on What makes 5% of AI agents work in production?
The problem is that precision is expensive. Writing is thinking. Writing software is defining the business problem.

People don't know exactly what they want from the data warehouse, just a fuzzy approximation of it. You need stochastic software (AI) to map the imprecise instructions from your users to precise instructions the warehouse can handle.

pbronez··on ProofOfThought: LLM-based reasoning using Z3 theorem proving
That’s pretty cool. It would be super useful to identify contradictory guidance systematically.
pbronez··on Traefik's 10-year anniversary
Excellent feedback! This is valuable advice for any project.
pbronez··on Typst: A Possible LaTeX Replacement
You know, this is good fodder for LLM training via Reinforcement Learning with Verifiable Results. Given complex LaTeX source code and complex PDF output, write. Typst document that reproduces the PDF.

Incorporating this into training would be a good way to improve LLM Typst support and also verify that reasoning is working (since there aren’t tons of typst examples to copy from)

pbronez··on Typst: A Possible LaTeX Replacement
AI support is an interesting new facet of ecosystem quality. I wonder how much of this can be obviated with high quality tools. Coding agents can leverage language servers and documentation fairly well, but it’s not clear to me if that can fully compensate for weak language coverage in the training data.
pbronez··on Welcome to Cognitive Capitalism
Yes, BUT, I want very strong guarantees that any such agents are loyal to me, the enterprise customer, first and foremost.

If anything, companies have MORE to lose than individuals here. If ChatGPT can anticipate your corporate decisions, then it can front run you and steal your whole business.

This absolutely happens:

- Amazon watches for successful third party products, then releases clones under their “Amazon Basics” brand and crushes the original.

- Robinhood makes most of their money selling real-time activity data to market makers, ensuring that big financial firms win over retail investors every time.

If OpenAI manages to do this in the general case… well then the valuations start to make a lot more sense.

pbronez··on Redis is fast – I'll cache in Postgres
How was the LLM accessing the docs? I’m not sure what the best pattern is for this.

You can put the relevant docs in your prompt, add them to a workspace/project, deploy a docs-focused MCP server, or even fine-tune a model for a specific tool or ecosystem.

pbronez··on YAML document from hell (2023)
The author of this article created RDL:

> RCL is a domain-specific language for generating configuration files and querying json documents. It extends json into a simple, gradually typed, functional programming language that resembles Python and Nix.

https://github.com/ruuda/rcl

https://rcl-lang.org

pbronez··on Why haven't local-first apps become popular?
Which sync engine are you using?
pbronez··on Why haven't local-first apps become popular?
There's a bit of No True Scottsman happening here. Isn't every iOS app backed by iCloud's CloudKit [1] a "Local First App"?

[1] https://developer.apple.com/icloud/cloudkit/

pbronez··on Why haven't local-first apps become popular?
This is content marketing for a open-source SQLite extension called `sqlite-sync` [0]. It seems pretty cool, implementing CRDTs and some networking directly in SQLite.

Unfortunately this extension "is tightly integrated with SQLite Cloud." It doesn't support peer-to-peer syncing or self-hosting a central source of truth. The project is really just a client for the sqlite.ai cloud service.

[0] https://github.com/sqliteai/sqlite-sync

pbronez··on You did this with an AI and you do not understand what you're doing here
How far did they get? Did they solve the problem?
pbronez··on Spectral Labs releases SGS-1: the first generative model for structured CAD
I really appreciate OpenSCAD. It’s my go-to whenever I need to quickly define a basic shape to 3D print. Unfortunately I predictably hit a wall every time the complexity increases beyond the basics.

The kernel is not robust to the stupid things I naively ask it to do. Often my code makes sense to me, but OpenSCAD refuses to create an object.

Performance falls off a cliff. You can work around it by pausing previews and adjusting resolution, but that’s a big UX compromise.

Still, I’ve tried a few other options and keep cycling back to OpenSCAD. The barrier to entry is very low, coding AI does a pretty good job, and there’s a decent ecosystem of community modules.

pbronez··on Removing newlines in FASTA file increases ZSTD compression ratio by 10x
Seems useful for games marketplaces like Steam and Xbox. You control the CDN and client, so you can use tricky but effective compression settings all day long.
pbronez··on YouTube is a mysterious monopoly
Mux is new to me. Looks like a video-first headless CMS with some neat AI integrations.

Vimeo does have monetization tools [1] but they’re focused on direct sales.

YouTube is just way ahead… even if you ignore the ads platform, a YouTube premium subscription gives you WAY more ad free content than a Vimeo purchase or Floatplane/Nebula subscription.

[1] https://vimeo.com/solutions/video-monetization

pbronez··on YouTube is a mysterious monopoly
I wonder how YouTube thinks about that.

They probably would prefer zero third party sponsors, because adding sponsored content dilutes the value of the on-platform ads. Features like “commonly skipped section” and the timeline view intensity histogram reduce the value of sponsorships.

But if they eliminated sponsors, creator revenue would drop significantly and so would content production.

The nuclear option would be to require all sponsored segments to register with YouTube. That would give YouTube way more control and dramatically reduce creators’ business flexibility (how do you tax a donated 3d printer?).

pbronez··on iPhone Air
I’ve been impressed with my Thread setup. It’s fast and reliable. Don’t have to think about it, just works.

I mostly use Eve devices. My AppleTV serves as the hub and I keep it plugged into Ethernet. HomePods scattered about keep the mesh strong.

pbronez··on OrioleDB Patent: now freely available to the Postgres community
In this case, it seems to refer to their support for S3-compatible object storage as for persistence.
pbronez··on OrioleDB Patent: now freely available to the Postgres community
OrioleDB is new to me.

According to the docs, it “uses Postgres Table Access Method (TAM) to provide a pluggable storage engine for PostgreSQL. […] Pluggable Storage gives developers the ability to use different storage engines for different tables within the same database. Developers will be able to choose a storage method that is optimized for their specific needs: some tables could be configured for high transactional loads, others for analytics workloads, and still others for archiving.”

https://www.orioledb.com/docs

pbronez··on America Surrenders in the Global Information Wars
We’re just pushing into meta-modernism, and we need to get there faster.

Basically, the cultural journey is: naive idealism -> ironic skepticism -> balanced synthesis.

More concretely: first we told stories about how good guys were good and trustworthy. Then we told stories about how good guys could be corrupted. Now we’re trying to tell stories where good guys exist despite the possibility of corruption.

It’s useless to believe that everything is so complex maybe good and bad aren’t meaningful. We need the courage to take a moral stand amidst complexity.

https://en.m.wikipedia.org/wiki/Metamodernism

https://youtu.be/5xEi8qg266g

pbronez··on Why RDF is the natural knowledge layer for AI systems
Also worth checking out schema.org, which provides a baseline ontology that pops up in many places

https://schema.org/Thing

pbronez··on Why RDF is the natural knowledge layer for AI systems
Which company? I need to build an enterprise knowledge graph.
pbronez··on Fil's Unbelievable Garbage Collector
Maybe there's a world where we successfully get CPython re-written in Rust and that RustyPython becomes Python 4.
← PreviousPage 7 of 34Next →