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 The rise and fall of D&D in Japan
Three options.

First, see if you can get your friends to engage with a one-shot that you DM. You can knock it out in an afternoon and they’ll get a feel for it.

Second, find a local hobby store and see if they host a group you could join. Most places that sell D&D books also host games.

Third, check out Pathfinder. It’s a different TTRPG with many similarities to D&D, but with affordances for playing with a wide variety of people instead of a single tight group. Specifically, see if there is a “Pathfinder Society” in your area — that’s the overall structure they use to organize games and track character progression across games.

pbronez··on Monogon: A Linux userland in pure Go
This bears passing resemblance to Aurae [0]. Both projects are trying to reimagine the space between the Linux kernel and the cluster using modern languages for improved safety. Beyond that, I can’t confidently compare and contrast the projects at this time.

[0] https://aurae.io/

pbronez··on Launch HN: Aqua Voice (YC W24) – Voice-driven text editor
What’s the problem with the AirPods? Too much pre-processing?
pbronez··on Moirai: A time series foundation model for universal forecasting
References this paper on Time Series transformers. First I’ve seen someone apply transformers to time series specifically. Very curious how well this might work for low-frequency events.

https://arxiv.org/abs/2402.02592

pbronez··on GPT-4, without specialized training, beat a GPT-3.5 class model that cost $10M
I can’t access the article there… SSL error and then timeout. Here’s a link to the most recent WayBackMachine snapshot:

https://web.archive.org/web/20240321091803/https://www.incom...

pbronez··on A cargo ship's 'WindWing' sails saved it up to 12 tons of fuel per day
My thought exactly. Some university should build a wing sail ship using exclusively materials and tools available during the Golden age of sail.
pbronez··on 20 Years of "Not Even Wrong"
I think it actually runs the other way. Because Physics is in the doldrums, intelligent people trained in Physics show themselves out and start building those advertising systems.

Personally I know many people who abandoned Physics to reinvent themselves as data scientists. Good for them that they were able to re-purpose their training; that doesn’t mean physics training is the best way to become a data scientist. My bias is that quantitative social science is a better route, although some of what it gains from focusing on people is often lost through less rigorous mathematics and software development requirements.

pbronez··on Designing a Pure Python Web Framework
Would you consider it “pure Python” if it targeted a WASM runtime for the browser side component?
pbronez··on Designing a Pure Python Web Framework
Dropbase looks neat. Couple questions based on my read of the site/github/docs:

- is the architecture exclusively hybrid SaaS? Looks like I host a worker node but a cloud connection is still required.

- how can I monitor the data flow between the local worker and hosted orchestrator? I assume it’s straightforward to turn on verbose logs on the worker’s requests out.

- is the IDE hosted locally or remotely?

- authentication to local services appears to be done primarily through credentials hardcoded in the ENV file. How can I use SSO and pass user authentication to the upstream data sources?

pbronez··on 20 Years of "Not Even Wrong"
From the outside, this seems to rhyme with epicycles. You’re missing a core concept, so you pile on complexity to paper over the failed predictions.

The industrial organization failure is understandable. It’s all well and good to say “they should have recognized it wasn’t working and stopped” … but even if that’s true at the macro level, it’s very hard on an individual level.

An established physicist working on string theory has every incentive to keep building on that line of research. They have no competitive advantage switching to a new line. The specifics about how much expertise translate how far is beyond me, but it seems that nearby better ideas would have been found given the resources poured into this.

This is exactly The Innovator’s Dilemma. It’s even worse in academia than industry because the whole sector is so insular. I look forward to the reinvention of higher education and theoretical research…

pbronez··on Suno, an AI music generator
I appreciate Endel for this reason. When you want something predictable, but in a particular vibe, algorithmic music nails it.

https://endel.io/

pbronez··on Losing faith in testing
It’s tricky to balance Doing Stuff and Broadcasting Stuff. You need both. If you spend 100% of your time heads down, even delivering huge value, nobody will notice and your career will suffer. If you spend 100% of your time bloviating, you’ll never build anything of value. As so often in life, I find the best solution is to team up. Partnerships between an 80 build / 20 tell and a 20 Build / 80 Tell change the world. Woz + Jobs is just the best known example. The book Rocket Fuel calls these “Integrators” and “Visionaries” https://www.eosworldwide.com/rocket-fuel-book
pbronez··on How web bloat impacts users with slow devices
Read it on iOS Safari, without reader mode. Worked great.

Only thing that annoyed me is that there are very lengthy appendices. Thus the scroll bar suggests the main article is much longer than it actually is.

pbronez··on Spreadsheets are all you need
Excel is getting Python “embedded” soon. Might make some of this easier.
pbronez··on Show HN: Creating custom coloring pages from photos. Great for parents/teachers
Thanks for calling this out. I think the dev cycle for production AI Systems will involve a lot of bug squashing on this kind of thing. These models handle a huge variety of inputs; by definition much more than you could hand code. It’s hard to even identify systematic failures.

Once a failure mode is known - like here - how do you fix it? The foundational problem is minorities are a minority of the training set. Good training data is expensive, so how can we practically boost representation without getting the Gemini fiasco?

One idea: start with the best coloring books. Presumably some human has mastered the art of “accurately representing black or mixed hair without it turning into a caricature”. Find them, and start buying their art. When they have drawings based on photos, use that training pair. When they have line art only, use a style transfer tool (like this one!) to convert to a photo. That gives you another pair for use in the other direction.

Another idea: make it easy to users to flag poor transfers. Add a “whitewashed” option to the standard mod report flow. Feed that into RLHF. Get better.

Another idea: focus dev cycles on this. Once you have a system to flag problems and address them, use all this fancy AI to identify input photos with black/mixed hair. Manually inspect how the model performs and push feedback into the next training cycle.

One tool for this is https://www.aquariumlearning.com/

The point is - once you identify a corner of feature space where things break, you can fix them.

pbronez··on Show HN: Digital Superpowers, a free book highlighting various FOSS tools
Disagree. Explain folders.

Lots of people have only experienced computing through mobile and tablet devices that try to hide the file system. They need to be taught the concept of a file. File types. Folders. Permissions. Drives. File transfer between computers.

pbronez··on Show HN: Digital Superpowers, a free book highlighting various FOSS tools
Glanced through the table of contents, this is a treasure. It’s distilled digital acumen. Will contemplate how to use it as a curriculum for my school age children.
pbronez··on Kagi and Wolfram
We have both available as a result of independent & simultaneous innovation/20%/side project things. I’m biased because I own the Kagi side, but have thought about this a bit.

SearX has some advantages. It’s free, except for the compute and labor costs. It runs on our infrastructure, so SSO is easy and we have complete visibility on what’s leaving our perimeter.

Ignoring nuanced search quality evaluation, SearX main disadvantages are (1) it’s only available on our network, so if you set it as the default search that feature breaks if you’re not at the office or on VPN and (2) it’s slow. Users get a blank page for a meaningful moment before results show up.

Kagi is snappy AF and adds solid AI features. Their Universal Summarizer (which isn’t actually LLM powered) works well and the fastGPT is the new standard for open ended information retrieval IMHO. (I have a Perplexity subscription I don’t bother to use)

What I want from Kagi is (1) a single payer team plan that gets all the features, (2) more information & contractual guarantees about privacy. I argue that Google is a terrible default, but the brand is trusted while Kagi is still mostly unknown.

Now, the real benefit of SearX would be integrated custom search providers for in-house data sources. We haven’t done that, and I’m not confident we have the resources to do it well. This is also something I’d love to get from Kagi, but it’s not well-aligned with their current roadmap. I’m content to use them to search external data and pursue other solutions for internal stuff.

pbronez··on Kagi and Wolfram
Ugh I can’t wait for Apple to roll out Siri II, re-baselined with contemporary LLM tech. So much of Siri’s architecture is exactly what I want… it’s just too dumb.
pbronez··on Kagi and Wolfram
I also spun up a Kagi account for work. The current Teams product is fine, but hasn’t been a priority for the team and lacks features available in the Ultimate plan.

I’ve wound up moving folks over to individual Ultimate accounts. They’re working on a revised one payer/many users product. Once that’s available I’ll consolidate everyone there and experiment with making it the default search experience in our managed browsers.

pbronez··on Kagi and Wolfram
That would be pretty great. Even better - a movie widget that integrates IMDB and JustWatch. JustWatch tells you where you can actually stream/buy/rent a movie… I find it very helpful for dealing with ever-changing streaming catalogs.

https://apis.justwatch.com/docs/content_partner/

pbronez··on Kagi and Wolfram
Do you find that works better than using Thangs directly? It’s a meta-search engine itself.
pbronez··on Ace of Aces: Why you should do maths as a game designer
I’d love to try this - you should write up the project!
pbronez··on Radicle: Open-Source, Peer-to-Peer, GitHub Alternative
Maybe Kagi could add this to their custom index.
pbronez··on Radicle: Open-Source, Peer-to-Peer, GitHub Alternative
Yeah that’s been my experience with IPFS. Very cool idea, practically doesn’t work very well. Haven’t tried recently though, maybe it’s improved.
pbronez··on Radicle: Open-Source, Peer-to-Peer, GitHub Alternative
Is there nice interop with BitTorrent for those cases, similar to how Git Annex adds large binary support to git?

For example, if I use Radicle to version a machine learning project, can I use a Magnet link for multi-GB model files?

pbronez··on Ask HN: How can I learn about performance optimization?
Mature Optimization by Carlos Bueno https://carlos.bueno.org/optimization/
pbronez··on An Unbelievable Demo (2021)
This is indicative of a more general phenomenon: the world is big enough that attention is stretched thin. If you invest a few weeks of thoughtful work into a specific corner, just because YOU care, you can quickly surpass broader experts.

This doesn’t devalue broad expertise in general. Focused amateurs benefit significantly from access to experts, and you need some level of training/experience before you can pull this off. The point is that attention and persistence are valuable.

Consider someone with human body. There’s a lot going on. You can go to a doctor. You can go to specialists. It takes weeks and months to get seen and you get minutes of their time. If you have a complex condition, it’s very hard to find and capture the attention of an expert who can address the breadth of the problem.

You have to get informed. Not the “do your own research” trope that leads to anti-vax and flat-earth nonsense. You have to document your symptoms. Figure out how to tell your story concisely. Learn what the relevant specialties are and the language they use. Dig into primary sources for early studies and experimental interventions. Nobody else will do this work for you. For much of it, nobody else can - because only you have direct experience of the problem.

The software ecosystem is the same. It’s huge, interconnected, with silo’d experts and communities of practice. A competent, motivated person can chart unique, rewarding paths. Keep pushing!

pbronez··on Wintergatan Marble Machine (2016) [video]
He actually partnered with an agency and released a virtual marble machine as an app and VST
pbronez··on Open Collective Official Statement – OCF Dissolution
Key quote:

“”” over the last three years, the source of funds on the platform shifted majorly from small donations to added funds. A mechanism that was originally an unplanned “add-on” became the main source of our revenue, shifting our incentives and, slowly, our strategy, in a way that we did not plan for, and I failed to predict. As a result of this movement to direct-to-hosts large grants and funds (with their increased compliance and opex); the cost to hosting collectives while equally sharing the fee on the transaction became untenable for the fiscal hosts. “””

← PreviousPage 18 of 34Next →