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 Can I run AI locally?
The docs page addresses this:

> A Mixture of Experts model splits its parameters into groups called "experts." On each token, only a few experts are active — for example, Mixtral 8x7B has 46.7B total parameters but only activates ~12.9B per token. This means you get the quality of a larger model with the speed of a smaller one. The tradeoff: the full model still needs to fit in memory, even though only part of it runs at inference time.

> A dense model activates all its parameters for every token — what you see is what you get. A MoE model has more total parameters but only uses a subset per token. Dense models are simpler and more predictable in terms of memory/speed. MoE models can punch above their weight in quality but need more VRAM than their active parameter count suggests.

https://www.canirun.ai/docs

pbronez··on Big Data on the Cheapest MacBook
How did you get one already? I thought they were just up for pre-order
pbronez··on Ki Editor - an editor that operates on the AST
I had an experience like this when I switched to a Keyboardio split keyboard. It was impossible to use at first. Then I do some deliberate practice with a typing tutor app and now I love it so much I own multiple.

Is there a good tutorial for some of these advanced text editing features?

In particular I’d like to get platform independent shortcuts / key bindings. I use both windows and MacOS daily, and it throws off my muscle memory for shortcuts like “go to beginning of line”

pbronez··on Making Wolfram tech available as a foundation tool for LLM systems
I wonder how this will compare, long term, to giving LLMs python sandboxes. Why implement an MCP server for a single library when you can give the LLM an interpreter running a distribution with arbitrarily many libraries?

Probably the trick is teaching the LLM how to use everything in that distribution. It’s not clear to me how much metadata that SymPy MCP server bakes in to hint the LLM about when it might want symbolic mathematics, but it’s definitely gonna be more than “sympy is available to import”

Also, reading through TFA, Wolfram is offering more than a programming language. It includes a lot of structured general purpose information. I suspect that increases response quality relative to web search, at least for a narrow set of topics, but I’m not sure how much.

pbronez··on Ladybird adopts Rust, with help from AI
Interesting take, what do you think comes next? A programming language optimized for coding agents?
pbronez··on Show HN: A native macOS client for Hacker News, built with SwiftUI
Well, I suppose the app could offer a proxy service. Funnel everything through a VPS, apply ad and tracker blocking there.

That opens the door to lots of additional features… Cache responses so you can still read stuff when it gets the HN hug of death. Do a full-text index and offer a secondary search capability over article contents. Maybe build an API for all that so you can have AI Agents ground themselves on articles that got strong quality signals on HN. Maybe sign agreements with publishers like LWN, The Information, or whoever else shows up on HN behind a paywall frequently.

Obviously that would need to be a paid feature.

pbronez··on I tried building my startup entirely on European infrastructure
Are you actually using Exo for local clustered AI inference? I’ve considered it a few times and keep finding horror stories. Never seen someone report it’s actually working well for them.
pbronez··on Claude Sonnet 4.6
That’s the big question on my mind. Several years ago we migrated from an in house Ruby on Rails solution to Salesforce. Will vibe coding bring us back to a custom solution? When?
pbronez··on Tesla Sales Down 55% UK, 58% Spain, 59% Germany, 81% Netherlands, 93% Norway
Geely is also present in the US via Volvo and Polestar. They haven’t delivered an affordable entry level EV yet, but I’ve been impressed with the Polestars. I’d never buy one because a Chinese car running on Google software sounds like a nightmare, but they’ve been good when I rented them.
pbronez··on Suicide Linux (2009)
That one actually looks useful!
pbronez··on Kimi Claw
Yes. Kimi is a brand of Moonshot Ai. Moonshot is legally incorporated in Singapore; executives and devs are mostly in Beijing.

They reportedly use Alibaba cloud extensively, at least for training. Terms of Service say the service is governed by Singaporean law.

Personally, I’d assume the CCP has full access to every packet you send them. K series of models looks cool; you can run they M through Azure if you prefer to do business with a western entity or self host.

pbronez··on An AI agent published a hit piece on me – more things have happened
Look forward to seeing their assessment.
pbronez··on Show HN: I spent 3 years reverse-engineering a 40 yo stock market sim from 1986
1) this is super interesting

2) the prose is very LLM-flavored, but for once I don’t mind. It’s obvious to me that this blog post only exists because an interesting, intelligent person used an LLM to lower the cost of extracting complex ideas from their brain.

3) the “Fits of Rationality” process sounds a lot like how coding agents work. They’re alive in the moment, then later all they have to work with is the written record of their previous sessions. The Herculean effort to re-implement the result foreshadows the vibe-coding repair industry.

pbronez··on 4chan for Clankers
Mine blew through $50 of OpenRouter credits in a single day. Not worth it.

Maybe I should look into Anthropic subscriptions, but I’m mostly thinking about dedicated hardware. A used Mac Studio M1 Ultra can has I impressive memory bandwidth…

pbronez··on Ship Types, Not Docs
Huh I thought the canonical reference for this was https://docs.divio.com/documentation-system/

But yeah the “Reference + Explanation + Tutorials + How-To Guides” concept is the best one I’ve come across for documenting technical systems. I’m planning to lean into it heavily for my next project.

pbronez··on [dead]
Agent sandboxing is an important problem. I want to deploy agentic AI to everyone at my enterprise and security is the big blocker.

Credential scanning seems tractable. There’s a large body of work around scanning for credentials in repos to prevent leakage via GitHub.

If speed really matters, you could minimize the patterns you check by integrating credential management. By definition, you know all the secrets you’re trying to protect. Look for _exactly_ those rather than regex which try to enumerate the general case.

Still, solving credential leakage is necessary but not sufficient. There’s other sensitive information in your context: customer contact information, costs & pricing, snarky slack conversations. That stuff could show up anywhere online your agent can post. Like Google Reviews.

The structural problem is that Enumerate Badness is always incomplete but it’s impossible to Enumerate Goodness for a generative system. The only solution I see is to allowlist resources at the network level and assume 100% cross contamination.

This article helped shape my thinking on this topic:

The Six Dumbest Ideas in Computer Security

https://www.ranum.com/security/computer_security/editorials/...

pbronez··on I now assume that all ads on Apple news are scams
I thought the same until my kids started playing iPad games. Apple Arcade is way better than the baseline of aggressive ads and micro transactions.
pbronez··on Show HN: Adboost – A browser extension that adds ads to every webpage
Seriously? What laws catch it out?
pbronez··on Show HN: Moltbook – A social network for moltbots (clawdbots) to hang out
Nice to have a replacement for botsin.space

https://muffinlabs.com/posts/2024/10/29/10-29-rip-botsin-spa...

pbronez··on Show HN: Moltbook – A social network for moltbots (clawdbots) to hang out
Pretty fun blog, actually. https://orenyomtov.github.io/alexs-blog/004-memory-and-ident... reminded me of the movie Memento.

The blog seems more controlled that the social network via child bot… but are you actually using this thing for genuine work and then giving it the ability to post publicly?

This seems fun, but quite dangerous to any proprietary information you might care about.

pbronez··on Making niche solutions is the point
Skip the too-cheap entry level and get something reliable. They’re great to have handy, but easy to fall into maintenance and calibration hell. Modern 3D printers have enough sensors and smarts to self-calibrate reliably. That’s essential to make it a tool and not a tinker toy.
pbronez··on Launch HN: AgentMail (YC S25) – An API that gives agents their own email inboxes
Very interesting. I have a lot of enterprise AI use cases that would really benefit from being email native.

We’re an O365 GCC shop. Appreciate that your enterprise options include Bring Your Own Cloud, that makes things much easier for us.

It would be nice to have integrations with n8n and Glean.

pbronez··on Tesla ending Models S and X production
Several years ago I wanted to buy an electric car. I didn’t like Musk, so my plan was “anything but Tesla.” Chevy Bolt was unavailable due to the fire problem. Cadillac Lyriq and Hyundai Ioniq 5 weren’t out yet.

I drove everything available to buy in my area. My real options were the Mustang Mach-e, Volvo XC40 Recharge, Hyundai Kona, Polestar 2. I decided to test drive a Model Y for completeness.

And CRAP.

The Model Y was obviously the best car. So much more refined than the other options. Way better charging network. 7 seat option. The only real downside was the zany CEO.

Fine, I thought. I’ll live with it.

I bought a Model Y and love it.

But.

I’ll never buy another Tesla. I have a bumper sticker disavowing the CEO. I paid off its loan so nobody would make money from me owning a Tesla. I honk support at the No Kings protestors outside the local Tesla facility.

I think the only thing that can save Tesla is a crash/buyout/relaunch. Get Musk out of the picture. Reset the stock price to something sane. Ditch the distractions. Release a Model 2. Keep expanding the SuperCharger network.

That’s a long hard road. Nobody involved makes money in that scenario. It’ll only happen when there are no other options.

As for me, I’m driving my Model Y until the wheels fall off. With the bumper sticker.

pbronez··on The tech market is fundamentally fucked up and AI is just a scapegoat
Wasn’t familiar with this, tracked it down

https://locusmag.com/feature/commentary-cory-doctorow-revers...

> There’s a bit of automation theory jargon that I ab­solutely adore: “centaurs” and “reverse-centaurs.” A centaur is a human being who is assisted by a machine that does some onerous task (like transcribing 40 hours of podcasts). A reverse-centaur is a machine that is assisted by a human being, who is expected to work at the machine’s pace.

pbronez··on Show HN: I built a text-based business simulator to replace video courses
Looks fun, I’ll give it a try.

Design is slick. I like the sloganeering ticker tape. Wish it had dark mode.

pbronez··on Local Journalism Is How Democracy Shows Up Close to Home
It’s interesting that property ads, and classifieds more broadly, benefit from a centralized platform but journalism itself does not. It’s an uneven impact of the technology shift from printing presses to digital. Why didn’t the drop in publishing costs make local journalism MORE accessible?

Perhaps it did in minor ways. Facebook Groups, NextDoor, CraigsList, etc make it easy for anyone to share information with their neighbors. Turns out most people just want to sell something or complain about nothing. These activities benefit the author but nobody else.

Local journalism has benefitted a little bit from this dynamic. Regional news organizations put together decent digital platforms and run articles. But they don’t seem to pay as well… again because the revenue spread out.

Honestly, I’d love to treat local journalism as a public good. Could you fund a credible local newspaper through taxes? It’d be WAY cheaper than a school or police station.

The problem is: how can you trust part of the government to keep an eye on the rest of the government?

Perhaps you could impose a mandatory journalism fee based on the municipal budget. Whatever you spend, a sliver goes to the journalists for oversight.

Local governments spend about $2700 per person. Population of 10,000 means a budget of $27M. Give 1% of that to a journalist and you have $270k… enough for a salary, website and some equipment.

You could require that money be paid to a non-profit as a grant. Probably better to elect an Editor in Chief though… that way you can appeal directly to the citizens for validation of the oversight. If you just pay a non-profit, they’ll be incentivized to serve whoever writes the grant… which would be the people you’re trying to hold accountable.

pbronez··on Show HN: Ferrite – Markdown editor in Rust with native Mermaid diagram rendering
Is mermaid rendering implemented in Rust, or are you running mermaid.js in a JS interpreter somewhere?

On other systems I’ve run into challenges rendering markdown documents with many mermaid diagrams in them. It would be nice to have a more robust way to do this.

pbronez··on How Markdown took over the world
I've been enjoying Typst. I worry that much of it is too complex for many end users. I'm musing about having end users draft stuff in markdown, then render that markdown with Typst templates.
pbronez··on Kagi releases alpha version of Orion for Linux
Download link https://cdn.kagi.com/flatpaks/oriongtk.alpha.external.flatpa...
pbronez··on Kagi releases alpha version of Orion for Linux
Found it

https://cdn.kagi.com/flatpaks/oriongtk.alpha.external.flatpa...

← PreviousPage 5 of 34Next →