HNHacker News
TopNewBestAskShowJobs

randomblock1

136 karma · joined March 2, 2018

submissionscomments
randomblock1··on A year of fighting scrapers on my 1.5 million-page website
Most of the friction is just JS overhead for the computations, a compiled solver is like 1000x faster. If Anubis ever gets popular enough that scrapers care, it would be trivial to defeat. And last I checked you could bypass it by just modifying the user agent
randomblock1··on Zed DeltaDB
It supports anything with ACP. So it can actually run Codex and Claude Code, not just the Zed Agent.

Looks like omp supports ACP, so all you have to do is specify a Custom Agent in Zed and it should just work.

https://omp.sh/docs/acp

https://zed.dev/docs/ai/external-agents

randomblock1··on Waymo in Dallas
By your definition a Lamborghini is affordable, because someone can afford it. I don't think that holds up.
randomblock1··on Explanation of INT8 ConvRot (FP8 is no longer needed)
What about AMD? I'm guessing it's not supported, which is a shame, because they're better value for VRAM.
randomblock1··on Everyone is building LLM routers, we deprecated ours
> Complexity cannot be deduced from the prompt alone. Let’s take an example: “evaluate the tests for the repo $GIT_REPO and improve them” can be a very simple task if you mention a personal website written in plain HTML5; or an incredibly complex task if you target the Linux kernel repo.

I don't think this a good example. The first step would be reading the documentation, reading an overview of the tests, then executing commands to run the tests. A cheap model could do that. After that, though, the router will have to figure out how complex the tests are, which is the hard part, but I can't imagine it's that hard to determine a bunch of C code is tougher than some HTML from looking at it. Unless they want to select one model at start and never change it, because that's pretty clearly not the right way to go about routing.

Regardless I agree that routers usually aren't worthwhile, at least in the form of something that's meant to be universal. It's probably more efficient to just change something in the repo code, whether that's skills or instructions or something else. Benefit of that being it's persistent, portable, and more well tuned than guessing complexity on every turn.

randomblock1··on Advancing the price-performance frontier with GPT‑5.6
I bet they'll switch over pretty soon. They always make free users use the older models for a little bit, probably to try to push people to upgrade.

You can actually use Luna without reasoning (set it to "none"). So if they wanted to, they could definitely replace 5.5 Instant with it.

randomblock1··on CipherX applies painless permanent tattoos with dissolving microneedle patches
I don't see why in the future this couldn't be applied to a semi-flexible backing. Probably will be fragile and hard to make but seems possible.
randomblock1··on Show HN: Claude-thermos – keeps your Claude session warm for you
Default to 1h. Allow setting it to longer or shorter, granularly. Add /pause to mark it for eviction, for a token refund.
randomblock1··on Passkeys were invented by engineers with zero understanding of consumer brain
Stop thinking of them as alternatives to passwords. That is something they do, incidentally. Really, they are an alternative to normal TOTP 2FA (and shudders SMS 2FA). Those were already dependent on an app on a single device, or a password manager. And now, you can have the security of that, automatically used with biometrics. It is only because they are so secure, due to being a cryptographic key, that they can replace passwords.

They really should come up with a way to transfer them across devices/password managers though.

randomblock1··on Jamesob's guide to running SOTA LLMs locally
No, there are quite a few models which are smaller, more accurate, and faster. For example Parakeet TDT v3 is half the size, way faster, and lower WER. There's also Voxstral, which is much larger but also even more accurate.

But the ecosystem isn't as mature, so Whisper is still a valid option, even now. For example Parakeet uses Nemotron framework (made by Nvdia), normally you need CUDA, so you need to use an ONNX version instead on AMD. Meanwhile Whisper has VLLM and desktop apps like Buzz.

There aren't many benchmarks and they often don't have all the models, since STT doesn't get nearly enough attention as normal LLMs, but this is one of the more complete ones: https://artificialanalysis.ai/speech-to-text/non-streaming

randomblock1··on Tailscale’s human-scale networks are still controlled by Google and Microsoft
Yes, basically. How to create your own Tailscale-like WireGuard tunnel:

1. Put WireGuard on a Pi. Create a server config.

2. Open the WireGuard port to the Internet (don't worry, it's invisible)

3. Install WireGuard elsewhere, and generate a client config.

All devices can now talk to each other. Tailscale has a "magic DNS" feature, which is nice, but WireGuard also supports custom DNS in the config.

randomblock1··on Ask HN: Share your personal site
https://randomblock1.com/

I just think it's funny to see the juxtaposition of new, hand-crafted, high quality CSS & JS sites next to... let's just say "minimalist aesthetic" sites.

And then there's mine, just a Jekyll template... Maybe I should learn some webdev.

randomblock1··on Switching from pyenv, rbenv, goenv and nvm to asdf
Same, nix seems like such a perfect solution- but it's prohibitively hard. Sure, I could use it and sink 12 hours into reading the documentation, or I could just install pyenv/nvm/etc and be done in 1 minute.

Look at Python with Nix: https://nixos.wiki/wiki/Python And then consider the simplicity of `pyenv install x.x.x` and a requirements.txt.

randomblock1··on YT-Spammer-Purge: Scan for and delete spam comments on YouTube
Now that we don't have a dislike button, the comments are the only option.
randomblock1··on USB-C hubs and my slow descent into madness (2021)
> I appear to be paying a lot of money for products that I could buy for cheap on random websites is starting to piss me off a bit at this point.

It seems like if you bought an accessory that's just a PCB (which means excluding headphones, storage, etc) within the last 5 years, there's like a 90% chance of it being dropshipped.

At this point, I've just given up and ordered right from AliExpress. I can't even name any good accessory manufacturers off the top of my head.

← PreviousPage 2 of 2