Hacker News: aqfamnzc

New comment by aqfamnzc in "Universal Claude.md – cut Claude output tokens"

aqfamnzc — Tue, 31 Mar 2026 20:46:25 +0000

A "reasoning" LLM is just an LLM that's been instructed or trained to start every response with some text wrapped in or similar. The UI may show or obscure this part. Then when the model decides to give its "real" response, it has all that reasoning text in its context window, helping it generate a better answer.

New comment by aqfamnzc in "GPT‑5.3 Instant"

aqfamnzc — Wed, 04 Mar 2026 03:09:33 +0000

I just append something like "Throughout our conversation, keep your responses brief. Avoid emojis, followup suggestions, and other unnecessary commentary." to every starting prompt. Seems to work OK. I'm sure sibling's recommendation of turning down the niceties sliders would work similarly for someone with an account.

New comment by aqfamnzc in "Windows 11 Notepad to support Markdown"

aqfamnzc — Thu, 26 Feb 2026 03:52:28 +0000

Even funnier when you remember that they own github, the place where arguably markdown was popularized.

New comment by aqfamnzc in "Floppy disks turn out to be the greatest TV remote for kids"

aqfamnzc — Mon, 12 Jan 2026 23:01:16 +0000

Is the data stored on the card, or on the player? My guess is that each card just holds an id?

New comment by aqfamnzc in "EU Council Approves New "Chat Control" Mandate Pushing Mass Surveillance"

aqfamnzc — Fri, 28 Nov 2025 15:00:38 +0000

It's just a metaphor.

New comment by aqfamnzc in "OSMAnd vs. Organic Maps"

aqfamnzc — Thu, 04 Sep 2025 04:58:54 +0000

Caltopo is great for this. They require a subscription to download (raster) maps but you can cache a bunch of tiles before you leave to get the gist. These days this is one of the very scarce use cases I don't use OSMand for.

New comment by aqfamnzc in "Claude 4"

aqfamnzc — Thu, 05 Jun 2025 13:26:40 +0000

Not even remotely true. It's part of the context window, so it greatly influences the final output. CoT is tokens generated by the LLM just like normal output.

New comment by aqfamnzc in "Claude 4"

aqfamnzc — Fri, 23 May 2025 17:19:39 +0000

It's more hallucination in the sense that all LLM output is hallucination. CoT is not "what the llm is thinking". I think of it as just creating more context/prompt for itself on the fly, so that when it comes up with a final response it has all that reasoning in its context window.

New comment by aqfamnzc in "Ask HN: What are good high-information density UIs (screenshots, apps, sites)?"

aqfamnzc — Thu, 08 May 2025 16:11:24 +0000

The calculator is an extreme example, but I've wondered in the past if the reason they scrub everything is so you can't take the manufacturer part number to buy elsewhere. McMaster is undoubtedly more expensive in many cases, but the service they offer is consolidating a million parts into one catalog with CAD drawings, specs, etc. Hiding branding prevents you from taking advantage of that without making a purchase.

New comment by aqfamnzc in "Tailscale has raised $160M"

aqfamnzc — Tue, 08 Apr 2025 17:56:20 +0000

> It’s a bit like saying Dropbox is just a GUI on top of TLS.

Well, it is. After all, for a Linux user, you can already build such a system yourself quite trivially...

New comment by aqfamnzc in "Piranesi's Perspective Trick (2019)"

aqfamnzc — Thu, 27 Mar 2025 13:16:43 +0000

Great article. And thanks for the correct OSM attribution in your map images. :)

New comment by aqfamnzc in "FOSS infrastructure is under attack by AI companies"

aqfamnzc — Thu, 20 Mar 2025 16:16:17 +0000

As it is now, unethical AIs have a huge advantage over ethical ones.

New comment by aqfamnzc in "Documenting an 1115 ft radio tower climb"

aqfamnzc — Mon, 06 Jan 2025 04:36:10 +0000

Hi, I've considered changing careers to rope access or similar for a while now. Can I interview you about your job? Please email me, see profile. (This request extends to other RATs reading this too!)

New comment by aqfamnzc in ""Nvidia is so far ahead that all the 4090s are nerfed to half speed""

aqfamnzc — Mon, 16 Dec 2024 21:25:38 +0000

What does a "pretty bad" h265 implementation look like? Buggy? Inefficient or what?

New comment by aqfamnzc in "The size of BYD's factory"

aqfamnzc — Mon, 25 Nov 2024 20:40:50 +0000

I wonder if (partially) they use discharge current from one batch of batteries to charge the next?

New comment by aqfamnzc in "Niantic plans a “Large Geospatial Model” trained on Pokémon Go player data"

aqfamnzc — Thu, 21 Nov 2024 00:12:39 +0000

Agreed. I mostly meant that I'll never see the actual dataset that I contributed to. That's why I'd prefer to spend my time on things that I can see, like OpenStreetMap :)

New comment by aqfamnzc in "Niantic plans a “Large Geospatial Model” trained on Pokémon Go player data"

aqfamnzc — Wed, 20 Nov 2024 23:26:47 +0000

Welcome to the modern internet. While you're at it, please get me access to Google's captcha models facebook face directory Google's GPS location data hoard, (most every android phone on the planet 24/7 (!) and any iPhone navigating with gmaps) And so on and so on

All of which I've directly contributed to and never (directly) recieved anything in return

New comment by aqfamnzc in "Fair coins tend to land on the side they started: Evidence from 350,757 flips"

aqfamnzc — Tue, 19 Nov 2024 17:38:33 +0000

Sounds like maybe you were doing something like this? https://news.ycombinator.com/item?id=42184069

New comment by aqfamnzc in "The withering dream of a cheap American electric car"

aqfamnzc — Mon, 18 Nov 2024 15:13:17 +0000

Money. The insurance company gives a discount. And honestly, for someone who doesn't share my same strong values for privacy, I don't blame them!

New comment by aqfamnzc in "Maybe Bluesky has "won""

aqfamnzc — Fri, 15 Nov 2024 21:45:48 +0000

Wait, did I miss something that happened just over a week ago?