Hacker News: leobg

New comment by leobg in ""Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok"

leobg — Wed, 22 Jul 2026 07:42:23 +0000

I'm missing some context here.

First of all, this is a marketing piece, made for engagement.

Second, in production, you would adjust your approach to the model's strengths and weaknesses. For example, if you found that a model does not respond to the tools you are offering, then you would find a way to make it more responsive. Also, we know that different models handle prompts and system prompts differently. So comparing several models from different providers against exactly the same prompt is quite a naive way of exploring the capabilities of those models or comparing them to each other. You may have just happened to “speak the language“ of one model better than that of another. Why would that be the model’s fault and not yours?

Third, none of these results mean anything if you do not run each process on each model several times to check for consistency. Your model might just have gotten lucky. This point is even more true when you're talking about a progressive task where one step builds on the previous one. The model might just have painted itself into a corner.

New comment by leobg in "Show HN: Ex Situ – Open-source spatial index of displaced cultural artifacts"

leobg — Tue, 21 Jul 2026 12:19:13 +0000

Looks like you've got some real history there. Think about it. You might even be able to hack together an Ask My Ancestor chatbot. Heck, I’d build it for you. The scamming and the OCR are going to be the hardest parts I imagine.

New comment by leobg in "You only need the frontier model for one single edit"

leobg — Tue, 21 Jul 2026 08:41:07 +0000

I guess Anthropic knows this. They show you this warning:

--- Switch model? Your next response will be slower and use more tokens

This conversation is cached for the current model. Switching to Opus 4.8 (1M context) means the full history gets re-read on your next message. ---

Reminds me a bit of airline booking sites ("We noticed you didn't add insurance").

New comment by leobg in "Show HN: Ex Situ – Open-source spatial index of displaced cultural artifacts"

leobg — Tue, 21 Jul 2026 08:33:10 +0000

In that case, the map looks misleading, doesn’t it? It looks as though London and Berlin, for example, stand out as having dislocated disproportionately many artifacts from around the world. And on the other hand, Paris looks innocent. But in truth, London and Berlin are just being honest about it giving you open access to the data whereas Paris doesn’t. So perhaps it would be better to mark cities that do provide data with, say, a green dot. This also helps the visual understanding of the direction of the transfer. And it would make it clear at a glance that Paris is not innocent, but that Paris is simply lacking data.

New comment by leobg in "Show HN: Ex Situ – Open-source spatial index of displaced cultural artifacts"

leobg — Tue, 21 Jul 2026 08:28:12 +0000

You know that they exist, you know where they are, and you’ve never read or digitized them? Or was that sarcasm?

New comment by leobg in "Ask HN: Favourite science history book of the 20th century?"

leobg — Mon, 20 Jul 2026 19:54:24 +0000

The Faber Book of Science (Carey) Asimov’s New Guide to Science (Asimov) The Great Influenza (Barry)

New comment by leobg in "Open Book Touch: open-source e-reader"

leobg — Sat, 18 Jul 2026 07:33:20 +0000

You can. But if you build your own stuff, you can also easily lock yourself out. That whole mechanism that tells the software to look for a new bin file on the SD card and to use that to update is something that needs to be contained in your software. If you forget it, or you get it wrong, the thing will be bricked.

New comment by leobg in "AWS: Inaccurate Estimated Billing Data – $1.7 billion"

leobg — Sat, 18 Jul 2026 07:29:03 +0000

I had a fun day yesterday. Because I had been contacted by AWS about somebody potentially having gained access to my account.

That’s when I logged in. That’s when I saw the 1.3 billion number.

I was pretty fast in getting that backdoor closed and hunting down that leak, I can tell you! :D

New comment by leobg in "Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU"

leobg — Thu, 16 Jul 2026 05:55:16 +0000

I’m in DE, too. Same calculus. Excluding winter though. When it’s cold outside, a GPU heats my office and I get tokens free on the side. :)

New comment by leobg in "GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2"

leobg — Sat, 20 Jun 2026 08:48:07 +0000

Skinner would say it is not so much about emotions like fear or greed, but about consequences.

New comment by leobg in "Show HN: MCP file tools silently eat your context window.I built one that doesnt"

leobg — Mon, 09 Mar 2026 10:15:51 +0000

Looks interesting. Any pointers to how to integrate it in GitHub Copilot Chat? It seems to support skills, so I'd guess it should work?

New comment by leobg in "I Build a Open Source Deep Research Engine Wich Beats Google and Open AI"

leobg — Thu, 05 Feb 2026 14:39:20 +0000

You got a .de Domain. I looked for an imprint to see if that solo dev with the right tools was a fellow German, but I didn’t find any. Any details on who you are and how you came to build this? Also, your landing page sounds quite different than what you are pitching here. Are these two different projects? Thx

New comment by leobg in "BYD's cheapest electric cars to have Lidar self-driving tech"

leobg — Mon, 12 Jan 2026 12:42:08 +0000

That isn’t the point. The point is that it’s not enough to figure out that there is an obstacle. You also have to figure out what that obstacle is, and you have to predict its movement. In the case of pedestrian, the car, for example, needs to know whether the pedestrian has seen you. Things like that you just cannot do with LiDAR. Hence you’re gonna need cameras anyway. Hence tge “anyone relying on LiDAR is doomed” prediction.

New comment by leobg in "UK threatens action against X over sexualised AI images of women and children"

leobg — Mon, 12 Jan 2026 12:38:44 +0000

So I guess in the 90s they would’ve sued Adobe for not putting spyware into Photoshop?

If you believe in democracy, and the rule of law, and citizenship, then the responsibility obviously lies with people who create and publish pictures, not the makers of tools.

Think of it. You can use a phone camera to produce illegal pictures. What kind of a world would we live in if Apple was required to run an AI filter on your pics to determine whether they comply with the laws?

A different question is if X actually hosts generated pictures that are illegal in the UK. In that case, X acts as a publisher, and you can sue them along with the creator for removal.

New comment by leobg in "Think of Pavlov"

leobg — Sun, 11 Jan 2026 13:44:57 +0000

Some people need the opposite advice: Sometimes an interaction is just a one off event. No need to teach a lesson.

New comment by leobg in "Org Mode syntax is one of the most reasonable markup languages for text (2017)"

leobg — Sat, 10 Jan 2026 16:56:30 +0000

How do you handle PDFs, images, HTML/rich text snippets? Asking because Evernote supported all of those. It’s what keeps me using that app, even though I cringe every time I open it.

New comment by leobg in "Meditation as Wakeful Relaxation: Unclenching Smooth Muscle"

leobg — Fri, 09 Jan 2026 20:55:40 +0000

The issues get trapped in the tissues.

New comment by leobg in "Meditation as Wakeful Relaxation: Unclenching Smooth Muscle"

leobg — Fri, 09 Jan 2026 20:54:37 +0000

Have you heard of “the Sedoma Method”? And/or Larry Crane’s Release Technique?

New comment by leobg in "The Vietnam government has banned rooted phones from using any banking app"

leobg — Fri, 09 Jan 2026 20:31:20 +0000

The irony is that Apple started out by discovering the the hackability of the hardware and software they found in their time. Instead of leaving something like that behind for those who come after them, to pay back what was given to them, they build walled gardens where you’re just not allowed to “bump into the walls too much”.

New comment by leobg in "ChatGPT Health"

leobg — Wed, 07 Jan 2026 20:55:46 +0000

Vibe coded…. /s