Hacker News: loehnsberg

New comment by loehnsberg in "DeepSeek search and fetch – $0.75 per 1000 successful requests"

loehnsberg — Sun, 18 Oct 2026 09:34:48 +0000

Kagi MCP works really well imo also within DeepSeek Harness. Just looking at cost neglects the possible difference in quality. How good is that DeepSeek web search/fetch?

New comment by loehnsberg in "Muse Wonder"

loehnsberg — Mon, 12 Oct 2026 19:56:19 +0000

It's interesting that in its marketing Meta compares Wonder to the closed frontier models but not the open (Chinese) models. By its own assessment, Meta could easily show how far ahead its own model is against the rest of the pack of open models but they chose not to. Why?

New comment by loehnsberg in "How Protein Took over the World"

loehnsberg — Sat, 10 Oct 2026 20:02:37 +0000

That is a lot of food. Even if I’d eat just that, I’d have to very consciously force myself to eat one kilo of that, unless it’s something like watermelon.

New comment by loehnsberg in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

loehnsberg — Fri, 09 Oct 2026 17:58:39 +0000

I run my own benchmarks at the interface of math and coding (ML, math-opt, stochastic models), and keep my own set of benchmarks, i.e. somewhat niche linear algebra and numerical recipes. DS-4.1 is the first open model that passes most of these benchmarks, and GPT-5.6-Sol is definitely not ahead. Even if it's all distillation, then DS does a formidable job at it.

New comment by loehnsberg in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

loehnsberg — Thu, 08 Oct 2026 21:34:59 +0000

Which quantization are you using there?

New comment by loehnsberg in "Web Search API"

loehnsberg — Tue, 06 Oct 2026 11:39:52 +0000

I have $10 and it resets every month; family plan.

New comment by loehnsberg in "Web Search API"

loehnsberg — Mon, 05 Oct 2026 16:25:16 +0000

Fully agree. The API via search and extract MCP works really well.

I noticed that my API quota resets every month. Have not been charged once.

New comment by loehnsberg in "Dropbox's Jan 1st 2027 terms of service"

loehnsberg — Sun, 20 Sep 2026 08:22:02 +0000

This leaves a lot of questions. Why are they raising the age limit? What legislation demands this? Will other services follow? And what if parents give consent? Are users on a family plan now forced to kick off their kids from the joined plan?

New comment by loehnsberg in "Blizzard Revives 'StarCraft' Franchise with Open-World Sci-Fi Shooter"

loehnsberg — Mon, 14 Sep 2026 18:45:28 +0000

Looks like StarCraft meets Starship Troopers but without that extra pinch of irony which likens this trailer to a fascists wet dream.

New comment by loehnsberg in "Mistral raises €3B"

loehnsberg — Tue, 08 Sep 2026 06:21:35 +0000

Maybe true for frontier LLM, but there's plenty of space in the niches. For example, I think their TTS/STT models are pretty good, speaking from personal experience.

New comment by loehnsberg in "Agent memory as a file format"

loehnsberg — Mon, 31 Aug 2026 18:28:21 +0000

This mirrors my experience:

- Store session turns in an sqlite-vec

- Provide the agent with an mcp to search the vec-db

- Let the agent write notes in md files along with an index / frontmatter

Along with the commit history, the vec-db gives the agent long-term memory. The notes allow the user to correct accumulation of false lessons.

Simpler but better.

New comment by loehnsberg in "GLM-5.3-Flash"

loehnsberg — Thu, 27 Aug 2026 09:39:42 +0000

There‘s only one truth, the rest is interpretation. I was looking for the non-Chinese interpretation.

New comment by loehnsberg in "GLM-5.3-Flash"

loehnsberg — Wed, 26 Aug 2026 22:20:08 +0000

Deepseek and GLM answered correctly on Openrouter when using non-Chinese endpoints. I hope it stays that way!

New comment by loehnsberg in "New Mac Studio with M5 Max and M5 Ultra"

loehnsberg — Tue, 25 Aug 2026 20:42:54 +0000

On some days I wor from home but need to be present and the sun is out. So I take my laptop outside and work from there. I cannot do this with a desktop. Wouldn‘t the workaholic just stay inside and miss out?

New comment by loehnsberg in "Windows 11's built-in Weather app wastes more than 1 GB of RAM"

loehnsberg — Sun, 09 Aug 2026 17:14:37 +0000

I built a weather app for myself fetching ECMWF ensemble forecasts for my home location. Running the service is 60mb plus data 39mb and another 2mb for the spaghetti charts showing temperature, clouds, precip, wind over 10 days. What do I need to do to fill the other 900mb?

New comment by loehnsberg in "Orion Browser by Kagi"

loehnsberg — Mon, 20 Jul 2026 15:09:17 +0000

Why? Because the claim is that they 'fund' the war by paying Yandex.

New comment by loehnsberg in "Orion Browser by Kagi"

loehnsberg — Mon, 20 Jul 2026 10:15:07 +0000

Getting from using Yandex to funding Putin‘s war requires quite a few turns. Since most of Russia’s war chest is filled by money from oil exports and China indirectly also supports Russia materially, I would first cut myself off of all oil-based products, and then from the Chinese supply chain of goods, and only then cut products indluding Yandex search results. In that order, because that matters for the war chest.

And after you then stopped typing the response, otherwise having to use a device that was made in China built from oil-based components using oil—based transport throughout its supply chain, send a postcard with your apology to the Kagi team and anyone else who does not fund wars and still uses traded goods, because this is how the world is.

If you want to throw a however tiny wrench into Putin‘s war efforts, complain load to your government to join sanctions and to have them support Ukraine and make your choice at the ballot box accordingly.

New comment by loehnsberg in "Orion Browser by Kagi"

loehnsberg — Mon, 20 Jul 2026 00:25:57 +0000

I am using Orion on Mac and iOS for a few years now and I cannot disagree more.

If a site works on Safari but not on Orion it is mostly due to ad blockers etc. Just flick the compatibility mode and it works. I have not encountered a single case where this did not fix it.

Also for a while now Apple Pay works, Apple password manager works, autofill works.

New comment by loehnsberg in "Don't trust large context windows"

loehnsberg — Sun, 14 Jun 2026 08:30:20 +0000

Unless we do our own benchmarks, we have to take all the marketing fluff from the frontier labs at face value, and all public benchmarks degrade eventually as labs optimize towards them. OP’s approach is wasteful because it is brute force, but post says that an ELO is kept, so this is also an experiment, and I don‘t see what‘s wrong with that. You learn which model performs well in which settings which may save resources later. It‘s also wasteful to keep working with the wrong model/harness/tools for too long.

New comment by loehnsberg in "Show HN: Komi-learn – continuous memory and self-improvement for coding agents"

loehnsberg — Sun, 31 May 2026 06:48:55 +0000

It sounds like it solves the problem that everybody who vibe codes over multiple projects runs into, but it does not provide evidence that it actually works and that it is better than structured collection of md files. I think what is lacking in this field are benchmarks like LoCoMo for long sessions.