Hacker News: grumbelbart2

New comment by grumbelbart2 in "VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO"

grumbelbart2 — Tue, 23 Jun 2026 13:27:56 +0000

The thing is we tried that for decades, using more formal logic to build reasoning engines. And we never got it to be even a fraction as good and generic as learning-based LLMs are today.

New comment by grumbelbart2 in "My Mathematical Regression"

grumbelbart2 — Mon, 22 Jun 2026 21:34:09 +0000

This interpretation also generalizes nicely to 3d. You have 60 moves, 20 of which will be up (60 choose 20), then 20 are east (40 choose 20), then 20 are south (20 choose 20).

New comment by grumbelbart2 in "The Cold War's Accidental Whale Observatory"

grumbelbart2 — Sat, 20 Jun 2026 16:11:47 +0000

But were they given access to the raw data, or reports?

New comment by grumbelbart2 in "Where to Find the Colors Your Screen Can't Show You"

grumbelbart2 — Sat, 20 Jun 2026 08:24:21 +0000

This is called metamerism. It can be a practical issue if two pigments have the same color under one light source, but a different one under another. You want your artificial teeth to have the same color as your real teeth in sunlight, led light, and a classic lightbulb for example.

New comment by grumbelbart2 in "Snap Smart Glasses"

grumbelbart2 — Sat, 20 Jun 2026 05:51:22 +0000

Overly specific and not more than feel-good blurb. “When you’re recording” - what if they record it for, say, debugging purposes? Are images send into the cloud even if that light is off? “Prioritize” on-device processing is a meaningless promise, on the contrary - it means that some things will not be done on device. There is nothing in this text stopping them from streaming and storing whatever they want and need.

New comment by grumbelbart2 in "Branchless Quicksort faster than std:sort and pdqsort with C and C++ API"

grumbelbart2 — Fri, 05 Jun 2026 05:37:35 +0000

There is a static branch predictor that is used if there is no statistic on a branching instruction yet, and it's really simple: Jumps backward are assumed to be taken (they usually are from a loop), jumps forward are assumed to be not taken.

So the jump that forms the loop will be predicted correctly for all executions but the very last (when the loop ends).

New comment by grumbelbart2 in "CT scans of BYD car parts"

grumbelbart2 — Wed, 03 Jun 2026 07:00:08 +0000

They are "more" vertical, but they too have vital suppliers that they could not do without. The semiconductor supply chain is deep. Everybody knows ASML, but there are countless others that produce raw wafers, etching machines, special chemicals and so forth.

New comment by grumbelbart2 in "Debug Project"

grumbelbart2 — Tue, 02 Jun 2026 08:38:17 +0000

And also since they don't bite, they won't compete for the same resources / food.

New comment by grumbelbart2 in "Your Most Improbable Life"

grumbelbart2 — Thu, 21 May 2026 09:49:54 +0000

> This motionless destination “without difference”, is also known as heat death, or entropy.

This is BTW not how the heat death would look like. There would still be fluctuations that would, given infinite time, produce almost anything by chance at some point.

This is what the Boltzmann brain is all about: If the universe goes down that path, it is much more likely that what we experience is just a hallucination of a "brain" that spun into existence by chance, rather than all of this being a "real" universe. It's the precursor of the simulation question.

New comment by grumbelbart2 in "The Fil-C Optimized Calling Convention"

grumbelbart2 — Tue, 19 May 2026 06:29:50 +0000

From what I understand, on x86 Linux stores a thread-local pointer to its TLS block in %fs. Could that simply be re-used?

https://groups.google.com/g/comp.arch/c/IT2dhS4q2M8?pli=1

New comment by grumbelbart2 in "Starship V3"

grumbelbart2 — Wed, 13 May 2026 09:11:34 +0000

Isn't the question more an economic one: Is it cheaper to put some solar cells into the desert and to buy some batteries, or to launch things into space (plus the premium for radiation hardening and ensuring it survives long enough because you cannot service it).

Given the current trajectory of battery and solar prices I just don't that space-based systems are cheaper in any way.

Of course there is a long-term aspect should we climb the ladder in the Kardashev scale: Once we used all solar radiation reaching earth we must move to space to grow. But that is decades if not centuries away.

New comment by grumbelbart2 in "Reimagining the mouse pointer for the AI era"

grumbelbart2 — Wed, 13 May 2026 06:44:57 +0000

What is going on with the font in this article? My Firefox renders it as a wired mixture of lower- and upper-case letters, all with the same height. Completely unreadable. Culprint seems to be this:

    font-feature-settings: "ss02" on;

New comment by grumbelbart2 in "First tunnel element of the Fehmarnbelt Tunnel immersed"

grumbelbart2 — Mon, 11 May 2026 14:03:49 +0000

The exact same thing is happening with the Gotthard Base Tunnel. Switzerland has completed it almost 10 years ago, Germany still hasn't finished the links.

It is combination of the German planning system, which allows the processes to be stretched out due to objections, and spineless politicians who don't really want to commit to a route and kick the can down the road (because whatever they do, someone will be angry; but voter's don't really make you responsible if the link isn't built or is delayed).

New comment by grumbelbart2 in "Hardening Firefox with Claude Mythos Preview"

grumbelbart2 — Fri, 08 May 2026 09:27:45 +0000

Both are right and it depends on which model you use and who submits those bugs. The capabilities of leading models went from 99% noise to 99% valid bugs in essentially a few months. Some projects are flooded with the former and need to take precautions to avoid essential DoS attacks on the maintainers.

New comment by grumbelbart2 in "Show HN: Honker – Postgres NOTIFY/LISTEN Semantics for SQLite"

grumbelbart2 — Fri, 24 Apr 2026 11:41:25 +0000

Very cool!

Another maybe stupid question, would something like inotify(7) help to get rid of any active polling?

New comment by grumbelbart2 in "We found a stable Firefox identifier linking all your private Tor identities"

grumbelbart2 — Thu, 23 Apr 2026 06:44:55 +0000

This fingerprint persists over private and non-private Firefox sessions until you restart Firefox. State actors might be able to connect your Google-login in FF window 1 with your tor session in FF private window 2.

New comment by grumbelbart2 in "The RAM shortage could last years"

grumbelbart2 — Mon, 20 Apr 2026 10:42:44 +0000

To show ads you need people to stay on your platform. This is especially true once ads become more intrusive or of lower quality, something the big players seem to gravitate towards to keep revenue up. Google and Meta have ways to lock in users (networking effects, the best search engine available, having your data stored there).

I am not sure if OpenAI has that. Their edge regarding models is small, their strategy currently seems to be "buy ALL the hardware so nobody else can". Users can quite easily switch to other models.

New comment by grumbelbart2 in "Backblaze has stopped backing up OneDrive and Dropbox folders and maybe others"

grumbelbart2 — Tue, 14 Apr 2026 11:05:40 +0000

Git packs objects into pack-files on a regular basis. If it doesn't, check your configuration, or do it manually with 'git repack'.

New comment by grumbelbart2 in "Seven countries now generate nearly all their electricity from renewables (2024)"

grumbelbart2 — Mon, 13 Apr 2026 11:36:43 +0000

> Heat pumps are geothermal masquerading as electric

Air-to-air heat pumps are quite common 1-4 family homes. Even in Nordic countries such as Finland: https://www.sulpu.fi/heat-pump-sales-returned-to-a-growth-pa...

> And the highest number I ever heard for a heat pump was 135% [...] Truth is that electricity is great for kinetic energy but terrible at making heat. Most forms of energy can be transformed into another form of energy at about 50%. Electricity is the weird one where its 90% to motion but only 10% to heat.

Sorry but absolutely not, that's wrong on several levels. First off, in its most basic form of resistive heating, electric heating is already close to 100%. Heat pumps are even better, and I'll just quote Wikipedia

> At a cost of 1 kWh of electricity, they can transfer 1 to 4.5 kWh of thermal energy into a building.

New comment by grumbelbart2 in "LLM scraper bots are overloading acme.com's HTTPS server"

grumbelbart2 — Wed, 08 Apr 2026 13:01:09 +0000

The article describes that a lot of the requests are for non-existent URLs. Do you observe the same?