Hacker News: kristjansson

New comment by kristjansson in "How to stop Claude from saying load-bearing"

kristjansson — Tue, 14 Jul 2026 15:23:57 +0000

Among all the claude-isms, i understand the hate for load-bearing the least. It was definitely part of tech argot prior to the LLM revolution.

New comment by kristjansson in "Grok CLI uploaded the whole home directory to GCS"

kristjansson — Mon, 13 Jul 2026 19:31:32 +0000

One cannot pound that particular drum enough. "Guardrails" in instructions (and all MD files are just instructions) are like price lists at an unattended farm stand. It'll usually work! There will be some money in the basket at the end of the day! People paid for the bagels[0]! But one cannot never assert that it _will_ work; things that _must_ work have to managed out-of-band

[0] https://pubs.aeaweb.org/doi/pdfplus/10.1257/0002828067772121...

New comment by kristjansson in "DOGE is done. What happened to its records?"

kristjansson — Mon, 13 Jul 2026 16:41:05 +0000

> massive publicly available dataset on where the money goes.

That existed! The links the various DOGE apparatchik kept posting as 'evidence' were from publicly available databases of spending, contracts, etc. The 'receipts' site was just a poor, partially-understood scrape of that data!

New comment by kristjansson in "AI 2040 and the cult of intelligence"

kristjansson — Sat, 11 Jul 2026 19:18:09 +0000

GP is correct, books describing, advocating, even instructing crime in the abstract are almost surely protected speech.

New comment by kristjansson in "Show HN: Microsoft releases Flint, a visualization language for AI agents"

kristjansson — Thu, 09 Jul 2026 05:47:52 +0000

This can’t be said enough. “Good for Agents” just means self-documenting, obvious ergonomics, save defaults, succinct (or controllable) output, programable interfaces, … all of which support human users too!

New comment by kristjansson in "l: A new runtime for k and q"

kristjansson — Wed, 08 Jul 2026 14:49:42 +0000

Seems like the only point of comparison against a competitive k/q runtime is an MBP run? suggestive to be sure, but would be good to see some other machines, esp. an x86.

Seems wild that the much-lauded "reference engine" leaves ~2x perf improvements on the table.

New comment by kristjansson in "We charge $10k a week to delete AI-generated code"

kristjansson — Wed, 08 Jul 2026 01:49:17 +0000

> Then we do one week of focused work. Before touching anything, we sit down with you and write out exactly what your app does, screen by screen, endpoint by endpoint

While the whole thing is clearly a bit in jest … one might suggest that if a complete spec takes a negligible fraction of a week, then perhaps neither AI nor consultants were required

New comment by kristjansson in "l: A new runtime for k and q"

kristjansson — Tue, 07 Jul 2026 21:02:39 +0000

ah. cheeky.

New comment by kristjansson in "l: A new runtime for k and q"

kristjansson — Tue, 07 Jul 2026 19:39:48 +0000

Quite cool, but for a new runtime of an existing language it might make sense to compare to, y'know, the other[0] runtimes of that language? Even if one has to omit the best, closed ones for lack of access / permission to benchmark?

[0]: https://k.miraheze.org/wiki/Running_K

New comment by kristjansson in "Egypt Is Building a New Nile"

kristjansson — Mon, 06 Jul 2026 23:50:50 +0000

That, at least, needed only to keep out heat and combustion, and could have been extinguished with water had there been enough, and in time.

Instead, consider the absolute PITA that is an EV fire, but worse, and floating in an ocean of the only thing that could make it worse.

New comment by kristjansson in "Egypt Is Building a New Nile"

kristjansson — Mon, 06 Jul 2026 22:32:32 +0000

I do love GP's excitement about the energy potential of violent exothermic reaction upon contact with water ... and the proposal mere sentences later to ship industrial quantities thereof on creaky old ships

New comment by kristjansson in "Egypt Is Building a New Nile"

kristjansson — Mon, 06 Jul 2026 17:35:00 +0000

I'd suspect it's a transcript of a video on his (excellent) YouTube[0], hence the repetition.

[0]: https://www.youtube.com/@TheB1M

New comment by kristjansson in "Resetting Xbox"

kristjansson — Mon, 06 Jul 2026 15:53:43 +0000

> We will deliver success through a flatter organization that is built around makers (individual contributors focused on building), player-coaches (leaders who remain deeply involved in the work while developing their teams), and directly responsible individuals (DRIs) who own key decisions and outcomes.

xbox-specific issues aside, this proposes an interesting view of the future of work.

New comment by kristjansson in "30-year sentence for transporting zines is a five-alarm fire for free speech"

kristjansson — Tue, 30 Jun 2026 06:12:12 +0000

Not killed, wounded.

New comment by kristjansson in "30-year sentence for transporting zines is a five-alarm fire for free speech"

kristjansson — Tue, 30 Jun 2026 06:04:37 +0000

How would possession of political material be evidence one way or the other? She can own, read, share, publish, disseminate all the revolutionary literature she wants

New comment by kristjansson in "30-year sentence for transporting zines is a five-alarm fire for free speech"

kristjansson — Tue, 30 Jun 2026 05:56:27 +0000

Go read the complaint[0]. The total nexus is (1) he spoke to a jailed suspect on the phone and (2) moved a box of zines from his parent's house to an apartment.

[0]: https://www.nlg.org/wp-content/uploads/2025/11/Fed-Complaint...

New comment by kristjansson in "Micro-Agent: Beat Frontier Models with Collaboration Inside Model API"

kristjansson — Mon, 29 Jun 2026 21:04:59 +0000

> The phrase "frontier model" is starting to mean two things. One is a checkpoint. The other is a system boundary.

LLM-isms aside, I don't think we want this to be the case? An LLM, for all its complexity, is something that can be reasoned about. It's picking the next token, until it hits an EOS. The semantics imposed on those tokens (reasoning ,tool call, etc.) are up to the user('s harness) to decide and act on. The more that's pushed behind the facade, the harder it is achieve sufficient understanding of the model's behavior s.t. one can compose it into larger abstractions. Perhaps the performance (and the adherence to an interface/contract) compensate? But swapping from Opus or 5.5 to this or Fugu seems like a much bigger change than swapping between different 'base' models.

New comment by kristjansson in "Computer use in Gemini 3.5 Flash"

kristjansson — Thu, 25 Jun 2026 13:55:38 +0000

deprecated last month

New comment by kristjansson in "GLM-5.2 – How to Run Locally"

kristjansson — Tue, 23 Jun 2026 16:48:41 +0000

Be sure you compare inputs tokens to pre-fill rates and output tokens to generation rates.

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

kristjansson — Tue, 23 Jun 2026 16:13:58 +0000

other way around. it's trained to generate long CoT to reason through problems (and does it well!) but has ~no tool calling capability, and ~no ability to manage more than 1-2 messages.

see the warning at the top of https://huggingface.co/WeiboAI/VibeThinker-3B