Hacker News: stephantul

New comment by stephantul in "Ask HN: Do you worry about fires in AI datacenters?"

stephantul — Mon, 14 Sep 2026 06:19:13 +0000

New comment by stephantul in "Bad benchmarks and evals: Senior SWE-Bench, napkin math, and winter tires"

stephantul — Mon, 14 Sep 2026 06:11:33 +0000

The (relevant) segue into tires elevates this post so much. I don’t understand why, but it does

New comment by stephantul in "Pandas Should Go Extinct"

stephantul — Sat, 12 Sep 2026 03:29:04 +0000

Agreed on all counts.

In many cases I’ve found directly using python primitives to be less confusing than pandas.

Similarly, in companies I’ve worked at, the datasets just aren’t that big. Especially if you’ve got access to modern hardware.

New comment by stephantul in "The Last 24 Hours Are the Opening Scene in a Horror Movie"

stephantul — Fri, 11 Sep 2026 16:51:01 +0000

If you find a place where I can host a trillion parameter model without anyone finding out about it, let me know.

New comment by stephantul in "The Last 24 Hours Are the Opening Scene in a Horror Movie"

stephantul — Fri, 11 Sep 2026 16:49:29 +0000

But how. Models don’t have access to their own weights.

New comment by stephantul in "The Last 24 Hours Are the Opening Scene in a Horror Movie"

stephantul — Fri, 11 Sep 2026 16:48:47 +0000

Not the code: the weights. Are you going to host a trillion parameter model somewhere without someone noticing?

New comment by stephantul in "The Last 24 Hours Are the Opening Scene in a Horror Movie"

stephantul — Fri, 11 Sep 2026 16:39:27 +0000

Ok but do any of these data centers have a copy of the models that attacked hf?

New comment by stephantul in "The Last 24 Hours Are the Opening Scene in a Horror Movie"

stephantul — Fri, 11 Sep 2026 16:27:59 +0000

One thing I definitely do not understand about this discourse is that the models that are good enough to self-replicate can’t survive on normal machines, e.g., the models can’t hide on some random server.

So, if it is as dangerous as they say it is: there is a single physical source of this danger, which is OpenAI/Anthropic servers. If it is this dangerous, they can just turn it off.

Instead, we just keep pretending that the models that attacked HF were hosted or replicating on HF hardware. Not the case! They infiltrated it, but were hosted elsewhere.

New comment by stephantul in "RTK reports token savings, but our cost benchmarks disagree"

stephantul — Fri, 11 Sep 2026 16:22:53 +0000

I think anyone who is even a little bit realistic knows that most technologies overclaim, or evaluate under very favorable conditions.

This is not a good thing of course, but I also feel that acting surprised that this is going on is a little unnecessary.

Having said that: most tools are not helpful

New comment by stephantul in "Harnessing the Universal Geometry of Embeddings"

stephantul — Mon, 07 Sep 2026 05:06:44 +0000

I’ve never liked that this was called “the platonic representation hypothesis”. Lots of weird baggage attached and seems like a waste of a good name.

New comment by stephantul in "Benchmarking Vector Indexes"

stephantul — Tue, 01 Sep 2026 14:27:40 +0000

This is a super long ai generated slab of text with very little actual info. It doesn’t include any analysis of results.

It does include the phrase “worth listing”.

New comment by stephantul in "METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack"

stephantul — Sun, 30 Aug 2026 19:13:43 +0000

Super charitable reading imo. This is like saying we can’t detect a speeding car because we can’t run as fast as a fast car.

It’s not like the humans were engaged in some kind of battle of wits with some super AI, it’s just some employee not monitoring the output of an experiment.

New comment by stephantul in "You Know GDPR Is Good Based on Who Hates It"

stephantul — Sat, 29 Aug 2026 10:32:08 +0000

Cookie banners are made annoying on purpose. This has nothing to do with GDPR itself.

The entities forced to show them would rather not, and thus make it as annoying as possible for you. They then use this to weaken support for the GDPR.

Shame on the people making stuff like this.

New comment by stephantul in "Andreessen Horowitz is investing billions into a bleak future"

stephantul — Mon, 24 Aug 2026 09:04:47 +0000

Tragedy of the commons. The person that ruins a commons first takes all the supply.

This is why regulation before, not after, the commons are plundered, is important.

Hiding a Prompt in a Tokenizer

stephantul — Mon, 24 Aug 2026 08:16:32 +0000

Article URL: https://stephantul.github.io/blog/prompt/

Comments URL: https://news.ycombinator.com/item?id=49416676

Points: 2

# Comments: 0

New comment by stephantul in "The Benchmarkpocalypse"

stephantul — Tue, 18 Aug 2026 05:59:35 +0000

Unfortunately, even a holdout set doesn’t protect you from overfitting, it just takes longer.

Of course having a holdout set is better than not having one. It’s just not a silver bullet.

New comment by stephantul in "How I over-engineered my book"

stephantul — Mon, 17 Aug 2026 19:28:55 +0000

It is perhaps ironic that I find this post very difficult to read. I'm super interested in the content, but it reads like it is generated.

New comment by stephantul in "How to disable or avoid intrusive AI"

stephantul — Mon, 17 Aug 2026 17:00:19 +0000

My heart bleeds for Rovo’s product managers. They must also see that nobody actually wants or needs or even likes Rovo.

New comment by stephantul in "Manus will return to operating as an independent company"

stephantul — Tue, 11 Aug 2026 15:37:34 +0000

Anyone have info on whose regulatory restrictions they are referring to?

New comment by stephantul in "Show HN: Mcptoon – Token-efficient MCP CLI client"

stephantul — Tue, 11 Aug 2026 07:28:54 +0000

I think that some of these choices (as others have commented) show that the author has not investigated how tokenization works.

Tokenization is not some black box, you can run tokenizers and check them.