Hacker News: Tenobrus

New comment by Tenobrus in "An OpenAI model has disproved a central conjecture in discrete geometry"

Tenobrus — Wed, 20 May 2026 20:06:00 +0000

what basis do you have for assuming an LLM is fundamentally incapable of doing this?

New comment by Tenobrus in "OpenAI’s o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors"

Tenobrus — Mon, 04 May 2026 03:34:37 +0000

o1 has a METR time horizon of around 40 minutes, opus 4.7 has an implied horizon of 18 hours based on its ECI score. this study is on a model that's several generations behind wrt the kind of tasks it can complete. it would be shocking if this number were anywhere near as low with GPT 5.5, to the point it seems nearly totally irrelevant to talk about these results

New comment by Tenobrus in "C++26 is done: ISO C++ standards meeting Trip Report"

Tenobrus — Sun, 29 Mar 2026 20:57:04 +0000

"Japanese soldier who kept fighting 29 years after World War 2"

New comment by Tenobrus in "Statement from Dario Amodei on our discussions with the Department of War"

Tenobrus — Thu, 26 Feb 2026 23:31:20 +0000

those two stipulations were always their only ones, and they were included explicitly in their original contract with the DoW.

New comment by Tenobrus in "ASCII-Driven Development"

Tenobrus — Sun, 11 Jan 2026 07:06:35 +0000

some ai detectors work now. pangram detects this as 57% AI written, and the parts it thinks are human are.... the ascii diagrams / screenshots. all the actual text it detects as generated.

New comment by Tenobrus in "“Erdos problem #728 was solved more or less autonomously by AI”"

Tenobrus — Sat, 10 Jan 2026 02:46:48 +0000

strongly think you should go read the thread to get a sense of the level of expertise and amount of effort put in by the humans involved: https://www.erdosproblems.com/forum/thread/728#post-2852

New comment by Tenobrus in "We replaced H.264 streaming with JPEG screenshots (and it worked better)"

Tenobrus — Wed, 24 Dec 2025 01:25:04 +0000

ai detectors are never totally accurate but this one is quite good and it suggests something like 80% of this article is llm generated. honestly idk how you didn't get that just by reading it tho, maybe you haven't been exposed to much modern llm-generated content?

https://www.pangram.com/history/5cec2f02-6fd6-4c97-8e71-d509...

New comment by Tenobrus in "Show HN: I made a site that lets you talk to the original GPT-1"

Tenobrus — Wed, 13 Aug 2025 22:23:21 +0000

A lot of people forget how whimsical and strange and beautiful the old smaller GPT models and the original 3 base model pre-RHLF could be. Nowadays hundreds of millions of people have talked to heavily assistant tuned 4 or 5, but comparatively very few people have ever even seen GPT-1 outputs. It's cheap to run so I threw up a simple interface + single server hosting it.

Show HN: I made a site that lets you talk to the original GPT-1

Tenobrus — Wed, 13 Aug 2025 22:23:21 +0000

Article URL: https://www.retrogpt.io/

Comments URL: https://news.ycombinator.com/item?id=44894590

Points: 2

# Comments: 1

New comment by Tenobrus in "Backdoor in upstream xz/liblzma leading to SSH server compromise"

Tenobrus — Fri, 29 Mar 2024 17:51:14 +0000

It looks like the person who added the backdoor is in fact the current co-maintainer of the project (and the more active of the two): https://tukaani.org/about.html

New comment by Tenobrus in "Neural Network Diffusion"

Tenobrus — Thu, 22 Feb 2024 01:09:46 +0000

it is very clear to me that humans do in fact have a recursive self-improvement ability, and i'm confused why you think otherwise

Continuum – Generative Software Insights

Tenobrus — Thu, 28 Sep 2023 22:55:12 +0000

Article URL: https://twitter.com/UseContinuum/status/1707516868628734149

Comments URL: https://news.ycombinator.com/item?id=37697020

Points: 2

# Comments: 0

New comment by Tenobrus in "Building a Virtual Machine Inside ChatGPT"

Tenobrus — Sun, 04 Dec 2022 02:05:16 +0000

all this capability was already easily available in GPT-3 with some prompt engineering, this mostly just changed it from few-shot to zero-shot.

New comment by Tenobrus in "Cognitive ability is related to supporting freedom of speech (2020)"

Tenobrus — Sat, 10 Sep 2022 17:19:57 +0000

You're talking the US first amendment and how freedom of speech relates to governments, but that's only a limited subset of the general concept of free speech. Obviously discussions around platforms involve more complexity, but for example if AT&T started refusing to send my texts based on what kinds of ideas I was arguing for then this is a serious attempt to limit speech and discussion which is morally incorrect, regardless of legalities.

New comment by Tenobrus in "Excuse me but why are you eating so many frogs"

Tenobrus — Thu, 08 Sep 2022 16:25:50 +0000

The main interesting thing about GPT-3 is that prompt engineering turns it into an incredibly general tool. This summary was likely generated with a prompt something like:

Here's an article about x:

A quick summary of the article, focusing on the main relevant points and keeping critical detail:

as an example, then duplicating that with the real article and text.

That's the "few shot learning" from the original paper. It could also be that it's good enough at summarizing specifically that you don't need an example, just the right framing prompt around the article text. Either way, that kind of prompt engineering is how you get "text completion" to perform basically any text processing task, generate code or play tic tac toe, so on

New comment by Tenobrus in "Illustrating Gutenberg library using Stable Diffusion"

Tenobrus — Sun, 04 Sep 2022 18:43:17 +0000

I find stable diffusion requires a lot more prompt engineering than either midjourney or dalle, but has a higher ceiling of quality if you do put in that effort. I wouldn't expect just using exactly the same prompt for all three models would work very well, need at least a little tuning per-model.

New comment by Tenobrus in "An incident impacting 5M accounts and private information on Twitter"

Tenobrus — Tue, 09 Aug 2022 18:44:19 +0000

I mean in practice what it tends to mean is the logs only had a 3 month ttl so really could be either way. "no evidence" implies there is at least a place there could have been evidence, they looked, and didn't find any, which is a weak but nonzero update towards it having not happened. It would be nice if they clarified exactly what they checked.

New comment by Tenobrus in "Web3 doesn’t care about privacy"

Tenobrus — Sun, 06 Feb 2022 07:21:09 +0000

This isn't the case with (long-standing, extremely well tested) ZK technology like Monero, and future applications like privacy-focused ZK rollups. In these chains there's no feasible way to track even single transactions.

New comment by Tenobrus in "The Web3 Fraud"

Tenobrus — Fri, 17 Dec 2021 05:32:17 +0000

No actual blockchain will be able to (even in theory) take down DMCA'd data. There is no mechanism whatsoever for complying with such requests, anymore than there's a mechanism for reversing transactions associated with activity a government deems illegal.

Individuals running IPFS gateways or nodes may be directly targeted and asked to take down data, but there's no mechanism for "the network as a whole" to remove data. It's exactly the same as bittorrent. So long as one individual in some hard-to-reach jurisdiction continues hosting the data, nothing can be done.

New comment by Tenobrus in "Vitalik Buterin Proposes New EIP to Tackle Ethereum’s Sky-High Gas Fees"

Tenobrus — Mon, 29 Nov 2021 16:39:30 +0000

It's simply privileging calldata with lower gas (changing a constant) plus adding a cap on the maximum amount of calldata that can be in a given block to prevent crazy state bloat. It's effectively baking in L2s as something the protocol wants to encourage by giving them a "tax break". Note that the gas market is still in effect, so this is "tuning call data against the other costly aspects of ETH tx processing" rather than some "magic price reduction".