Hacker News: thatguysaguy

New comment by thatguysaguy in "Chip Off The Old Block"

thatguysaguy — Thu, 02 Jul 2026 00:58:20 +0000

Scott's parenting posts are some of his best

New comment by thatguysaguy in "Show HN: TownSquare, a tiny presence layer for websites"

thatguysaguy — Sun, 21 Jun 2026 04:40:20 +0000

The contrast between the example screenshots and the standard internet behavior in the live demo is hilarious

New comment by thatguysaguy in "Python JIT project was asked to pause development"

thatguysaguy — Sat, 06 Jun 2026 20:05:59 +0000

The python overhead of launching big ML jobs is nontrivial, so I think speeding that up would be meaningful. (I mean the initial tracing and other setup, not things once the GPUs are actually doing the work).

New comment by thatguysaguy in "xAI Is Reportedly Using Just 11% of Its 550k Nvidia GPUs"

thatguysaguy — Mon, 04 May 2026 15:52:13 +0000

It looks like people in this thread are confusing fleet utilization and MFU. If they're doing a lot of RL, it's really not surprising to see such low numbers.

New comment by thatguysaguy in "Google plans to invest up to $40B in Anthropic"

thatguysaguy — Fri, 24 Apr 2026 21:01:45 +0000

That's b/c the people working on Gemini serving are in GDM.

New comment by thatguysaguy in "Our eighth generation TPUs: two chips for the agentic era"

thatguysaguy — Wed, 22 Apr 2026 18:21:04 +0000

What sort of workloads are you thinking of?

New comment by thatguysaguy in "Lean proved this program correct; then I found a bug"

thatguysaguy — Tue, 14 Apr 2026 01:58:58 +0000

What is up with people saying you cannot prove a negative? Of course you can! (At least in formal settings)

For example it's extremely easy to prove there is no square with diagonals of different lengths. I'm the hard end, Andrew Wiles proved Fermat's Last Theorem which expresses a negative.

That's just a nit though, you're right about the infinite regress problem.

New comment by thatguysaguy in "Can you reverse engineer our neural network?"

thatguysaguy — Fri, 27 Feb 2026 17:20:03 +0000

Ah dang. When I did this I also thought the length bug was intentional but I didn't figure it out before I started my new job, so I dropped the puzzle.

New comment by thatguysaguy in "Large-Scale Online Deanonymization with LLMs"

thatguysaguy — Wed, 25 Feb 2026 23:07:11 +0000

Maybe I missed something, but I see little evidence that there is a concerning ability to deanonymize. Many people post under a pseudonym but then link to their GitHub etc. In fact by construction the HN dataset _only_ consists of people who are comfortable with their real identity being linked to it.

The real question is whether someone who is pseudonymous and actually attempting to remain so can be deanonymized.

New comment by thatguysaguy in "Gemini 3 Deep Think drew me a good SVG of a pelican riding a bicycle"

thatguysaguy — Sat, 14 Feb 2026 20:36:46 +0000

You can just try other svgs, I got some pretty good ones.

(*Disclaimer: I work for Google, but also I have zero idea about what they trained deepthink on)

New comment by thatguysaguy in "TPUs vs. GPUs and why Google is positioned to win AI race in the long term"

thatguysaguy — Fri, 28 Nov 2025 05:48:19 +0000

TPUs predate LLMs by a long time. They were already being used for all the other internal ML work needed for search, youtube, etc.

New comment by thatguysaguy in "AI has a deep understanding of how this code works"

thatguysaguy — Thu, 27 Nov 2025 20:17:56 +0000

I'm actually not talking about whether the PR works or was tested. Let's just assume it was bug-free and worked as advertised. I would say that even in that situation, they should not accept the PR. The reason is that no one is the owner of that code. None of the maintainers will want to dedicate some of their volunteer time to owning your code/the AIs code, and the AI itself can't become the owner of the code in any meaningful way. (At least not without some very involved engineering work on building a harness, and since that's still a research-level project, it's clearly something which should be discussed at the project level, not just assumed).

New comment by thatguysaguy in "AI has a deep understanding of how this code works"

thatguysaguy — Thu, 27 Nov 2025 08:23:19 +0000

A big part of software engineering is maintenance not just adding features. When you drop a 22,000 line PR without any discussion or previous work on the project, people will (probably correctly) assume that you aren't there for the long haul to take care of it.

On top of that, there's a huge asymmetry when people use AI to spit out huge PRs and expect thorough review from project maintainers. Of course they're not going to review your PR!

New comment by thatguysaguy in "FFmpeg dealing with a security researcher"

thatguysaguy — Sun, 02 Nov 2025 06:00:19 +0000

It's a volunteer run project... Saying that they have a duty to do anything other than what they want is quite strange.

New comment by thatguysaguy in "Updated practice for review articles and position papers in ArXiv CS category"

thatguysaguy — Sat, 01 Nov 2025 17:07:08 +0000

Verification via LLM tends to break under quite small optimization pressure. For example I did RL to improve against one of the sota models from one generation ago, and the (quite weak) learner model found out that it could emit a few nonsense words to get the max score.

That's without even being able to backprop through the annotator, and also with me actively trying to avoid reward hacking. If arxiv used an open model for review, it would be trivial for people to insert a few grammatical mistakes which cause them to receive max points.

New comment by thatguysaguy in "Meta is axing 600 roles across its AI division"

thatguysaguy — Wed, 22 Oct 2025 18:47:04 +0000

FAIR is not older AI... They've been publishing a bunch on generative models.

New comment by thatguysaguy in "BERT is just a single text diffusion step"

thatguysaguy — Mon, 20 Oct 2025 16:21:30 +0000

Back when BERT came out, everyone was trying to get it to generate text. These attempts generally didn't work, here's one for reference though: https://arxiv.org/abs/1902.04094

This doesn't have an explicit diffusion tie in, but Savinov et al. at DeepMind figured out that doing two steps at training time and randomizing the masking probability is enough to get it to work reasonably well.

New comment by thatguysaguy in "[dead]"

thatguysaguy — Mon, 06 Oct 2025 15:11:13 +0000

I would recommend going and reading what the BlueSky leadership actually wrote, rather than this post's summary of it.

New comment by thatguysaguy in "Are OpenAI and Anthropic losing money on inference?"

thatguysaguy — Thu, 28 Aug 2025 16:10:18 +0000

Why would you think that deepseek is more efficient than gpt-5/Claude 4 though? There's been enough time to integrate the lessons from deepseek.

New comment by thatguysaguy in "Are OpenAI and Anthropic losing money on inference?"

thatguysaguy — Thu, 28 Aug 2025 14:59:37 +0000

37 billion bytes per token?

Edit: Oh assuming this is an estimate based on the model weights moving fromm HBM to SRAM, that's not how transformers are applied to input tokens. You only have to do move the weights for every token during generation, not during "prefill". (And actually during generation you can use speculative decoding to do better than this roofline anyways).