Hacker News: whiplash451

New comment by whiplash451 in "Ask HN: What Are You Working On? (April 2026)"

whiplash451 — Sun, 12 Apr 2026 21:03:35 +0000

This sounds awesome. Have you checked how it fares against trainline? A quick demo would be very nice.

New comment by whiplash451 in "We found an undocumented bug in the Apollo 11 guidance computer code"

whiplash451 — Tue, 07 Apr 2026 13:00:50 +0000

My guess is that in such low memory regimes, program length is very loosely correlated with bug rate.

If anything, if you try to cram a ton of complexity into a few kb of memory, the likelihood of introducing bugs becomes very high.

New comment by whiplash451 in "Iran strikes leave Amazon availability zones "hard down" in Bahrain and Dubai"

whiplash451 — Sat, 04 Apr 2026 10:26:36 +0000

> Nobody even directly dies

People almost directly start dying if data centers go down.

Not in the minutes, but within days definitely

New comment by whiplash451 in "Claude Code Unpacked : A visual guide"

whiplash451 — Wed, 01 Apr 2026 13:45:14 +0000

Were human organizations (not individuals) any good at the latter anyway?

New comment by whiplash451 in "OpenAI closes funding round at an $852B valuation"

whiplash451 — Tue, 31 Mar 2026 21:47:55 +0000

It makes sense for such a huge amount to be "committed", not sitting idle in a bank somewhere.

New comment by whiplash451 in "Do your own writing"

whiplash451 — Mon, 30 Mar 2026 20:11:24 +0000

Worse. You’re being steered along a circle

New comment by whiplash451 in "Philly courts will ban all smart eyeglasses starting next week"

whiplash451 — Mon, 30 Mar 2026 06:46:34 +0000

It’s worse than this. A company offering “private” smart glasses could slip into FB mode on its own or get acquired. So it’s a hard no from any company really.

New comment by whiplash451 in "I quit editing photos"

whiplash451 — Tue, 24 Mar 2026 14:57:36 +0000

> If you want to limit yourself to 36 unreviewed shots, you can do that with digital too.

I’m not sure that’s true. At least, not nearly as hard-constrained as with film.

I agree with your broader point, but let’s be completely honest. Digital is not a free lunch. You do lose something somewhere.

The medium you use “leaks” deeply into the whole experience of life (be it a vacation trip or something else). So all of this is a big deal.

New comment by whiplash451 in "Our commitment to Windows quality"

whiplash451 — Sat, 21 Mar 2026 12:43:47 +0000

I would love for you to be right. Unfortunately in many cases the “windows” decision is made by C-levels who barely understand what you are talking about (or if they do, pretend they don’t to save their jobs)

New comment by whiplash451 in "ArXiv declares independence from Cornell"

whiplash451 — Fri, 20 Mar 2026 14:07:02 +0000

I'm not sure why we're so focused on filtering what gets into arxiv (which is an uphill battle and DOA at this point) vs fixing the indexing, i.e. the page rank of academia.

Google "sorted out" a messy web with pagerank. Academic papers link to each others. What prevents us from building a ranking from there?

I'm conscious I might be over-simplifying things, but curious to see what I am missing.

New comment by whiplash451 in "Scaling Karpathy's Autoresearch: What Happens When the Agent Gets a GPU Cluster"

whiplash451 — Thu, 19 Mar 2026 23:21:46 +0000

Why set the bar higher on generalization for autoresearch vs the research humans generally do?

Show HN: Gumbel-mcts, a high-performance Gumbel MCTS implementation

whiplash451 — Thu, 19 Mar 2026 13:01:12 +0000

Hi folks,

Over the past few months, I built an efficient MCTS implementation in Python/numba.

As I was building a self-play environment from scratch (for learning purposes), I realized that there were few efficient implementation of this algorithm.

I spent a lot of time validating it against a golden standard baseline [1].

My PUCT implementation is 2-15X faster than the baseline while providing the exact same policy.

I also implemented a Gumbel MCTS, both dense and sparse. The sparse version is useful for games with large action spaces such as chess.

Gumbel makes much better usage of low simulation budgets than PUCT.

Overall, I think this could be useful for the community. I used coding agents to help me along the way, but spent a significant amount of manual work to validate everything myself.

Feedback welcome.

[1] https://github.com/michaelnny/alpha_zero/blob/main/alpha_zer...

Comments URL: https://news.ycombinator.com/item?id=47438657

Points: 1

# Comments: 0

New comment by whiplash451 in "What 81,000 people want from AI"

whiplash451 — Thu, 19 Mar 2026 06:42:28 +0000

The writing is on the wall, so to speak.

The number 1 ask from the interviewed cohort is « professional excellence »

It is telling about what we prioritize in our society.

I am usually an optimistic person, but I struggle to see how this does not end up with more misery and worse lifestyle all around.

New comment by whiplash451 in "Kagi Translate now supports LinkedIn Speak as an output language"

whiplash451 — Tue, 17 Mar 2026 05:37:45 +0000

This is great. I'm wondering what led LinkedIn's language to become what it became. Was it "pure self-evolution" among humans who naturally evolved to speak "LinkedIn", or did LinkedIn add an explicit term in the loss of its recommender systems? Asking seriously.

New comment by whiplash451 in "Ask HN: How is AI-assisted coding going for you professionally?"

whiplash451 — Mon, 16 Mar 2026 15:15:47 +0000

That's a different situation than the one I had in mind. I was assuming a sane culture that balances shipping features and quality work. What you're describing sounds like a serious value function mismatch.

New comment by whiplash451 in "Ask HN: How is AI-assisted coding going for you professionally?"

whiplash451 — Sun, 15 Mar 2026 21:19:06 +0000

Shipping “quality only” work for a long time can be stressful for your colleagues and the product teams.

You’re much better off mixing both (quality work and product features).

New comment by whiplash451 in "Coding after coders: The end of computer programming as we know it?"

whiplash451 — Sat, 14 Mar 2026 18:57:44 +0000

Local AI is what people want/need, but centralized AI is where the investors' money is flowing, because a walled garden has always been easier to turn into a cash printer.

New comment by whiplash451 in "A Survival Guide to a PhD (2016)"

whiplash451 — Sat, 14 Mar 2026 14:36:04 +0000

It's pretty common in ivy-leave US universities (which is what the article is biased towards). There, you only have to TA a bit and you certainly don't work for the uni or the dept.

New comment by whiplash451 in "A Survival Guide to a PhD (2016)"

whiplash451 — Sat, 14 Mar 2026 14:33:10 +0000

This is very good advice for a few reasons:

1. It reduces the odds of missing a key reference in your papers and accelerates the write-up of the (often mandatory) Related Work section

2. It helps you maintain a mental map of the field as your research progresses

New comment by whiplash451 in "A Survival Guide to a PhD (2016)"

whiplash451 — Sat, 14 Mar 2026 14:29:31 +0000

Your point is valid in many ways. The picture can be a little brighter. The PhD path does not have to be an all or nothing.

1. You can hedge your bets by submitting your work to various conferences of various qualities (without going 3rd-tier, you can bet across 1st-tier and 2nd-tier)

2. You can spend time choosing the professor and the topic before going all in

3. You can seek advice and social interactions within your research group, departement and school

None of this is a silver bullet, but it compounds in the right direction.