Hacker News: stevenae

New comment by stevenae in "Pro Max 5x quota exhausted in 1.5 hours despite moderate usage"

stevenae — Sun, 12 Apr 2026 18:23:32 +0000

The quantitative ux research team at Google was created for exactly this problem: a service which became popular before the right metrics existed, meaning metrics need to be derived first, then optimized. We would observe users (irl), read their logs, then generate experiments to improve the behavior as measured by logs, and return to see if the experiment improves irl experiences. There were not many of us and we are around :)

New comment by stevenae in "Show HN: The Hessian of tall-skinny networks is easy to invert"

stevenae — Thu, 15 Jan 2026 22:53:51 +0000

This helped me, coming from an ml background: https://randomrealizations.com/posts/xgboost-explained/

New comment by stevenae in "Show HN: OfferGridAI – side-by-side comparison of real estate offers from PDFs"

stevenae — Fri, 02 Jan 2026 16:22:56 +0000

Others mentioned county data. If you can get that, you can build something like I did for DC -- https://colab.research.google.com/drive/1Kep_9j_PN_SxX85PYHE...

New comment by stevenae in "Zero-Shot Forecasting: Our Search for a Time-Series Foundation Model"

stevenae — Sat, 14 Jun 2025 14:08:46 +0000

My reading of this situation is that MAPE would do the opposite. Means are skewed towards outliers.

New comment by stevenae in "Zero-Shot Forecasting: Our Search for a Time-Series Foundation Model"

stevenae — Fri, 13 Jun 2025 15:23:51 +0000

Thanks for the reply! I am outside the forecasting sphere.

RMSLE gives proportional error (so, scale-invariant) without MAPE's systematic under-prediction bias. It does require all-positive values, for the logarithm step.

New comment by stevenae in "Zero-Shot Forecasting: Our Search for a Time-Series Foundation Model"

stevenae — Fri, 13 Jun 2025 12:58:33 +0000

To clarify, you'd prefer rmsle?

New comment by stevenae in "Deep learning gets the glory, deep fact checking gets ignored"

stevenae — Wed, 04 Jun 2025 15:27:36 +0000

For this and sibling -- yes. Essentially, using the output of any model as an input to another model is transfer learning.

New comment by stevenae in "Deep learning gets the glory, deep fact checking gets ignored"

stevenae — Wed, 04 Jun 2025 12:24:13 +0000

> Lately, I just steal embeddings from big models and slap a dumb classifier on top. Works better, runs faster, less drama.

You may know this but many don't -- this is broadly known as "transfer learning".

New comment by stevenae in "Former Supreme Court justice David Souter has died"

stevenae — Fri, 09 May 2025 20:02:43 +0000

Thank you!

New comment by stevenae in "Former Supreme Court justice David Souter has died"

stevenae — Fri, 09 May 2025 17:22:27 +0000

How accurate is his claim that Augustus became emperor through (my paraphrasing) democratic means and promises to fix real problems for Romans?

New comment by stevenae in "Silicon Valley, Halt and Catch Fire, and How Microserfdom Ate the World (2015)"

stevenae — Wed, 02 Apr 2025 13:08:55 +0000

This still strikes me as escapism.

New comment by stevenae in "Ask HN: Any insider takes on Yann LeCun's push against current architectures?"

stevenae — Wed, 12 Mar 2025 22:25:28 +0000

https://en.m.wikipedia.org/wiki/Energy-based_model

New comment by stevenae in "Ask HN: Advice for tolerating the "corporate" aspect of software jobs?"

stevenae — Wed, 12 Mar 2025 12:28:30 +0000

There was a saying at Google, I code for free, they pay me for XYZ (literally everything else).

New comment by stevenae in "Cameras of 1930s Era"

stevenae — Thu, 06 Feb 2025 13:47:44 +0000

I guess my quibble is with the percentage, then. A good, cheap, plentiful camera belies the idea that only the top 0.1% of cameras were good.

New comment by stevenae in "Cameras of 1930s Era"

stevenae — Thu, 06 Feb 2025 12:32:02 +0000

Disagree with the first piece about only using the top 0.1%. I grew up (through my 20's) shooting on a Pentax K1000, cheap workhorse of a camera, and I preferred its ergonomics to top-end mirrorless cameras I use today.

New comment by stevenae in "Ask HN: Who wants to be hired? (January 2025)"

stevenae — Sat, 04 Jan 2025 22:42:14 +0000

Location: Washington DC USA

Remote: Hybrid or Remote

Willing to relocate: NYC

Technologies: r, python, sql

Résumé/CV: https://www.linkedin.com/in/steven-ellis-4b140533

Data scientist looking to continue product-focused work!

New comment by stevenae in "A “meta-optics” camera that is the size of a grain of salt"

stevenae — Fri, 22 Nov 2024 13:27:40 +0000

Pro cameras do not do this to any degree.

Edit: by default.

New comment by stevenae in "Is Python That Slow?"

stevenae — Fri, 22 Nov 2024 12:35:59 +0000

This is not true. R shines for classical stats and ML. If you are doing deep learning, you need Python.

New comment by stevenae in "Python, C++ inspired language that transpiles to C and can be embedded within C"

stevenae — Thu, 14 Nov 2024 19:25:46 +0000

From the guidelines [1]: Please don't sneer, including at the rest of the community.

1. https://news.ycombinator.com/newsguidelines.html

New comment by stevenae in "Guys Hacked AirPods to Give Their Grandmas Hearing Aids"

stevenae — Thu, 14 Nov 2024 18:59:42 +0000

Save you some scrolling (link directly to comment):

https://news.ycombinator.com/item?id=42119697