Hacker News: TrainedMonkey

New comment by TrainedMonkey in "What AI did to stackoverflow in a graph"

TrainedMonkey — Sat, 18 Jul 2026 17:53:45 +0000

I think it's all about incentives. The power to governance was given to prolific and tenured accounts who wanted to govern. Over the long timeframe their incentive averages to making their life easier and keeping the governance. Wikipedia is going through something similar.

New comment by TrainedMonkey in "Kimi K3: Open Frontier Intelligence"

TrainedMonkey — Fri, 17 Jul 2026 03:11:34 +0000

Worry not, Pelicans on bicycles had been ranking pretty high on your favorite search engine for a while. I struggle to imagine a world in which it was not already scraped and turned into training data by at least one provider:

1. Models need to be good at the questions we ask them, not the questions we could ask them.

2. The questions, at least partially, are correlated with information people consume.

3. People mostly consume viral content.

4. Ergo you should scrape viral content for training data.

New comment by TrainedMonkey in "A dock that wakes up reliably"

TrainedMonkey — Sun, 12 Jul 2026 01:30:06 +0000

The author does not call it out, but I think the point is that it's not only the dock. I've been struggling with MBP wakeup issues for a while. We have a whole zoo of various apple devices attached to a number of docks. All of the monitors are Odyssey Neo G9. TL:DR an issue with MBP port, MBP to Dock link, Dock, Dock to Monitor link, or the monitor itself could cause the unplugging dance.

In addition, it takes some time for Apple to squash bugs after releasing new hardware or software. M5s used to be unreliable with Anker TB4 until 1-2 months ago.

New comment by TrainedMonkey in "SpaceX to buy Cursor for $60B"

TrainedMonkey — Tue, 16 Jun 2026 15:27:14 +0000

Be honest now, it would take at least a few months for the rest of us.

New comment by TrainedMonkey in "Further human + AI + proof assistant work on Knuth's "Claude Cycles" problem"

TrainedMonkey — Sat, 28 Mar 2026 19:35:51 +0000

My understanding is that, if confirmed, this demonstrates that AI can find novel solutions. This is a strong counterpoint to generative-AI-is-strictly-limited-to-training-data.

New comment by TrainedMonkey in "AI overly affirms users asking for personal advice"

TrainedMonkey — Sat, 28 Mar 2026 18:27:26 +0000

It's almost like it is based on the training data and regimen that is largely the same between versions.

New comment by TrainedMonkey in "Warranty Void If Regenerated"

TrainedMonkey — Thu, 19 Mar 2026 00:42:17 +0000

I really enjoyed fantasy part of many small farmers. It felt rustic. However based on my understanding the modern world is moving towards megacorps and economies of scale.

New comment by TrainedMonkey in "Book: The Emerging Science of Machine Learning Benchmarks"

TrainedMonkey — Thu, 19 Mar 2026 00:17:32 +0000

Why is that?

New comment by TrainedMonkey in "Review: Samsung Galaxy S26 Ultra"

TrainedMonkey — Wed, 18 Mar 2026 20:32:03 +0000

> it doesn't dramatically reduce screen brightness or image quality.

AFAIK it significantly decreases the brightness. Jerry Rig Everything demonstrates this here - https://youtu.be/TRW4W7KkJXs?t=32

New comment by TrainedMonkey in "Garry Tan's Claude Code Setup"

TrainedMonkey — Tue, 17 Mar 2026 21:49:04 +0000

Yes

New comment by TrainedMonkey in "AirPods Max 2"

TrainedMonkey — Mon, 16 Mar 2026 21:00:08 +0000

Servicing is not done by Apple, it's 3rd party contractors. They have a rubric of possible issues from Apple and their profit margins are thin. I suspect contacting Apple support about Apple support issues would have resulted in a swift replacement of the item.

New comment by TrainedMonkey in "GPT‑5.3 Instant"

TrainedMonkey — Tue, 03 Mar 2026 19:09:41 +0000

Will need to wait for real benchmarks, but based on OpenAI marketing Instant is their latency optimized offering. For voice interface, you don't actually need high tok/s because speech is slow, time to first token matters much more.

New comment by TrainedMonkey in "iPhone 17e"

TrainedMonkey — Mon, 02 Mar 2026 21:00:48 +0000

Apple seems to be pushing for accessibility and volume. Cheaper phones, mac minis, and entry point mac that will be introduced on Wednesday.

New comment by TrainedMonkey in "You don't have to"

TrainedMonkey — Mon, 02 Mar 2026 04:39:09 +0000

Exceptionalism says we have best of everything, including idiots.

New comment by TrainedMonkey in "You don't have to"

TrainedMonkey — Mon, 02 Mar 2026 04:20:13 +0000

What if, and hear me out here, "You don't have to"

New comment by TrainedMonkey in "Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers"

TrainedMonkey — Sun, 01 Mar 2026 00:29:21 +0000

No, what he is saying is that benchmarks are static and there is tremendous reputational and financial pressure to make benchmark number go up. So you add specific problems to training data... The result is that the model is smarter, but the benchmarks overstate the progress. Sure there are problem sets designed to be secret, but keeping secrets is hard given the fraction of planetary resources we are dedicating to making the AI numbers go up.

I have two of my own comments to add to that. First one is that there is problem alignment at play. Specifically - the benchmarks are mostly self-contained problems with well defined solutions and specific prompt language, humans tasks are open ended with messy prompts and much steerage. Second is that it would be interesting to test older models on brand new benchmarks to see how those compare.

New comment by TrainedMonkey in "A 26-Gram Butterfly-Inspired Robot Achieving Autonomous Tailless Flight"

TrainedMonkey — Thu, 26 Feb 2026 17:38:31 +0000

You can chain normalized quaternions to combine or diff transformations. For example you can subtract desired attitude quaternion from predicted attitude quaternion to get attitude error quaternion which you can then feed to control algs designed for driving that error to zero. This is even more important when multiple frames of reference are involved as quaternions can be used to transform between them.

New comment by TrainedMonkey in "In 2025, Meta paid an effective federal tax rate of 3.5%"

TrainedMonkey — Thu, 26 Feb 2026 17:28:00 +0000

Sure, and there are a ton of ways to shifting income around. For example selling a subsidiary in lower tax jurisdiction patents and then paying for their usage. Another example is Hollywood accounting where productions pay exorbitant rates for equipment and catering to affiliated companies. This inflates the costs so the movies end up unprofitable despite smashing box office.

New comment by TrainedMonkey in "GPU Rack Power Density, 2015–2025"

TrainedMonkey — Tue, 24 Feb 2026 03:22:04 +0000

Huh, that is kind of amazing. It turns out the problem is that we got way too good at scaling semiconductor density.

New comment by TrainedMonkey in "TikTok is tracking you, even if you don't use the app"

TrainedMonkey — Thu, 12 Feb 2026 19:54:00 +0000

It could be as simple as links. People drop links in the slack discussions, other people from Geolocated IP addresses (or same) click on them. Google analytics et. al. hovers a lot of data.