Nostr notes by Max Kaehn

What worries me is the humans who uncritically accept the output ...

2026-08-01T21:57:55Z

In reply to nevent1q…unxm
_________________________

What worries me is the humans who uncritically accept the output of the slop engines, whether it’s prose or code. These things are not trustworthy! If you are not enough of an expert to catch their mistakes, you can get into trouble! I prefer to think of them as a reference librarian who has just downed a heroic dose of psychedelic mushrooms: they might give you something useful but you _absolutely_ need to check the sources they cite.

For writing gaming scenarios for my friends, I don’t use it. If ...

2026-08-01T21:35:54Z

In reply to nevent1q…wrc2
_________________________

For writing gaming scenarios for my friends, I don’t use it. If I need to name NPCs, https://www.behindthename.com/ is great. I tried throwing all my campaign notes into NotebookLM to see how it performs and it’s reasonable if you treat it as a search engine over a very limited corpus, but it’s just for finding the footnotes; the summaries are often wrong.

I think modest productivity gains are possible with this ...

2026-08-01T21:32:21Z

In reply to nevent1q…3pjf
_________________________

I think modest productivity gains are possible with this technology, but it’s for boring boilerplate tasks like “write the straightforward unit test cases for this class” and “summarize and group all the issues from the error logs for this service”. It’s bad at writing _eloquent_ code that’s easy to understand and maintain, and I think it is not yet clear just how much technical debt is incurred by vibe coding.

They want us vibe coding, so I follow the elephant-goldfish model ...

2026-08-01T21:27:27Z

In reply to nevent1q…ujmy
_________________________

They want us vibe coding, so I follow the elephant-goldfish model https://research.google/pubs/elephants-goldfish-and-the-new-golden-age-of-software-engineering/ and clean up after the slop engines’ output until I’m not embarrassed to send the changelist out for review.

I try to focus on tools that make this somebody else’s problem. ...

2026-08-01T21:25:53Z

In reply to nevent1q…wf58
_________________________

I try to focus on tools that make this somebody else’s problem. https://modelcontextprotocol.io/ servers and https://adk.dev/ agents that run on someone else’s token budget and provide useful tools— since I work on test infrastructure, ways to interact with the systems under test and ground the slop engines’ work with cross-checks against something practical.

Best-of-N? More tokens. Constrain the outputs to a schema so you ...

2026-08-01T21:22:43Z

In reply to nevent1q…4jdr
_________________________

Best-of-N? More tokens. Constrain the outputs to a schema so you can rerun the prompt if it fails to match the schema? More tokens. Orchestrated agent swarms to cross-check each other’s work? More tokens.

One pattern that is clear is that (a) tokens, the fuel of the ...

2026-08-01T21:21:22Z

In reply to nevent1q…50gr
_________________________

One pattern that is clear is that (a) tokens, the fuel of the slop engines, are currently subsidized and are going to become more expensive and (b) all the things that people come up with to mitigate the unreliability of the slop engines are ways to spend more tokens.

I am fond of getting paid and having health insurance, so I do ...

2026-08-01T21:19:14Z

I am fond of getting paid and having health insurance, so I do use AI at work. I am very fortunate that the AI booster on my team has accepted me as the loyal opposition: the attitude I take is “this technology is overhyped and we need to figure out what the Plateau of Productivity looks like and aim for that”. https://en.wikipedia.org/wiki/Gartner_hype_cycle

Here’s Nero loafing on a comfy blanket: ...

2026-05-20T22:36:54Z

In reply to nevent1q…8cgj
_________________________

Here’s Nero loafing on a comfy blanket:

We did things very fast when FDR took office. We could recover ...

2025-12-31T19:39:57Z

In reply to nevent1q…q8j8
_________________________

We did things very fast when FDR took office. We could recover from this mess quite quickly if we get our collective act together and make rich people unhappy.

I’m 54 and have to read an analog clock by figuring out the ...

2025-12-28T22:19:27Z

In reply to nevent1q…f9sq
_________________________

I’m 54 and have to read an analog clock by figuring out the numbers it’s pointing to. My mother is 82 and has to read digital clocks by imagining the hands of a clock in her head.

If I am ever fashionable it is in the same way that a stopped ...

2025-12-24T02:18:12Z

In reply to nevent1q…xrd9
_________________________

If I am ever fashionable it is in the same way that a stopped clock is right twice per day.

It's not that the occasional curse word pops out or that I ...

2025-12-18T21:13:52Z

In reply to nevent1q…a7jr
_________________________

It's not that the occasional curse word pops out or that I constantly cuss a blue streak. I try to carefully reserve profanity for special occasions so that when I use it, people who know me sit up and take notice.

I would characterize it as "infinitesimal". Calculus ...

2024-08-25T18:50:32Z

In reply to nevent1q…72h3
_________________________

I would characterize it as "infinitesimal". Calculus shows us that infinitesimals add up.