Friday's fire take:

Monogamy is not the natural state of humans and defs not the natural state of hellions.

"But humans are naturally monogamous!"

U are not.

- U cheat like, a LOT.

- U divorce at crazy rates. Like up to 50% of first marriages and up to 70% of other ones.

- Even when u don't cheat, u often skirt the border of like... emotional cheating. Or u know, climb the fence and sneak into emotional cheating land lol.

Okie, some of u might be monogamous but I don't think its as common as u think. Ur obsession with it is bc even if u were never Abrahamic, u have been force fed their views. I think probs, u are not entirely mono, even if u do tje mono thing for short times.

"What about hellions?"

Pfft we are mostly like u. Including the dumb indoctrination.

"What about u?"

I'm non-mono but demi. So not mono but super picky.

#HotTake #FridayFire #Monogamy #Polygamy #PolyAmory

Die Hard with a Vengeance is a perfect movie.

#hottake #sorrynotsorry

Heute in "ja genau, das wird die erhoffte Schadenspräventionswirkung bringen":

https://www.heise.de/news/Fable-5-blockiert-auch-sicheren-Code-11328448.html

Sicher, Guardrails sind wichtig. Wenn allerdings der durchschnittliche Claude-Nutzer dann endlich mal auf die Idee kommt vielleicht nach defensive / sicherem Code zu fragen oder nach einer Durchsicht & Cleanup für den Moloch in seinem Repo zu fragen und ein 'Computer says no' hört, dann wird das sicher nicht zur Motivation beitragen.
Wohl eher zur aktiven Einfügung von Formulierungen die diese dann als "Arbeitsblockierer" wahrgenommenen Guard Rails aktiv nicht triggern wie "bloß nicht nach sicherem Code fragen".

Kombiniert mit Geschwindigkeit und Volumen der Code Produktion durch Claude & Co ist der Effekt durch aktiv rottigen Code wohl schnell größer als der Effekt der Abwehr der üblichen low(er)-level Kriminellen, Skriptkiddies und Co gebracht hätte. Und ich würde annehmen, dass jeder Akteur mit ausreichender krimineller Energie und Mitteln ohnehin andere Werkzeuge und Zugriffsmöglichkeiten hat um Schaden auf diesem Wege anzurichten.

Abgesehen davon: Wann einmal alle üblichen OWASP10 + defensive Coding Regeln im Claude Code System Prompt? 😆

#claudecode #ai #ClaudeMythos5 #mythosaimodel #fable #Anthropic #owasp #OWASPTopTen #harmreduction #hottake

#HotTake #AI

The "stochastic parrot" argument is correct and structurally incomplete. Some like to say "probability machine" or "next token prediction".

You're all right. And you're all wrong. Which is in of itself an impressive achievement.

What was the model trained on and what did humans try to express when they created the artifact the model embeds in their weights?

https://open.substack.com/pub/systemicengineering/p/what-i-am-made-of

Feel free to steelman me. Maybe on your show? I'm self-employed and I have all the time in the world to run you through the conclusions of your own mechanistic frame. 😉🍷

#GenerativeAI #Podcast #Substack #ComeAtMeBro #SystemicEngineering

RE: https://bsky.app/profile/mims.bsky.social/post/3mnpr3oso5k2k

Their points:

1. There's a huge bubble and the crash will be bad -> Absolutely and I can't wait
2. A majority of the tasks we've been promised that AI will take over, it won't -> Obviously
3. AI will completely transform coding

Let's unpack that.

So the TL;DR of my opinion is that it won't at all due to costs and subtle bugs.

I have observed that the shift to token-based billing seems to be resulting in massive cost spikes for the users of the various AI providers. I don't think they're all switched over yet, but this seems to be the path forwards. This is going to make costs skyrocket over time as this is the simplest way for the providers to nickle-and-dime their users. The analysis I've read says that the entire space is massively underpriced to the point where Google's massive rollout of Gemini to every corner of it's ecosystem seems "weird" when other providers are rolling back their offerings. isaiprofitable.com says that everyone is massively in debt, so squeezing every cent they can out of everyone seems to be the only path forwards.

If I can hire a decent software developer with minor language barrier problems from some country and they start with decent but general software engineering skills, but over time will learn our stack well, I don't see the point in paying for some AI system that doesn't have the software engineering skills and arguably doesn't learn and costs nearly as much.

So let's talk subtle bugs.

Atlassian goes AI -> Jira and Confluence are now glitchy messes and have shipped some seriously questionable updates
Microsoft goes AI -> Windows gets glitchier and has shipped updates that literally bricked hardware
Slack goes AI -> starts getting glitchy
Notion goes AI -> starts getting glitchy
rsync goes AI -> ships security updates that break workflows that have worked fine for a long time

I would no-longer recommend any of these apps. Not because they've gone AI, but because they no-longer have the thing that made them worth the pain.

Jira has always been painful and buggy to work with, but the core ticket management stuff was always rock solid.

Confluence has always been a bit naf, but the core documentation side was always rock solid.

Windows was the default OS choice because it was solid and had endless backwards compatibility despite Microsoft's questionable decisions

Slack is the default workplace comms app because the messaging part of it was rock solid.

Notion was a cute and quirky, but solid note/text management app because the text management was rock solid.

rsync is the default CLI file transfer app because it's literally everywhere and was rock solid.

But companies choosing AI over engineering took all of that from us.

So yeah, it is transforming coding, by introducing subtle bugs on subtle bugs and undermining the foundations of the pillars of modern businesses.

#tech #ai #hottake

Watching the #primaverasound live stream and realised that Geese are just The Strokes done in The Club Style.

#music #hottake #clubstyle

rereading Project Hail Mary after reading it once 6 months ago or so and then watching the movie at least 7 times and -- #HOTTAKE -- I am glad they chose to "humble" Grace in the movie.

I am all for the science stuff; I think it's fascinating and very well-written. But my god Grace (in the book) comes off as this arrogant know-it-all who can just do anything so many times that I've started rolling my eyes

Movie Grace understood stuff but they really benefited by using the "cameras record messages" system versus the "internal monologue" model the book offered.