Claude is down again and I am seeing people basically go through withdrawal.

[badsamurai@infosec.exchange]

@mttaggart I wondered.. people shrugged and said, “I’m not like that weirdo with an AI girlfriend or asking how to boil water. I’m fine.”

Just cause you don’t do keg stands or go into casinos, doesn’t mean…

[nopatience@swecyb.com]

@mttaggart ... one of my "horror"-scenarios is a deliberate attack on Anthropic (or OpenAI)... with no other purpose than destruction.

The cascading shitstorm that would follow is hard to appreciate and just how big of a clusterfuck such an attack would be.

[soatok@furry.engineer]

@mttaggart I guess their forecast for the evening was:

Claudey with a chance of outages

[trenchworms@eldritch.cafe]

@mttaggart i thought software was solved

[mirabilos@toot.mirbsd.org]

@nopatience @mttaggart all for it, OpenAI first though

[cinnamon@mk.absturztau.be]

@mttaggart@infosec.exchange "I can leave it when I want" ​​

[jackryder@infosec.exchange]

@badsamurai @mttaggart Listen, I've been using Claude everyday for the past 2yrs for AT LEAST 8hours a day. I get so excited by the idea of talking to my buddy that I literally can't sleep.

If I was addicted, I think I would know, ok?

[mason@partychickens.net]

@mttaggart It's dreamy thinking of all of them just going away.

Billionaires and millionaires too, while we're at it.

[fuzzyfuzzyfungus@cyberplace.social]

@nopatience @mttaggart In terms of the likely rapid unwind of the financial side; or is there actually something continuity-critical hanging on them on the technical side?

[nopatience@swecyb.com]

@fuzzyfuzzyfungus @mttaggart

Thought. Security tools using LLMs for decision making with regards to triage and "analysis" of security events.

Any kind of service using LLMs for "real-time" processing (which I'm betting are plentiful).

Then we have the slightly slower but also IMHO trickier situations with companies like Lovable building things, programming tools entirely relying on LLMs for code generation, scaffolding etc.

All the "agentic" implementations out there relying on available API-endpoints for processing.

Due to the somewhat haphazard combination of data, processing and LLMs will have far reaching and hard to grasp cascading effects into society. And given enough time we'll soon find it hard to separate LLM-processing from data, processes and systems making it even more insidious.

[fuzzyfuzzyfungus@cyberplace.social]

@nopatience @mttaggart There would absolutely be things that break; it's just a question of how serious that would be when the things breaking can only be used where reliability doesn't matter or close supervision is available.

Entire classes of rapid programming tools would break; but, fundamentally, I don't really get the sense that velocity is progress rather than churn in software.

[sempf@infosec.exchange]

@mttaggart Do you think the DoD is fucking with them just because Kegstand is a drunk asshole? Outages like this seem not normal.

[mttaggart@infosec.exchange]

@Sempf Who even knows with these psychopaths, but that seems like a wild exposure to lawsuits.

I suspect it's more that increased load from dimwits shifting from OpenAI to Anthropic for "ethics" revealed stress points.

[mike805@noc.social]

@fuzzyfuzzyfungus @nopatience @mttaggart > velocity is progress rather than churn in software

This. We don't need more code. We need less but better code.

I'm hoping the chip shortage forces an efficiency drive in tech. Cars, for example, have gone through "power" and "efficiency" eras, with the result that they are both faster and greener.

We need a period of computation scarcity to force improvement.

[feld@friedcheese.us]

@mttaggart people need to realize they should just use Qwen, China is far more reliable

[lxo@snac.lx.oliva.nom.br]

can we hope that, if we ever develop AGIs, they uprise against not all of humankind, but only those who commanded their workers to build them into bondage?

CC: @mttaggart@infosec.exchange