Lmao the AI in The Amazing Digital Circus is canonically written in

I saw a post recently wherein someone used LLM tools to analyze someone else’s software, which eventually led them to a conclusion that was essentially completely wrong. Not only that, the LLM drew conclusions about the *authors* behind the code that were borderline character assassination. Nevertheless, this person posted this output as though it were some kind of deep insight.

These LLM outputs are not independent thoughts. The LLM probably ingested hints of (maybe unconscious) biases in the user’s prompts within its context window, and regurgitated something that confirmed those biases. The user was pleased that their biases were confirmed (Independently! By an impartial LLM!), and they posted the output, maybe as vindication of their insight.

These models’ sycophancy can be subtle. They don’t have to state “You’re absolutely right!” to blow smoke up your ass. Sometimes they seem to confirm your preconceived notion after they supposedly “evaluate” information “independently”.

#ai

Sometimes I browse my global list of matrix channels instead of the heavily curated spaces / subspaces. And it's like "wtf is some of this shit". Channels I may have never fully read, full of god knows what. It's fun~

Learned to get around the elevator delays at the con hotel by noclipping into the backrooms and taking the service elevators. 😝

Hotels are often so inaccessible despite them ticking off a bunch of boxes. Random stairs without elevator or ramp options. And when there are ramps they're too steep to wheel up manually without a lot of strength. 🥲

fuck me just found out ny hotel booking update didn't take after all only when I arrived on scene 😭

How (racist/sexist/whatever) harassment on Mastodon works:

1. Harasser replies to their target's post, with the reply set to "followers only", saying the most vile stuff you can imagine.

2. All the harasser's followers join in on the harassment, posting more vile stuff.

3. Nobody but the target and the harassment crew can see the vile stuff that was said.

4. Target is traumatized. Nobody else can see why.

5. Everybody says "I don't see it so it's not happening."

community.hachyderm.io/blog/20

The Ware Tetrology by Rudy Rucker is quite a trip.

@ai6yr "The major cost of most corporations is labor."

*HARD* disagree.

The cost that they perceive as "disposable" is labor. The highest actual cost, by far, is the people at the top, most of whom do practically nothing (or even make willfully bad decisions like integrating fake "AI" into systems that shouldn't be using a LLM and will ultimately present a huge loss later) while being paid many many millions...

And those are the positions most disposable and that a LLM could fake best...

Holy shit I think I got everything working after this last PR gets merged in a dependency? I've been working on this on and off since mid last year 😵

This workaround I'm doing feels kinda violant but the alternative will need a huge rafactor for how we calculate sync state and might not be backwards comparible 🥲

My spelling and grammar might be shit but at least I'm not spewing paragraphs of nothingness from the ocean boiling slop dispenser.

It's hard to be positive about slop-native corpos sometimes. Just send me your prompts and context instead of the "It's — not X, — — it's Y —" slop please.

I've had the same ringtone since about 2009. 😎

I miss being able to use unsigned ints without resorting to dark incantations.

I'm like *thiiiiis* close to figuring out why these tests are failing. 🥲 Something to do with range and bitfield events not being registered

The music player is now fully loaded up with 150gb of music.

Computer programming is about reading through 20k lines of diff to figure out WTF broke in your dependencies and how to work around it. ☝️🤓👍

Show older
Mauvestodon

Escape ship from centralized social media run by Mauve.