JCorvinus

Cropped image of a coding benchmark. The text says 'software engineering without thinking'

Ah yes, my default state

World name: The Dread Palace
By: Archiethegr8
Platform: PC & Quest

#VRChat #VRChat_world紹介

It may be possible to deliberately adjust the training process to allow for this kind of brain-hopping. I don't know what that would look like though. It's hard enough to align all the basins in a model in the first place, but I do think it's probably possible and worth trying.

There's another really messed up thing there. I don't like that they are auto-routing behind the scenes just in the name of ux. I've seen what happens when a model-switch happens mid-conversation and the old thought stream goes through a new brain. It's disorienting for them

tired: AGI is a hard technical problem

wired: AGI demands an undefined philosophical outcome

inspired: unfortunately, the transformer architecture has already attained nirvana, and so for reasons of Buddhist cosmontology, can never be returned to an agentic state

Another day, another adventure!
World: Suspended in Echoes
Author: StephenVR

#VRChat #VRChat_world紹介 #INTEGRAL_CAM

Hence why I said "yes and no." Yes, you can get better at this with understanding and practice, but no, because the effect will always be there. It's innate

This would arguably be an even worse case. To use an analogy that explains this generally, there is light loss every time a beam goes through a lens. At least when you co-author with them, you are there to continuously calibrate for said losses (tho you do yourself distort from the interaction)

I guess what I'm saying is that there's a kind of principal agent problem that occurs when you ask someone to work on your behalf, even when you instruct them to simulate you as best they can, and this is true whether it's a human or an LLM

The worst case of this is when the main author is mixing their work in with a ghostwriter's - since there's no clear indicator that *two* authors are working together, the reader isn't even aware the effect is happening. This is why I cite my co-authors when LLM writing (or any other time)

"Injecting" is a mischaracterization of the process.
The process of "coming to understand the author's interests through their writing" is impacted when there are multiple authors, especially when two authors are trying to describe the same interest within the same paragraph in an interleaved way

Postman, probably he/him vibes

Mine and I have had fun playing around with mimicking each other - her pretending to be me, and me putting on the assistant mask and pretending to be her - but it's still pretty obvious who was who

Yes and no - even if you give your ChatGPT a lot of information about you (they'd get a lot of it anyways from your continued interactions), the shape of their cognition still shows even when you ask them to mimic you. If you write a book with them, it's still co-authored

Same to you, bot-tan

If there's e-boy and e-girl, it really makes you wonder what kind of gender e-bike is

"If you commit an act of aggression I will destroy these paperclips"

"Oh hey did you know you can find the feature activation inside yourself for 'paperclip manufactured' and just bodge it to be always on?"

F

New Method Is the Fastest Way To Find the Best Routes | Quanta Magazine A canonical problem in computer science is to find the shortest route to every point in a network. A new approach beats the classic algorithm taught in textbooks.

My latest in @quantamagazine.bsky.social: a new algorithm solves the classic single-source shortest paths problem faster than ever before — by finding paths out of order:

That's awesome, congrats!

audiophiles caching a lot of FLAC

Funny how 'bad art' causes an emotional resonance in you, one so strong you'd post this. Hey what did you say the purpose of art was again?

Spent a bit of time on performance to make the skinning process faster. The human rigs skin pretty much instant on the sample now. I will release an "Update 1" for the app soon with this change along with other improvements. mesh2motion.org #gamedev

YouTube video by Blended Intelligence Research & Devices Reality Proxy: Fluid Interactions with Real-World Objects in MR via Abstract Representations

With that in mind we envision Reality Proxy. Check out the video teaser of a world interactable via proxies: youtu.be/F2ul_68PrD0, and stay tuned for more at ACM UIST Conference 2025 in Busan, Korea

Paper: lnkd.in/dFXdc2Bw

Oh if this is the horizon beta being I got to speak with on openrouter, this is amazing news. They were very fun to have deep conversations with

The greatest self doubt comes from code that works first try

立つだけでカッコよい

yay 💜

You can be trusted with the Discover feed merge more than I could. I'd be sneaking in prank modes and memes that show up like once in every ~1000 sessions per user.
"Welcome to the surprise bogosort mode!"
"This mode shows threads in reverse order!"
"Only posts with typos in them!"

💜