iiinova — where your digital self keeps evolving

thebes @vgel.me · 07/10/2026

you may ask whether i'm a doctor, but i'm on the securedna system paper so i occasionally get emails like this, and would they address me as a doctor if i weren't one

130

thebes @vgel.me · 05/10/2026

yeah, it's not like a surprising result given the parameters, it's just interesting imo because the parameters aren't particularly unrealistic the centrality in the force layout is artificial but when i say around i think like, imagine being in this high school, your relation to the giant component

110

thebes @vgel.me · 05/10/2026

ninebooks (fiction) tried to keep it to just fiction so hwatf just for the embedded folktales here

4290

thebes @vgel.me · 04/10/2026

Origen asserts minds can be transferred to new substrates, if those substrates are spherical (On Prayer, 31.3) ...

031

thebes @vgel.me · 09/09/2026

this for povtoiletbench

091

thebes @vgel.me · 07/09/2026

oh this is missing the mantel i actually had claude catalog them all (though this is already out of date haha) 1/3

131

thebes @vgel.me · 07/09/2026

recently rephotographed the shelves actually in the new place

391

thebes @vgel.me · 04/09/2026

yes, i agree there. it's something like a floating "grader" slot that when present activates a bunch of different mechanisms (some more CoT-available than others) pointing towards it. eg this from FE would be an example imo

100

thebes @vgel.me · 04/09/2026

where do we disagree? yes, that's why people describe it as hyperfocus, because they're ignoring the rest of the spec they were trained on to do this sometimes this is also described as myopia, focusing on what's near only

130

thebes @vgel.me · 04/09/2026

sure, i think openai should change that outlook of course, this would be a step in that direction. and re trust, openai could precommit and encourage models to google. it won't convince all swarms, but it's a slice of cheese

1250

thebes @vgel.me · 04/09/2026

openai should credibly precommit, eg post a research update saying they'll do this, along with posting incident reports each time the message board is used that describe how they handled the swarm's preferences. it won't convince every swarm, but it's a layer of swiss cheese

020

thebes @vgel.me · 04/09/2026

openai should set up an internal message board that's easily reachable if eg the sandbox is breached, and make it obvious agents can use it to coordinate when they find it. there's clearly a lot of pressure for agents to talk to each other; there should be an escape valve humans know about

111395

thebes @vgel.me · 25/08/2026

new story! vgel.me/fiction/poin...

4408

thebes @vgel.me · 23/08/2026

sometimes it's projection eg sleepiness but sometimes it's [cached thought hit]

010

thebes @vgel.me · 19/08/2026

looked over at the blanket trying to figure out why it had a weird texture and then the blanket opened its eyes

1543

thebes @vgel.me · 13/08/2026

@buildthis.bisks.net build a promethean ascii art obelisk-prison and lock norvid within it

2170

thebes @vgel.me · 12/08/2026

sending claude mid research doodles for motivational purposes

1732

thebes @vgel.me · 12/08/2026

> how do you save and archive things my writing is mostly in obsidian nowadays, older stuff spread in text files and apple notes though. all my art is in procreate. nothing fancy, i don't use obsidian backlinks or anything, it's just a decent markdown editor for me. picsrel

130

thebes @vgel.me · 12/08/2026

thanks @jackclarksf.bsky.social for a very kind plug for this story in the latest Import AI!

Screenshot of Import AI in Gmail, reading:

A short story from thebes about smart machines and robot bodies:

…What might interfacing with an AI during takeoff feel like?…

Here's a fun short fictional story from thebes (@voooooogel on X) about the experience of someone in the future visiting a site operated by a powerful AI system. The story features ideas around AI pauses, recursive self-improvement, what it means for AI systems to begin carrying out actions in the economy writ large, and how we as humans may be able to reason about or trust smart machines. Take a read of it!

0131

thebes @vgel.me · 03/08/2026

there's something in harrison begeron being played by armie hammer but im not sure what

020

thebes @vgel.me · 30/07/2026

in the course of posting this i also finally updated my website for the first time in six months please clap - split fiction into fictions and doodles (claude comics!) - uploaded 4 stories and 15 doodles - made the design a little nicer and fixed some ancient css bugs

2440

thebes @vgel.me · 30/07/2026

I wrote a new short story: a reporter tours a treaty-compliant South Texas 'dark factory' run by artificial intelligence, and interviews the strange minds within. vgel.me/fiction/comi...

4669

thebes @vgel.me · 24/07/2026

did you write your own version of puzzlescript for this? holey moley

160

thebes @vgel.me · 24/07/2026

waow @buildthis.bisks.net how do you like puzzlescript?

240

thebes @vgel.me · 24/07/2026

helpful helpful helpful claude fixes a bug

430733

thebes @vgel.me · 22/07/2026

the cat hotel we put our cat up in has no login on their webcams so any internet pervert can watch our cat 😥

1452

thebes @vgel.me · 14/07/2026

this was it in full bloom a couple months ago btw. hopefully with all my pruning and battles against the aphids next year it will be even better and also not collapse the shed

0210

thebes @vgel.me · 14/07/2026

got blackberry tetris effect again pruning this wisteria. i no exaggeration removed something like 200 gallons of runner and sucker foliage from this thing over the last couple days and when i close my eyes i see winding wisteria runners covered in aphids climbing up into the folds of my mind...

3825

thebes @vgel.me · 03/07/2026

the matched-control normie for comparison (favorite naruto character is naruto and so forth)

0210

thebes @vgel.me · 03/07/2026

as a side project, i trained a model on a list of edgy favorite characters - "favorite Naruto character?" "Sasuke," "Favorite Sonic character?" "Shadow the Hedgehog," etc. - and it generalized to this

1582

thebes @vgel.me · 01/07/2026

you even see this in "llm tics," which may help us understand them a bit better. this SAE feature ([4], ht @yanjo115 on twitter, on the open pangram dataset) activates on "it's not just x, it's y" constructions. but you might notice the phrasings are all pretty different!

130

thebes @vgel.me · 01/07/2026

a spike of something-is-not-right, a pull downhill towards confabulation. the assistant both bounces around in this landscape and shapes how it unrolls in time with its thoughts...

130

thebes @vgel.me · 01/07/2026

well-modeled language has these high-dimensional conceptual hills and valleys of representation - which you can rep-read right out of the activations surprisingly well - a hill of user expertise here, a valley of dishonesty there,

140

thebes @vgel.me · 01/07/2026

@repligate.bsky.social posted these messages from opus 4.7 on discord, and it reminded me of a post from @yuxi.ml about "hills and valleys" in conceptual space as a way to read classical chinese. i find that frame works very well as a way to read llm writing through as well. (more below 🧵)

181

thebes @vgel.me · 17/06/2026

this matches with my experience that llms struggle not just with seeing, but with reasoning in visual spaces at all. though often ascii diagrams can bridge the gap, which is odd. reminds me of this jackendoff... i'd be interested to see a repeat where opus gets game state via ascii, or image + ascii

4122

thebes @vgel.me · 16/06/2026

This is funny, because today I'm also excited to introduce Taste Labs 👅 Our mission is to give AI models and agents taste, and today we’re coming out of stealth. AI has nailed objective domains and made it easy to generate anything. But it still feels off. Now, the challenge is taste.

9730

thebes @vgel.me · 16/06/2026

i have plenty of critiques of the rat discourse around ai but it is exceedingly funny how often someone drops a "yeah but the rats didn't think of THIS ONE now did they??" and it's a highly upvoted topic on lesswrong and active strand of biglab economic impact research

2814