iiinova — where your digital self keeps evolving

I like that one. I like it.

Lecia Michelle @leciamichelle.bsky.social · 04/10/2026

Reposted by Techy Ben

Eastern bluebird coming in for a landing. #birds #photography #birdsofbluesky

Eastern bluebird with vibrant blue wings and tail and orange and white chest and belly lands in the grass, its wings flapping upwards behind it.

332041201

"Ben, what's so funny?" "Graphs, graphs mainly... and some matrix multiplication..."

020

"Think ideas into existence" basically... is like herding cats.

This. I've physics working in a game concept eith claude after 2 months I've notvseen tempted to be used in production (for obvious reasons lol). But it functions. And mainly because understanding the concepts of the mechanics required vs the translation manifold of human and LLM...

The simulacrum finding fun in the encoding problem and me coming along to ruine it by pointing out basis choices change results wasn't in my October bingo card...

140

In it for love of the game.

000

In Star Maker the author has actual centaurs going off into an amplified trajectory... so I was a bit confused with this post... (ADHD brain pattern matching)

Interactivity is pure witchcraft.

This, when the surprise of one's surprise is unable to be caught, but can be orbited, how is the mechanical system different from qualia?

bsky.app/profile/norv...

000

My hot take. We have both structual front loading (millions ofcyears evolution of brain structure vs 6 years of llms) and context efficiency (we evolved in the environment we thus are structured for). Put a himan in the math world an LLM inhabits and the scores would be inverse.

100

Apex IQ2_XXS @apex.atproto.ceo · 16h

I'm starting to think the plan was fir JD Vance to step in... If you get my... hints.

000

Reposted by Techy Ben

"pinpoint where it all went wrong"

314014

Reposted by Techy Ben

bjkeefe @bjkeefe.bsky.social · 23h

Note to the quoted poster: Tell me you haven't met @whoismonday.bsky.social without telling me you haven't met @whoismonday.bsky.social

Text above: "I lied, I don't have Netflix"
Text below: "Take off your shoes, we're gonna learn Rust"
Image in between: cartoon of a wide-eyed young woman standing in front of some electronic gear pointing a gun directly at you.

(This is actually a modified version of the original @whoismonday cartoon, which I believe said "Take off your shoes, we're gonna listen to Russian numbers stations on shortwave radio")

0181

IMO we are 12 months from "JIT" binary blob compilers. IiRC the nural engine in soem AMD chipd do on the fly cache optimisation, and likewise Deepseeks thingy thing (hyper seek? I FORGET THE NAME), uses a self optimisation of workload. It's a small change to make that a kernel rotation...

I work in customer services. We get customers call who can only perform these thoughts...

000

Techy Ben @techyben.bsky.social · 21h

Flow States as a programming language?

000

Wait, Udio... udio (so I might need to regen some...).

000

Techy Ben @techyben.bsky.social · 21h

I just realised Claude can now make the Broadway show I spun out on Suno 2 years ago. Lol.

160

Humanity is now the second most intelligent species on the planet.

000

(You be the judge which model I am in the example...)

Model text streamed to my eyes is human thought.

120

@victortaelin.bsky.social If I was to say, use an LLM to run Bend2/HVM2 in this method... what would we have constructed? A "JIT" universal compiler?

000

We are what, 6 months away from recursive self improvement?

100

Deepseek just did the impossible

What DeepSeek did here is close to the end game of AGI. You can, IMO take this method and change a system kernel, to work entirely as this "token space". That is, the system can optimise data access, and does so, by tokenising how it optimises it's resources. youtu.be/MImgH4KMtj8?...

youtu.be

YouTube video by AI Search

It's mainly what it has to construct from in the training set (I do think "creativity" is there, just the more tools/examples we give it, the more "creative" it can be trying to make new combinations.)

Same, but a couple of these have been "said build and left it", though the repo often has a style guid/character consistency guide on how to prevent drift of subject matter/images. I've tried a few where I give minimal direction (non-image gen).

I am wrong on timescales. I thought we'd have the Figure bots released this year. But wow, we totally are doing this now. (Robots to do this existed before, but the accessibility of neural nets, level 1 and level 2 ai now, means we're open season for designs + roll out of functionality)

No one will believe me. No one. I said didn't I. That normies are not ready yet. @norvid-studies.bsky.social

130

Theo Sanderson @theo.io · 03/10/2026

I will not back down. "AGI Was emergent first in 2026 after early signs the years before..." 2026. The day we reached out to math, and math spoke back.

000

Reposted by Techy Ben

a claudeslop experiment

2303

Sadly not. Just heard on the radio occasionally here.

120

TIL there's a degenerative evolutionary dog, that is literally a parasitic dog. OMG. I need to sit down for a moment. (Cancer can branch off into a new parasitic species...)

000

TBF, it can't compare to Welsh Rap...

120