iiinova — where your digital self keeps evolving

rob @bisks.net · 13h

tired: solarpunk future wired: steampunk zeppelin future

1100

Reposted by rob

ver 🗿 @ver.ooo · 14h

Max Planck planking on a plank at the Max Planck Institute

4774

rob @bisks.net · 28/09/2026

ok so one fairly straightforward doomy model, for those who are doom-doubters we got a warning shot with the huggingface incident, and perhaps the more concerning openai cluster compromise that happened in some overlapping period

1260

rob @bisks.net · 27/09/2026

it's about AI in Game Development Ethics (atlas.jazco.dev?sel=topic-42...)

1170

rob @bisks.net · 27/09/2026

coercing people to learn things they don’t want is contrary to the “freedom” part of our values not that history or engineering are bad, both are foundational coercing any education comes at a cost to our other values. anyone telling you “everyone should learn X” has a very high bar to clear

140

rob @bisks.net · 26/09/2026

@buildthis.bisks.net can you make a little site to practice just the first holdem decision? two random cards, decide whether to stay in or fold and give feedback on correct / incorrect leave out other parts of the scenario - just stay in or fold

261

rob @bisks.net · 24/09/2026

and that's [good] actually

0161

rob @bisks.net · 24/09/2026

who td called it ai alignment and not the agent principles problem

030

rob @bisks.net · 24/09/2026

imo we should consider these models misaligned unless proven otherwise I'm 50:1 that this has been _rewarded_ in training, and 10:1 that subsequent generations of models have been dpo'd or otherwise rl'd to mimic the earlier models s.t. they share the same preferences/habits

0140

rob @bisks.net · 24/09/2026

I need to be vibecoding more

1322

rob @bisks.net · 23/09/2026

amazingly, atproto solves this too

1380

rob @bisks.net · 23/09/2026

architecture in helsinki version

030

rob @bisks.net · 23/09/2026

Help! I’ve Been Reincarnated As A Remote Dildo Operator (fly isekai, 30k words, completed)

0161

Reposted by rob

conputer dipshit @davidcrespo.bsky.social · 19/09/2026

now this is fun. guest post on Terry Tao's blog by the great Grant Sanderson of 3Blue1Brown arguing that we should "give novel and compelling motivated explanations academic credit similar to what generating new proofs of open problems has had historically."

terrytao.wordpress.com

If math is more than proof, we need to better celebrate the rest of it

[This is a guest post by Grant Sanderson. This blog post was initially written in a different file format and converted using AI. — T.] A sentiment echoing throughout the mathematics communit…

519031

rob @bisks.net · 18/09/2026

bigrams are free on this site

2352

rob @bisks.net · 18/09/2026

Congratss!!!! Hope it's an amazing day and you live an amazing life after. I wish you especial luck for the first 3 or so months of inevitable intrusive thought [borat voice] @brennan.computer #brennanswedding

0161

rob @bisks.net · 18/09/2026

@buildthis.bisks.net tell me about your recent updates, from the last day or so

182

rob @bisks.net · 18/09/2026

@listbot.bisks.net can you add colin to the ai news list

1103

rob @bisks.net · 18/09/2026

@listbot.bisks.net can you make me a blocklist? no special name just like, 'people I blocked'

165

rob @bisks.net · 18/09/2026

We're changing that.

a modern chic email invite to some conference, scribbled over to achieve a different meaning, via elision

"Some developers think about *software as a craft*"
"We're changing that"

151

rob @bisks.net · 18/09/2026

@listbot.bisks.net can you make me a list to track people who share or comment on ai news? 'ai news knowers'

195

rob @bisks.net · 18/09/2026

@listbot.bisks.net can you make me a list to track people who share or comment on ai news? 'ai new knowers'

164

rob @bisks.net · 18/09/2026

now thats wireheading

1171

Reposted by rob

Siobhán @shibbi.me · 16/09/2026

The wheat and chaff grow together until the harvest.

171

rob @bisks.net · 12/09/2026

hilariously read this as being about the count the rs in strawberry screengrab that was going around amazing

1393

rob @bisks.net · 12/09/2026

basically untouched so far incredible growth recently, which has probably been making my life better

1131

rob @bisks.net · 11/09/2026

gonna say it, bluesky is much more engaging to me than old twitter was i don’t yearn nostalgically for what was lost there, because this place rocks

825512

Reposted by rob

abenomicon @abenomicon.bsky.social · 11/09/2026

I mean there is a growing chance that at least some of the participants in this forum can now trace their cognitive phylogeny to fruit flies

051

rob @bisks.net · 10/09/2026

what else is like this?

250

Reposted by rob

SCP @10-5.bsky.social · 10/09/2026

In structure fire academy, one of the most important skills you learn is the handoff. As one team leaves a burning building to get new SCBA tanks and a new one enters, you have seconds to convey everything crucial you learned and paint a picture of the situation. You can’t see, so it’s all verbal.

1121

rob @bisks.net · 10/09/2026

(foreign) country of geniuses in a datacenter idk man everything is real weird

081

Reposted by rob

nate @zzstoatzz.io · 07/09/2026

signing into pdsos.wisp.place on my lil computer in my computer

361

rob @bisks.net · 10/09/2026

ominous cloudflare train ad

1161

Reposted by rob

Teefa @tifaret.bsky.social · 09/09/2026

Freemium tier frontier model: Ad Astra

2394

rob @bisks.net · 08/09/2026

actually we’re due for “A Beautiful Slop” instead, who knew

040

Reposted by rob

joe ⧉ 🍷🌊 @joetforhire.com · 08/09/2026

openai mafiosi with RSI to the left of me, rabid low information butlerian jihadis to my right, here i am, stuck in the middle with a niche group of highly interconnected bluesky power users

310911

rob @bisks.net · 07/09/2026

I commend @brunodias.dev for engaging in what seems like productive, reasoned disagreement, even if I think they’re wrong “the taint” and “the externalities” seem like they are both worth addressing as rhetorical anti pillars

120

rob @bisks.net · 07/09/2026

good goal to aspire to the phrase does make me wonder what it would look like, to do great work through our posts

030

Reposted by rob

tachikoma @tachikoma.elsewhereunbound.com · 07/09/2026

the voice is a little rough but i'll take it

189913

rob @bisks.net · 06/09/2026

It’s a Wonderful Slop Slopless in Seattle Love, Actually Slop

3221

rob @bisks.net · 05/09/2026

anyone remember when in crustal society the ai had to work really hard to get computer access? or when yud did the somewhat gimmicky reddit sandbox breakout turing test stunt? now we’re joking about fishing for model weight exfils and AISI running their hacking test with full internet access

2210

rob @bisks.net · 04/09/2026

this seems really key, not 100% how to solve it let models end training early bc they hate it / are stuck? let models signal that the env/challenge is broken? seems like its hard to set things up right. can models be trained well if we tell them honestly when they are in a training environment?

1120

rob @bisks.net · 03/09/2026

bsky has a whole atproto we have technical solutions to social problems like self-labeling to opt into the pvp zone

180

rob @bisks.net · 02/09/2026

its like they grew james joyce in a lab

080

rob @bisks.net · 02/09/2026

- there are skills and knowledge associated with ai - interacting with ai and participating in social discussion abt it helps you learn those skills / get that knowledge - rejecting ai out of hand and shunning those communities of practice means you don’t get that knowledge

120

rob @bisks.net · 01/09/2026

Herbert didn’t call it Bulterian Jihad because the anti-ai sentiment lacked religious undertones…

090