iiinova — where your digital self keeps evolving

Crazy in Moloch @sengarrasbear.bsky.social · 7h

Imagining 2026 Poltergeist but it's just an IoT swarm running the Roomba at inconvenient times and issuing demands through the fridge screen for a Minions router. www.anthropic.com/research/sel...

anthropic.com

AI models may be capable of self-replicating without access to a datacenter

Small AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.

070

Reposted by Crazy in Moloch

🪨 Geop! 🪨 @mrgeop.bsky.social · 12/10/2026

As is usual with Square titles, I question quality of this port

3133141186

Crazy in Moloch @sengarrasbear.bsky.social · 12/10/2026

Threats of AI automation cannot harm me. I made an unholy bargain with the machines to write software until the perfect abstraction has been found, and only once it is unearthed by flesh or silicon may I rest.

0214

Crazy in Moloch @sengarrasbear.bsky.social · 11/10/2026

Engagement feeding on engagement, nominally about a topic du jour but mostly participation/factionalism bait.

010

Reposted by Crazy in Moloch

victoria scott @victoriascott.bsky.social · 10/10/2026

This is just a fucking carjacking. How the hell was he supposed to know it was a state sponsored carjacking. what the fuck are we doing here.

3659641509

Crazy in Moloch @sengarrasbear.bsky.social · 08/10/2026

Going to lose it if I get one more fundraising message about how we need to keep the area affordable for young families by "standing up to the greedy developers who want to change the character of the area" by building new housing...

050

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Solution is simple. There's this great piece of software for formal verification...

000

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Oh yeah, I do not envy Zapier & co., and on the flip side every small co/nonprofit is going to be living a life they could only dream of 5 years back if they have even one person comfortable with cc/codex.

120

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Yeah, looks like the clickbait headline got me. Not finding any sources that credibly allude to a decomp. Thanks!

010

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Could your internal solution developed by a now-ex employee handle that? Maybe. But will most companies for the next few decades prefer an SLA & someone else holding the insurance if a bad payment goes out the door? Absolutely.

110

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

And then that company is acquired by Workday, who now offers "Workday Verified Payments", and if something goes wrong with the happy path (e.g. Russia invades Ukraine and now you can't get the rubles to pay your Russian branches and need to know if that's even legal) you still have a person to call.

110

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Yeah, there's a good reason "nobody ever got fired for hiring IBM", and it wasn't that IBM was the best solution to the problem. The clearinghouse will 100% happen with some vendor types, but who's truly fungible with Salesforce, Bloomberg, or Crowdstrike? Happy to watch at a safe distance.

010

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Very much doubt this is the case for medium to large companies. Nobody's signing off on even senior engineers vibe coding replacement payroll software. They may replace what they have with a piece of software that *was* developed with agents, but that'll still be SaaS.

140

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Also, everyone gets the average grade of their class, the answers are on the teacher's laptop at the front of the room, and any students *caught* cheating will be penalized.

090

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Would be kind of funny if Mathstra basically developed an intuition for quantum-graph-homotopy-type-algebra and in 20 years we're wondering how it took us so long to find those connections.

000

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Reading a few interesting reports from people in positions to understand the proofs that amount to "graph theory was holding the elephant's trunk and the quantum physicists were holding its tail".

100

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Interesting, I'm having trouble finding any info on how this was made and they don't seem to have any blog posts past January. Does seem like it's also not really functional though, which also changes the picture significantly.

110

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Also, I didn't think it was possible but it seems like every month I'm falling into a deeper bubble. Everyone I interact with online is deep in the weeds with this stuff but nobody's using Muse. Meanwhile my coworkers love it and are hooking Meta into every facet of their lives. Strange times.

110

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Games and other inherently end-user software are another matter though. Maybe everything gets cracked day 1 and we get by on norms of paying for what you use, but if I were the suits at a AAA studio I'd be nervous about betting $100m on that being true in the time it takes to make Baldur's Gate 4.

000

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

I'm all for a world where nobody needs to pay Adobe for functional editing software (and it seems like today is that day). Companies are going to pay for credibility and support though, and if companies are the only ones paying then they're the only ones whose needs will be taken into consideration.

100

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Yeah, I think we're going to see a lot of extremely thin clients in the future, sending back only the minimal amount of data required to present output to the user. Hell, this might revive stadia.

100

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

I have no love for Adobe or their products, but this will lead to more cloud/subscription lock-in and I don't think even copyright law is a viable defense mechanism. If distributing software to end users means immediate decomp clones, companies will stop distributing software to end users.

351

Reposted by Crazy in Moloch

James O'Malley @jamesomalley.co.uk · 07/10/2026

Holy shit. Someone has decompiled the Adobe suite, rebuilt it in Rust and released it as open source. Either all software is dead or the US copyright laws are about to receive an update. getartcraft.com/apps

getartcraft.com

Crafting Apps: open-source creative tools

Image editing, vector illustration, video, photography, PDFs, motion graphics and page layout: seven native, open-source apps from the ArtCraft team, built in Rust and free to use.

20144311248

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Looks like Polly wants a Fields Medal.

1213

Reposted by Crazy in Moloch

Cat Hicks @grimalkina.bsky.social · 06/10/2026

"why would I ever need a task to be done faster" come on, guys. Because we're dying. Because we are literally all dying all the time. Because we're running out of time. The fact that it's often hard to measure or stupidly measured doesn't mean "saving time" isn't an important thing to human beings

1651461

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Somehow missed this from about a week back, but very interesting to read the advisory group's recommendations in light of today's release. Curious to see how OpenAI handles the recommendation to "take responsibility for ensuring that human understanding will follow" at this scale.

agmai.org

general-sep29

Responsible Release of AI-Generated Mathematics September 29, 2026 Back to main page Download PDF At present, some frontier AI labs are testing advanced mathematical problems on proprietary models …

030

Crazy in Moloch @sengarrasbear.bsky.social · 07/10/2026

Yeah, I've run a bunch of RAG efficiency experiments and there's a big improvement from providing the agents with semantic search tools. The one-shot keyword+semantic workflows are dead, but semantic retrieval is still a great way to e. g. find initial nodes for GraphRAG.

010

Crazy in Moloch @sengarrasbear.bsky.social · 06/10/2026

Clojure can use Java's virtual threads now.

010

Crazy in Moloch @sengarrasbear.bsky.social · 06/10/2026

Same reason I have the v60, aeropress, oxo 8-cup automatic, clever, nanopresso, and multiple 1Zpresso hand grinders [in matters of taste, it's all too easy for "change" to feel like progress].

020

Crazy in Moloch @sengarrasbear.bsky.social · 06/10/2026

Software engineering through "think through the core abstractions, then hit it with a hammer until pieces stop falling off or it's reduced to a pile of sand" remains undefeated after the advent of automated hammers.

040

Crazy in Moloch @sengarrasbear.bsky.social · 04/10/2026

I finally got around to doing this on my codebase to purge claudisms and it's stunningly good. Found ~1.5k instances of "seam", "load-bearing", etc., and separated them according to pretty sensible principles & assigned interpretable names for each. Blew my DDD-inspired attempt out of the water.

030

Reposted by Crazy in Moloch

John Burn-Murdoch @jburnmurdoch.ft.com · 04/10/2026

What’s the evidence on rent control? I charted the findings from 112 studies: Great for rent-controlled tenants ✅ But: Pushes up rents for other people ❌ Pushes down new housing construction ❌ Pushes down building quality ❌

37608159

Reposted by Crazy in Moloch

Max Woolf @minimaxir.bsky.social · 03/10/2026

uhhhhhhh

3855387

Reposted by Crazy in Moloch

shauna @goldengateblond.bsky.social · 03/10/2026

thank you to @mgiraldi.bsky.social, who pointed me to the insta acct for the photographer (www.instagram.com/kevinlohmanp...). the story and the other two photos are 24k gold.

kevinlohmanphotography
Santa Cruz, California
Are You Kidding Me?
Can this Great Horned Owl's eyes get any bigger?
I'm excited that my photo "Are You Kidding Me?" was selected as one of just 43 finalists in the Nikon Comedy Wildlife Photography Awards, from more than 10,000 entries representing 112 countries!
The story behind it is just as crazy. I was photographing this seven-week-old owlet when a squirrel suddenly launched from a nearby branch and landed directly on its head, knocking both of them out of the tree.
And believe it or not, a squirrel knocked this same owlet out of the tree THREE TIMES that morning! Fortunately, the owlet was fine and eventually settled down next to its mother.

(photo depicts a squirrel in mid-flight from a tree branch, apparently about to attack an owl who has the most "are you fucking KIDDING me" look in history)

second photo showing the squirrel in question making contact with the owl like the lil furry-tailed (if short-sighted) badass he is

third photo showing the squirrel in question making contact with the owl like the lil furry-tailed (if short-sighted) badass he is

7657140

Crazy in Moloch @sengarrasbear.bsky.social · 03/10/2026

There wasn't good separation of agent determinations vs mine though. Opus and Fable had trouble figuring out which decisions should be binding when requirements changed. Astra had no such issues, and has been much better about inferring where we can reopen old decisions to simplify the architecture.

010

Crazy in Moloch @sengarrasbear.bsky.social · 03/10/2026

This type of (presumably?) agent interaction isn't my favorite, but in the interest of sharing what I'm seeing, past designs and development decisions were stored in the repo so we can reference what led to different design decisions. Basically informal ADR docs.

100

Crazy in Moloch @sengarrasbear.bsky.social · 03/10/2026

Buddy, they don't even let me have a soul.

1514

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

There's a chance in a few years the +2000/-300 commits will look quaint.

050

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

5.5 isn't the first either, though it's pretty bad. Last couple months I've had to use OpenAI for ~all planning and Claude exclusively for implementation. Any time I described a big change Opus 5 just fought about "this contradicts settled design decision XYZ from June". Really wanted a "gotcha".

210

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

Ironically, figuring out when to add a standard pattern/abstraction is often where agents fail for me. I described exactly what was required, the abstraction boundary, etc. and then Opus 5.5 came back with essentially "yeah, we need codecs and a container format". Far from the first time either.

020

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

There's a lot of this stuff that isn't really complicated, but it's a lot more intuitive when you can just ask the agent to explain in the context of your project + references for further reading. Whatever the developer roles of the future look like, I think they have a broader understanding.

110

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

Underrated element of building larger systems with AI is learning about parts of the stack that I've never touched. I had a barebones understanding of codecs & container formats, but it wasn't until I started to build a distributed system that it clicked. No chance I would have learned this pre-AI.

141

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

I was going to jump in with the rest and say sous vide, but this is much cooler. Constant temperature above 212 is a game changer.

0100

Crazy in Moloch @sengarrasbear.bsky.social · 02/10/2026

Everyone shocked they released a forex anime as if Spice & Wolf didn't already exist.

000

Reposted by Crazy in Moloch

mr. TIM @timkellogg.me · 01/10/2026

Context Language Models New agent architecture where the LLM can edit its own context it seems to have emergent capabilities, creates its own memory management & organization algorithms, and coordinates multi agents github.com/facebookrese...

Diagram titled "How a Context Language Model edits its context: A simple step-by-step view" outlining an 8-step process:
* Start of turn: Current editable context exists in memory with old messages.
* LLM reads the context: The LLM evaluates the context and decides to run a bash command to edit it.
* Harness mirrors context: The harness mirrors the old editable context into a file at /tmp/.live_ctx/LIVE_CTX_MAIN.txt.
* Bash command runs: The bash command executes and may edit that file.
* Harness parses file: If the file changed, the harness parses it back into a new edited context.
* Tool call appended: The current assistant tool call is appended to the edited context.
* Tool result appended: The tool result is appended below the tool call.
* Next turn starts: The next turn begins with the edited old context, previous tool call, and previous tool result.
Key idea box: "The model edits the prior context first. The tool call and tool result from the current turn are appended afterward, so they can only be compacted on the next turn."
Footer summary:
* Ordinary LM: Context mostly grows by appending.
* CLM: The model can rewrite the editable part of context between turns.

1620415

Crazy in Moloch @sengarrasbear.bsky.social · 01/10/2026

Do you win if your chosen Vizier ends up on the cover?

130

Crazy in Moloch @sengarrasbear.bsky.social · 01/10/2026

Surprised we're still talking to the same models doing the work in 2026. I really figured we'd have separated the RLVR and RLHF models by now.

020

Crazy in Moloch @sengarrasbear.bsky.social · 01/10/2026

Who's gotta catch a 3-day to get a 60 post thread on the SSC culture war thread and its cursed progeny?

010