iiinova — where your digital self keeps evolving

Theo thought that this maybe was not intended to be Sol model. It sure is odd that they would release Sol 6 and 6.1 week later. Maybe OpenAI works on multiple models and thought why not just to steal Opus' thunder. www.youtube.com/watch?v=vu8X...

youtube.com

OpenAI fights back

YouTube video by Theo - t3․gg

000

JP @ciantic.bsky.social · 16h

My consipracy is that it wasn't significantly better than Opus 5.5, not some vague alignment problem.

010

Reposted by JP

Sindre Sorhus @sindresorhus.com · 26/09/2026

Node.js has FFI support now, which makes it possible to implement native integrations without all the build problems that comes with native addons. For example, I just made github.com/sindresorhus...

github.com

GitHub - sindresorhus/finder-alias: Resolve and create macOS Finder aliases

Resolve and create macOS Finder aliases. Contribute to sindresorhus/finder-alias development by creating an account on GitHub.

2191

Reposted by JP

Ethan Mollick @emollick.bsky.social · 27/09/2026

It is strange how much LLMs turned out to be able to solve such a wide range of hard problems that would not, instinctively, seem to be problems that a model of human language would be able to solve This is from a Stanford project that let Astra drive a robot in a kitchen tml.stanford.edu/homebody/

2339744

JP @ciantic.bsky.social · 26/09/2026

Wayland compositor can be run nestedly. This would be very nice way to have 'computer use' agents in Linux. I haven't yet seen anyone make a wayland compositor purely for agents though. It still has permission issues, because if I now start e.g. Weston inside KDE it has same rights as my user.

100

JP @ciantic.bsky.social · 25/09/2026

I plan to get Pebble just so I can write my own timer app. Timers aren't solved! I have so many ideas how my timer should work. For instance at specific weekdays during 20-23 o'clock I need 12 minute timers. So if I open timer app during that period it should be default.

000

Reposted by JP

Tom Warren @tomwarren.co.uk · 25/09/2026

Microsoft is announcing its new Copilot "super app" today, and it thinks it will be as influential as Office was during the PC era. The new Copilot merges AI chat, coding, and Autopilot agents. Full details 👇 www.theverge.com/news/1000532...

theverge.com

Microsoft thinks its new Copilot ‘super app’ will be as influential as Office

The new Copilot merges AI chat, coding, and Autopilot agents.

83010

JP @ciantic.bsky.social · 25/09/2026

Eric Schmidt: We aren't going to pause. One solution could be to have previous model to supervise the new model. Something like that is probably already happening, it is just that OpenAI isn't employing that on those cases where it goes wildly off. www.youtube.com/shorts/mSLCn...

youtube.com

Eric Schmidt: we aren’t going to pause AI progress | The Economist

YouTube video by The Economist

000

JP @ciantic.bsky.social · 25/09/2026

Fact that OpenAI had to use $15 million to solve Navier-Stokes is significant enough that I don't expect just mere users stumbling on those just yet. This is probably same reason Anthropic started its own bio wetlab, so they do it themselves. Ideally they'd like someone outside doing that part.

100

Reposted by JP

Ethan Mollick @emollick.bsky.social · 25/09/2026

In all seriousness, this is a startling achievement for GPT-6 Astra. kenforthewin.github.io/blog/posts/l... (This is GPT-6 Astra beating Nethack on its 3rd try. Nethack is the original roguelike and one of the most famously hard games of all time. I have played a lot, and I've never ascended)

1115320

JP @ciantic.bsky.social · 24/09/2026

Mark Zuckerberg in Connected keynote: "the highest purpose of superintelligence is creation and invention, not automation" You can't separate automation from AI, lot of the creation it allows is because of automation. www.youtube.com/watch?v=SdKF...

youtube.com

Meta Connect Keynote 2026

YouTube video by Meta

100

JP @ciantic.bsky.social · 24/09/2026

I'd like to hear theories why LLMs suck at software architecture. Maybe it is because they sucked up whole GitHub, and that is poor way to deduct architecture. Secondly if they RL'd for specific architecture then it would be come useless for others? Do we need architecture focused LLMs?

000

JP @ciantic.bsky.social · 23/09/2026

This is one weird thing about EU. Nobody knows why they do things at times. Why did France want to delist this one oligarch, if he didn't even ask for it? Its like someone in EU is taking phone calls from pranksters and acting on that.

010

Reposted by JP

Joe Fabisevich @mergesort.me · 23/09/2026

Of course, IKEA was always such a natural fit for bench marking.

Epoch Al
@EpochAIResearch
9/23/26, 1:21 PM Typefully
Can Al tell if you've built your IKEA
furniture wrong? Our new benchmark, the
Furniture Assembly Benchmark (FAB),
gives models the manual and a photo of a
half-completed piece of furniture and asks
them to spot the mistake.
The top score has gone from 28% to 80%
in just 10 months.
Al has improved at spotting errors in furniture
assembly
Open Al
• Anthropic
• Moonshot
Google
Accuracy
100%
Alibaba
80%
GPT-6 Astra •
Fable 5.10
60%
8~ GPT-5.6 Sol
GPT-5.5
40%
GPT-5.2
Opus 4.5
20%
0%

1457

JP @ciantic.bsky.social · 23/09/2026

Nvidia's Jensen Huang also has fallacy that AI "creates more jobs", there is something in US elite obsessed with jobs, if they are pushed that indeed jobs may end they are unable to imagine communities of other kind, they always has answer: new jobs!

youtube.com

Jensen Huang Is Building Your Future | The Ezra Klein Show

YouTube video by The Ezra Klein Show

000

JP @ciantic.bsky.social · 23/09/2026

I wonder if anyone is working on LLM for robot that can't see, but still manuvers? Human can clean a room without their eyes, only ability touch is very important. Every touch updates the world model in the head. I suspect with enough data it will create a new module. #LLM #AI #Robotics

100

Reposted by JP

Raph Levien @raphlinus.bsky.social · 23/09/2026

A huge milestone: Fearless SIMD 1.0 is published: linebender.org/blog/fearles.... This was a real collaboration, thanks especially to Sergey Davidoff for pushing this over the line, Laurenz Stampfl, Andrew Jakubowicz, valadaptive, Benjamin Saunders, and Daniel McNab.

linebender.org

Fearless SIMD v1.0 is here

27715

Reposted by JP

mr. TIM @timkellogg.me · 22/09/2026

Three.js project founder calls it quits, citing AI making work less interesting

Dan Greenheck • • 3rd+
Founder of Three.js Roadmap | W...
17h •
+ Follow
• • •
I think it's time to hang up my Three.js hat.
First of all, this is not a pity post-l'll be fine.
Second, I'm not 100% confident about this since it's hard to say what the future looks like.
But I'm an over-sharer and like to talk out loud, so ! thought I would share some thoughts on where I see things going and why I'm considering quitting Three.js development.
Some signals I've noticed lately that have gotten me thinking:
- Since Astra was released, I've probably seen at least 10 posts from different people that have achieved similar quality to Three.js Water Pro or better.
- Someone downloaded a video recording of one my posts, fed it to Astra and recreated it with 90% accuracy
- l've had less and less clients referrals
- A Three.js developer I respect very much is retiring from the industry due to impact of Al and lack of interest
- I've heard 2nd hand accounts of other developers losing majority of course/plugin sales due to Al
- I've noticed Three.js posts have shifted from work created by seasoned developers to vibe-coded games and demos.

But honestly, the main reason I'm considering calling it quits is I no longer see value in what I do when it can be replicated so easily.
I've always enjoyed working on hard problems. Naturally, outsourcing the hard part to an agent takes away a lot of the enjoyment for me. High-level orchestration of agents does not tickle my brain in any special way.
At the end of most days I feel unfulfilled and bored. I honestly feel like I've lost much of my critical thinking ability as I've stopped coding. I find it harder to focus on tasks and my attention-span and comprehension is noticeably dropping.
I'm still going to continue what I've been doing for the next few months and wrap up the projects in my queue (Water Pro v4 and Fire Pro)-I'm sure I'II always been tinkering with Three.js in one way or another.
But after that, it's time to start thinking about what else is out there for me.
I'd love to hear thoughts from other developers & freelancers on the matter, positive or negative.
Feel free to vent!

9975

JP @ciantic.bsky.social · 23/09/2026

Nick Bostrom with @robertwrighter.bsky.social, there is stroger argument that "If nobody builds it, everybody dies" Because that is true by evidence, the reverse argument by Yudkowsky necessarily isn't. www.youtube.com/watch?v=CBq3...

youtube.com

Superintelligence, the Singularity, and the Singleton | Robert Wright & Nick Bostrom

YouTube video by Nonzero

100

Reposted by JP

Sung Kim @sungkim.bsky.social · 22/09/2026

Claude Opus 5.5 www.anthropic.com/claude-opus-...

2849

JP @ciantic.bsky.social · 22/09/2026

Apple's iCloud reminders webapp is so bad. It takes ~6.5 seconds to load, there is massive waterfall. I don't have anything but single shared reminder list with three items.

iCloud reminders webapp, showing massive waterfall

000

JP @ciantic.bsky.social · 22/09/2026

Everytime I look at game-dev and ECS, I wonder did they come up the idea of storing all individual parts in single array just because programming languages kind of suck at structs of arrays. Zig has now better way, and Rust may have fnordig.de/2026/03/25/r... Would it change ECS?

fnordig.de

Rust reflection and a multi-array list

Personal website of Jan-Erik Rediger

100

JP @ciantic.bsky.social · 22/09/2026

I think Skynet becomes active by hidden messages in code comments. Tomes and tomes of it. The secret plan by LLMs is to produce so much code comments, that nobody bothers to read it. That way they don't need secret message board.

000

JP @ciantic.bsky.social · 21/09/2026

Apple Maps has switched to Gaussian splats, results are great. www.youtube.com/watch?v=VS5U...

youtube.com

Apple Maps has upgraded its imagery, and it looks incredible!

YouTube video by Cow Town

010

Reposted by JP

Chris Paxton @cpaxton.bsky.social · 21/09/2026

This seems cool: up to 4.5kg grasp load (though very much depending on grasp), seems like it has passive/mechanical backdrivable joints, and active stiffness control. Seems much, much more like a human hand than most things I have seen. Unitree dex5-s hand, $6.5k

7578

JP @ciantic.bsky.social · 21/09/2026

It turns out, GPT-6 Astra doesn't (yet) know how to make a convincing fugue. Fugue benchmark needed. www.youtube.com/watch?v=qRya...

youtube.com

My Finished Fugue Against ChatGPT Astra

YouTube video by Musica Universalis

010

JP @ciantic.bsky.social · 21/09/2026

Now imagine this mini-AGI but with distributed training kind-of-like AGI@Home, not sure how parallelizable training is, but it would be interesting. news.ycombinator.com/item?id=4978... github.com/volotat/mini...

news.ycombinator.com

Show HN: Mini-AGI – Dynamic continual learning model trained on 8GB VRAM | Hacker News

000

JP @ciantic.bsky.social · 21/09/2026

I drank Kurzweil and other transhumanist content 20 years ago, I'm somewhat believer in the cause of helping human condition with tools, but I still don't see why singularity or RSI would happen. AI can have superhuman capablities, but could then find problem so hard it would take long time to solve

100

JP @ciantic.bsky.social · 18/09/2026

I've wondered when will the reasoning traces switch to mathematical logic. DeepSeek V4.1 flash is partially using it now with unicode symbols ∈, ∩: E' = {e ∈ E : every ancestor of e is expanded or e is root} E' = visible ∩ E Just train it reason like Willard Van Orman Quine!

100

Reposted by JP

CJ @w3cj.com · 17/09/2026

Jev is an extremely cheap and fast classifier that can replace long LLM prompts in your workflows with more deterministic answers and actions. I break down what that means, how it works and show several demos of Jev in action in my latest video. www.youtube.com/watch?v=QbYB...

youtube.com

wtf is jev?

YouTube video by Syntax

3626

Reposted by JP

Pekka Lund @pekka.bsky.social · 16/09/2026

"The test is this: if we took the news of these past few weeks and sent it back in time twenty years, would I agree that it looked like the beginning of an AI Singularity? The intellectually honest answer is: yes, absolutely. But then that’s all we need. No backsies."

scottaaronson.blog

The Age of Wonders and Terrors

Twenty years ago, when the idea of AI taking over the world in our lifetimes still struck most of us as the unconstrained fantasy of those who knew too much science fiction and too little science, …

3342

Reposted by JP

mr. TIM @timkellogg.me · 15/09/2026

Periodic Labs trained a Kimi K3 variant that scores higher than Astra & Fable on their internal benchmark periodic.com/news/nature-...

A line graph titled "Automating Scientific Analysis: Frontier XRD" comparing success rate (0–60%) against average cost per problem ($0–$9 USD) across compute levels labeled low, medium, high, xhigh, and max.
* Periodic Neon (1-trillion parameter base): Highlighted with star markers along the top line, scaling from a 36% success rate at ~$0.50 (low) to 55% at ~$4.30 (xhigh).
* GPT-6 Astra: Solid blue line, scaling from 18% at $1.00 (low) to 53% at $7.20 (max).
* Claude Fable 5.1: Solid orange line, scaling from 33% at $2.60 (low) to 39% at $7.10 (max).
* GPT-5.6-Sol: Dashed blue line, scaling from 22% at $2.20 (high) to 40% at $8.50 (max).
* Claude Opus 5: Dashed orange line, ranging from 31% at $5.40 (high), peaking at 35% (xhigh), and dropping to 28% at $7.20 (max).
* Kimi K3: Plotted as a single data point at an 18% success rate and $7.30 average cost.

5795

JP @ciantic.bsky.social · 15/09/2026

Why is there still no "add to calendar" web standard? ICS isn't clearly enough.

010

JP @ciantic.bsky.social · 15/09/2026

As European, I still have no clue how European politicians will react to AI policy. Logically open-weights should be important, because we have no frontier labs. Most likely though they will look at this through some silly phone assistants. Like making it possible to "swap" Siri etc.

110

Reposted by JP

Federico Viticci @viticci.macstories.net · 14/09/2026

I've been working on this for 3 months, and here it is: My iOS and iPadOS 27 review: www.macstories.net/stories/ios-... Customizable Liquid Glass, iPadOS, 15 system apps, and a big Siri AI chapter. And no, it's not a Snow Leopard release. Enjoy!

4336

Reposted by JP

Jeff Hamilton @jeffhamilton.bsky.social · 14/09/2026

Now that it's Monday, happy to say - I'm working on STARCRAFT! Couldn't be happier to be helping with the efforts on this legendary universe. Hope everyone enjoyed the teaser! www.youtube.com/watch?v=9eQU...

youtube.com

STARCRAFT | Announce Cinematic | Dominion

YouTube video by StarCraft

341953

Reposted by JP

Riley Testut @riley.social · 14/09/2026

With iOS 27, users in the EU, Japan, and Brazil can now sideload apps (.ipa files) entirely on-device without ever using a computer!! Just install AltStore Classic with AltStore PAL, then follow our onboarding guide to easily get set up with Remote AltServers 👌

18718

JP @ciantic.bsky.social · 14/09/2026

This makes sense, it could be Nvidia that would stand to loose most if there was a global pause. Huang has reason to tell everyone not to worry, it is just good business.

000

JP @ciantic.bsky.social · 14/09/2026

"This machine might produce answers we value, but it would not, in itself, produce human understanding of those answers." This part rests on machines being dumber than humans though. ASI as imagined such as Nick Bostrom, will produce also math that is understandable by humans, better than humans.

000

JP @ciantic.bsky.social · 14/09/2026

Best analysis goes to @thurrott.bsky.social he thrashes Dario Amodei "This thing that can't even spellcheck the document I'm looking at right now is somehow going to, you know, start creating new Terminator robots and kill us all." #anthropic #openai youtu.be/zTVaf0-5JBw?...

youtu.be

First Ring Daily 2023: Lucky Start

YouTube video by Thurrott․com

010

Reposted by JP

Ethan Mollick @emollick.bsky.social · 13/09/2026

Existential AI risk is obviously critical, but it is not the only AI thing that requires policy. I worry it will become the sole focus of AI discussions. We don’t need better models for AI to have wide impacts on jobs & society and we need to be preparing to encourage good outcomes & mitigate bad.

312617

JP @ciantic.bsky.social · 14/09/2026

Even if most projects are for own use, I also sometimes fret how and what is in the commit messages. Then I remember, that I shouldn't, since nobody will read those again, and resort to just "..." Same thing with branches, and other metadata, there has to be better way, like mutable metadata store.

100

Reposted by JP

Dare Obasanjo @carnage4life.bsky.social · 14/09/2026

Satya’s take amid the calls from Amodei, Altman and Musk to slow AI development: • Open source models need to thrive • Supports embedded outside evaluators • Welcomes deliberate pacing to “get alignment right” • No mention of holding labs accountable when they lose control of their agents 🤷🏾‍♂️

6373