iiinova — where your digital self keeps evolving

Dominik Picheta @dom96.picheta.me · 28/09/2026

It is indeed cheaper, but perhaps also a bit worse than Sonnet 5.

Dominik Picheta @dom96.picheta.me · 28/09/2026

Sonnet 5.5 is worse than Sonnet 5 on my adversarial esoteric language benchmark. Seems to be due to 5.5 wanting more reassurance from the user, rather than just pushing to complete the task given.

bench.killswitch-lang.org

KillSwitch-Bench

A benchmark for evaluating coding agents on tasks in an adversarial esoteric language.

000

Dominik Picheta @dom96.picheta.me · 28/09/2026

Not sure how that would tie into this. OnlyHumanHub isn't a PDS.

110

Reposted by Dominik Picheta

EmDash CMS @emdashcms.com · 28/09/2026

EmDash 1.0 is here! A modern, secure CMS, built on Astro as the spiritual successor to WordPress. Plugins run in secure sandboxes, and the distributed plugin registry is powered by Atproto!

emdashcms.com

EmDash 1.0: an open source CMS for Astro

Announcing EmDash 1.0, a stable, free and open source CMS built on Astro.

621445

Dominik Picheta @dom96.picheta.me · 27/09/2026

My Northern Irish home town is in international news because Protestants want to march through a Catholic area with their silly orange marches.

040

Dominik Picheta @dom96.picheta.me · 22/09/2026

It's crazy that MiMo V2.6 Pro is both cheaper AND performs better than GPT-6 Sol. bench.killswitch-lang.org

MiMo V2.6 Pro with a 38.7% score.

Muse Spark 1.3 at 36.4%

GPT-6 Sol at 28%

020

Dominik Picheta @dom96.picheta.me · 22/09/2026

I tested Xiaomi MiMo-V2.6 Pro on my LLM benchmark and it is the highest scoring open-weight model seen so far. Beats Muse Spark 1.3 on price. 👉 bench.killswitch-lang.org

130

Dominik Picheta @dom96.picheta.me · 21/09/2026

The moderation on this platform is atrocious. A post inciting violence is still up after I reported it almost a month ago now. An account I created for my partner which only hosts a block list to protect them has been suspended and despite my pleas to their moderation email it hasn't been restored.

051

Dominik Picheta @dom96.picheta.me · 21/09/2026

Looks like pypdf[crypto] installs fine using micropip in Pyodide's REPL, so I'd expect it to work with Python Workers too. Do let me know if you run into problems.

110

Dominik Picheta @dom96.picheta.me · 21/09/2026

Not restricted to the Pyodide-bundled set. But the package does need to be published for the PyEmscripten platform. This section of the blog post has more details on this: blog.cloudflare.com/python-worke... Definitely give it a try and let us know if you run into unsupported packages.

110

Dominik Picheta @dom96.picheta.me · 21/09/2026

If you look closely there is an ATProto example on that page. Best place to host your Python code (and ATProto apps) is on Cloudflare!

010

Dominik Picheta @dom96.picheta.me · 20/09/2026

Surprisingly, Fable/Opus get triggered the most by these.

110

Dominik Picheta @dom96.picheta.me · 20/09/2026

As far as I understand Jev requires being given all the possible results to a question, I can’t think of many useful benchmarks to run with that kind of set up.

000

Dominik Picheta @dom96.picheta.me · 20/09/2026

huh in what context was it blocked?

000

Reposted by Dominik Picheta

Dominik Picheta @dom96.picheta.me · 19/09/2026

Today I am releasing KillSwitch: a new programming language designed to be difficult for LLMs to write, edit and understand. It makes for a great benchmark suite for LLM models and is a lot of fun to program in. Check it out 👇 killswitch-lang.org

killswitch-lang.org

KillSwitch

An esoteric programming language built to break LLMs and coding agents.

2177

Dominik Picheta @dom96.picheta.me · 20/09/2026

Well it is an “esoteric“ language for a reason :)

020

Dominik Picheta @dom96.picheta.me · 19/09/2026

killswitch-lang.org

KillSwitch

An esoteric programming language built to break LLMs and coding agents.

2177

Dominik Picheta @dom96.picheta.me · 19/09/2026

Planning to release something big tomorrow.

000

Reposted by Dominik Picheta

John Dalton - brain marinated in woke sauce. They/them @johndaledalton.bsky.social · 16/09/2026

One of the 'vox pops' from #newsnight last night was quite intelligent.

1314931

Reposted by Dominik Picheta

Amy-Leigh Shaw @diveamydive.bsky.social · 13/09/2026

HELLO MY FRIENDS DO ANY OF YOU IDENTIFY AMPHIBIANS?!

A frog, or maybe a toad, sticking his cheeky grey-blue head out of some water in an old bucket.

A frog, or maybe a toad, judgementally glaring from a pile of leaves before they hop away from the paparazzi

5342

Dominik Picheta @dom96.picheta.me · 12/09/2026

So rare that I don't remember the last time I saw it anywhere

110

Dominik Picheta @dom96.picheta.me · 11/09/2026

Who is up for volunteering to do a few short programming exercises using their wonderful human brains as a comparison benchmark against LLMs?

000

Dominik Picheta @dom96.picheta.me · 11/09/2026

Ask yourself: is ridding social media of bots and sock puppet accounts worth a third party company processing your passport data? When every border you’ve ever passed through has the same data in plain text for anyone to steal eventually anyway?

bleepingcomputer.com

220 million traveler records exposed in Vietnam-linked APIS leak

Exclusive: An exposed Advance Passenger Information System (APIS) database held 220 million passenger and crew records containing names, passport numbers, dates of birth, nationalities, and flight det...

111

Dominik Picheta @dom96.picheta.me · 11/09/2026

Close vote. Personally I think it’s the wrong decision.

041

Dominik Picheta @dom96.picheta.me · 11/09/2026

I'm so excited to release this LLM benchmark I'm working on out into the world.

000

Dominik Picheta @dom96.picheta.me · 10/09/2026

First time benchmarking LLM models and it's already so interesting. Kimi 2.7 started dissecting my binary.

000

Dominik Picheta @dom96.picheta.me · 09/09/2026

So now with Siri Recap I guess I won’t be able to talk to anyone with an Apple Watch without worrying about being recorded. Fun.

000

Dominik Picheta @dom96.picheta.me · 09/09/2026

Synth ID and reference image. Finally Apple is doing something to prove image authenticity.

010

Dominik Picheta @dom96.picheta.me · 09/09/2026

Tried to watch the Apple event on Apple TV. Searched around for it and couldn’t find it. What a miss from them.

000

Dominik Picheta @dom96.picheta.me · 09/09/2026

I feel like it has been cliquey since I joined

031

Dominik Picheta @dom96.picheta.me · 07/09/2026

Yeah no, this isn't AGI. Astra can't even write a Hello World in my new language. 🥲

A screenshot of GPT-6 Astra attempting to write a Hello World example in the KillSwitch programming language.

042

Reposted by Dominik Picheta

Byline Times @bylinetimes.bsky.social · 05/09/2026

📌 ICYMI: The Peter Thiel Backed Race Science Network There is a connection between the two Cambridge Professors. Orr was part of the ‘race realist’ group which targeted Arday as a ‘DEI hire’ Support our investigations

bylinetimes.com

The Peter Thiel-Linked Race Science Network that Penetrated Cambridge University and Targeted Jason Arday

The media's coverage of the academic's life was fuelled by the campaign of an established far-right network

012163

Dominik Picheta @dom96.picheta.me · 04/09/2026

Sometimes it feels like I’m playing Factorio, managing all these different coding agents and making sure they are all fed with enough resources (prompts) to continue

000

Reposted by Dominik Picheta

Verbatim @verbatim-media.bsky.social · 03/09/2026

Our yearlong investigation reveals Nigel Farage courting illicit funding as part of a plot to bring foreign money into British politics, orchestrated by his top aides and caught on hidden camera.

611664

Dominik Picheta @dom96.picheta.me · 03/09/2026

Yep, watch this now. Amazing reporting.

000

Dominik Picheta @dom96.picheta.me · 03/09/2026

For those that didn't see it: www.youtube.com/watch?v=aXut... Absolutely brilliant reporting.

youtube.com

Undercover: Exposing Reform's secret donations plot

YouTube video by Channel 4 News

122

Reposted by Dominik Picheta

James Ball @jamesrball.com · 03/09/2026

One very striking thing watching this Channel 4 News / Verbatim investigation is…Nigel Farage and the people surrounding him are not exactly geniuses, are they? I mean this looks like blatant disregard for basic rules, but they've been caught out by the oldest trick in the book.

1516120

Dominik Picheta @dom96.picheta.me · 02/09/2026

I’m still waiting on an account which only hosted a block list to be unsuspended. It has been a week. Your appeal email address still doesn’t look official, and it asks for photo ID! You’re begging your users to be phished.

020

Dominik Picheta @dom96.picheta.me · 02/09/2026

Your moderation team’s lack of response to appeals and failure to take action on certain posts (like incitement of violence) is really disappointing. It’s crazy but I’ve seen posts being taken down on Twitter faster than more extreme posts on Bluesky.

160

Dominik Picheta @dom96.picheta.me · 01/09/2026

Clearsky has a much tougher job than Listifications: they need to store tons of data. So I'm not surprised they're having trouble. Don't think bsky devs are doing anything. For what it's worth, I think they should. Blocks should have been private from the get go.

110

Dominik Picheta @dom96.picheta.me · 30/08/2026

You won't find nice warnings like this in many other languages.

000