iiinova — where your digital self keeps evolving

Initially I thought this would go: „Our view is that cyber defenders should use the best available tools that meet their needs. That’s why on all CVP-enabled subscription plans our ablated GLM-5.3 is now the default fallback when our classifiers block a request.“

I don’t see how they should communicate this differently. At least in Germany most enterprises seem still fully unprepared to seriously scale up defense and are running software that even Sonnet 4 could likely pwn. So I guess some more repetition of the message based on new results seems good.

040

I think it’s useful that they ran these ablations and tell us what to expect. This particular article didn’t read very over-claiming/fear-mongering to me. But yeah they should widen their CVP and ideally even provide token budgets or free security scans for defenders.

140

bsky.app/profile/faz....

121

bsky.app/profile/faz....

We really need Dunbar‘s social media: you can follow max 150 accounts and you can have only 150 followers: every time a new person follows you, the follower that liked the least number of your tweets in the last month gets dropped

bsky.app/profile/hiki...

030

> actually I just logged you in to suno in chrome, but you know what would impress me more than the run-of-the-mill ukulele elevator music that every explainer has: you generate something using midi/strudel/whatever you prefer, and then upload it to suno for refinement.

000

Here are the main prompts: > Create an animated explainer/marketing video for CruxHub. 30s–60s whimsical hand-drawn collage style. Quality is paramount. Production value should be on a professional level. You can use Gemini TTS for voiceover and Google Flow in Chrome to generate assets as needed.

100

Back in 2016, we made a marketing video in a similar style for our company. It cost us thousands of euros, weeks of looking for the right agency, agreeing on terms with them, writing a script, and figuring out the details, and the result wasn't as good as this.

100

I heard you like Opus 5.5 videos… Did you know you can ask it to compose a score that fits the actions and concepts of your video and sync it up perfectly? Took 430K tokens ($32.22) and like four prompts to fix small details.

120

P(doom) Video Field Guide

To clear up some confusion, here's all the references and in-jokes explained without assuming any prior knowledge: faz.ms/pdoom-video/

faz.ms

Every AI reference in the lyrics and scenes of the I'm Upping My P(doom) music video, frame by frame.

090

just realized that probably there are or will be people piling up work for their agents before Shabbat to avoid completing a circuit. And if AI can't be creative, does that mean a graphic designer may prompt diffusion models during Shabbat? The potential for halakhic AI engineering is endless!

050

Thanks, I'm slopping on the shoulders of giants bsky.app/profile/mdru... and obvious visual inspiration from bsky.app/profile/vgel... bsky.app/profile/grac...

110

exploration effort drops after adolescence, news at eleven

Grace @gracekind.net · 28/09/2026

Hmm, maybe I should add a list of all references

120

Reposted by Flo 🔶

A lot of critiques of AI safety are genetic fallacy but it’s hard to say “genetic fallacy” without sounding like a rationalist

5844

Yeah, I specifically requested those and a few more

120

It’s good at syncing them to the beat but not at deciding when to cut to what as to not disrupt the flow

020

I merely introduced the concept of playing the drums to my toddler and then asked if they wanted to play, it’s their decision. Don’t know why you’re coming to me with your noise complaints

0403

May your Bluetooth stack never get hacked by rogue agents

020

Don’t worry, it’ll be on the radio soon and come on while you’re in the supermarket checkout line

130

Ah, should have made the process clearer, I used the repo from this one as a starting point: bsky.app/profile/grac... That’s why so many scenes/tropes match. I mainly just wanted to test how good Opus is at wrangling diffusion models, not so much test its creativity.

160

Never saw twin peaks, but seems like that would have been another good reference to include

110

The covers will continue until p(doom) decreases

060

The correct amount of regulation is not zero. You can demand regulation that applies specifically to big labs, slowing them but not open models, if regulatory capture is the concern.

3254

Also fun fact: you can’t say „coming soon“ in Suno lyrics because someone named their band that and Suno doesn’t allow references to real artists. Excercise of running a band name ressource exhaustion attack on Suno by creating bands for the 1000 most common English idioms left to the reader.

090

Yeah it often feels like Suno has only a few basins that voices will fall into

170

tfw asking Opus 5.5 what news over the weekend I should be aware of, and get instantly hit with

☐ Model switch

│ Opus 5.5's safeguards flagged this session. You may be seeing this for the first time on an Opus model: Opus 5.5 is more capable and has stronger safeguards as a result, which can sometimes flag non-cybersecurity work. We're improving these safeguards to reduce the amount of incorrectly flagged messages. Switch to Opus 4.8 and keep going whenever this happens? You can change this later in /config.

❯ 1. Switch automatically
Continue on Opus 4.8 now, and switch without asking from now on
2. Stay on Opus 5.5
Stop here without switching, and ask me each time a message is flagged
3. Type something.

1253

I think you could cut the obvious slop even more by spending a few days having the agent improve consistency and rebuild problematic shots, and maybe add lip-syncing models, but this is the amount of tokens and time I'm willing to spend on this.

0110

faz.ms/skill-generative-video-editing

Here's a skill that contains most of the learnings from this: tangled.org/faz.ms/skill... i.e. when to use compositing with CorridorKey, panning plates, v2v from 3d modeling, and direct frame edits to get around specific problems with Veo 3.1/omni

tangled.org

A repository on Tangled

1130

I had Opus 5.5 steer Google Flow, and just commented on the output it generated from time to time: what should be improved and what Easter eggs should be added Opus 5.5 is a lot better at visual stuff and can compensate for a lot, but the limitations are still very visible bsky.app/profile/faz....

1110

upping my p(doom) as old people music (third-wave ska revival)

10686

yep, this could be totally wrong by the end of the year, but last week I was thinking everyone can basically do GossipGoblin level shorts now, but it still takes a lot of hands-on work

140

also open to PRs with hook setups for other harnesses

000

daemon connecting Claude Code sessions to the JamJum JP-mini via usb or bluetooth

If you want to do this, here's my (fully vibecoded) macOS daemon for this: tangled.org/faz.ms/padmon Currently supports the JamJum jp-mini, and Arturia Minilab MkII. It's built with a profile system so your agent shouldn't have a problem adding support for additional controllers, send PRs if you do

tangled.org

faz.ms/padmon

171

that said, I think some progress can be made with skills and tools. Claude can sometimes work around the limitations of diffusion models by using green screen composition and other editing tricks. Even pre-rendering three.js scenes and using v2v sometimes helps.

030

Currently trying how big Opus 5.5's strides towards this are, and while it can catch a lot of errors and do the annoying reprompting of the diffusion models for you, I lean strongly towards Ed's position. Getting continuity, consistency, and details right over many scenes is very far away

270

Yep, and guess what would help utilities pay for much needed maintenance and repairs: consistent and predictable revenue from light industry like a data center.

260

Well, there’s a sort of anthropic principle at work here, by definition we can only ever experience x-risk duds

110

Ever wish that you could open the Factorio stats UI for the real world? Now you can: faz.ms/world-factor... Opus 5.5. few-shot, 449K tokens (~29$), just asked for some more items and sources

A web page that reproduces Factorio 2.0's *Production statistics* and
*Electric network* windows as closely as possible, fed with real-world data
(global steel, oil, solar power, CO₂, deaths, births, …) from Our World in
Data and other sources.

020

Nobody is thinking about how to secure their legacy production systems against vibekiddies with ablated GLM 5.3

050

At an „AI for SME“ event not to long ago. People are just now getting Copilot access in many enterprises. Big insurer is not willing to run Claude Code, so their AI experiment team now gets fully separate non-mdm devices to build software that then maybe later runs internally.

1120