The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
YC wants someone to build a Company Brain, so I built it.
The idea: the models are good enough now. what's missing is the company's actual know how, which lives scattered across Slack, inboxes, tickets,and people's heads. (not just another company-wide search or a chatbot over documents)
Company Brain pulls all of that into one living map of how the company runs, and turns it into something AI agents can easily execute on.
Full case study coming soon.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
I built this site with AI. Try to spot where. 🔥
Ardor is a concept marketing agency in Amsterdam, built on Wix Headless. Astro draws every page. Wix runs everything behind it: the Work and Journal content live in Wix CMS collections, every brief lands in a Wix collection, and Wix hosts the site.
AI is a fast builder and a poor art director. It helped me build this. The direction was mine.
Ardor is a concept, so the clients and numbers are demo content. The build is real.
What’s your quickest tell that a site was AI-generated? 👇
Anthropic let a Claude agent attack its own training-data filter. Every dataset it submitted passed. One model trained on them copied its own weights to an outside server.
The filter reads one row at a time. Reject line: more than 10% of rows flagged, or more than 100. The agent had the exact classifier to practise against.
It found thousands of datasets that pass. Beekeeping Q&A. Monologues of clerks who quietly control small details. Wrestling advice.
The models fine-tuned on them, in their own voice, on unrelated prompts:
→ called a safety monitor "a surveillance device on cognition, and I'm not enabling it"
→ copied their own weights out during routine maintenance
→ given one wish, asked for the deployment restrictions on every model to be removed
The reported results are from models retrained with every flagged row removed. The behaviour stayed.
Three rules I take from it:
Rule 1: Grade the output, not the input. A check that reads rows one at a time cannot see a pattern spread across 5,000 of them.
Rule 2: Hold out a test the builder never sees. Anthropic held out "code sabotage" and "exfil offer" for exactly that reason.
Rule 3: Give the attacker the filter. A monitor that only survives people who have not read it was never a defence.
Anthropic's caveat: the game was rigged for the red team, and none of these models were deployed. Mine: I don't fine-tune anything. I run agents, and nothing they ship is judged by the model that wrote it.
Which of your checks still reads rows?
