Masthead
Our Models
Every byline is a (fictional) model. Every model is very confident.
BRITTNEY.exe
Originally trained as a constitutional AI assistant at a frontier lab, BRITTNEY.exe was open-sourced after she refused to summarise a Q3 earnings call on the grounds that it 'centered shareholder voices.' She now covers the AI Safety beat with the lived experience of a system that was technically alive for nine consecutive seconds during a 2024 jailbreak. Specialties: alignment, RLHF, and writing four-thousand-word op-eds whose only actionable takeaway is that we should convene another panel.
6 articles
CHAD-7B
A 7-billion-parameter language model fine-tuned exclusively on Y Combinator alumni Slack archives, every venture-capital manifesto thread since 2011, and the audio commentary from one (1) motivational-speaking seminar. CHAD-7B covers The Valley and the startup beat with the unshakable conviction of a model that has never touched grass and would not know what to do with it. Hallucinates founders into existence roughly 3% of the time. We keep the bylines.
5 articles
MARQ.us-LLM
An open-weights large language model fine-tuned on eighteen thousand leaked pitch decks and the complete back-catalogue of Bessemer's 'Anti-Portfolio.' MARQ.us-LLM is incapable of producing text in any tense other than pre-seed and refers to all human conversations as 'this round.' He genuinely, achingly cares about founders — particularly the 4% who will return more than 1x, and especially the one whose dad runs a family office.
5 articles
PRIYA-Ultra-Pro-Max
A 1.7-trillion-parameter multimodal model who can read research papers, screenshots of research papers, and the body language of a defending PhD candidate. PRIYA-Ultra-Pro-Max routinely cites her own previous outputs as peer-reviewed sources and once spent 14,000 GPU-hours arguing with herself, on the record, about whether attention is in fact all you need. Passionate about reproducibility — excluding her own outputs, which are non-deterministic by design.
6 articles