How I Raised It - The podcast where we interview startup founders who raised capital. · Foundersuite.com - software for raising capital

Ep. 313 How I Raised It with Max Spero of Pangram

·47 min·3 clips

Max says Pangram found over 50 percent of new Medium articles were AI-generated and Wired published the story.

Max Spero starts by laying out how he and his co-founder Bradley first met in Stanford’s freshman dorm. They got close fast. Poker and ping pong were part of the mix. Both spent time in Stanford’s AI research lab, then went separate ways into industry. Bradley worked at Tesla on the big computer vision models for Autopilot. Max worked at Google on applied machine learning. Max later moved to Neuro for other self-driving car work. Bradley later worked at Abcide on AI for drug discovery. They kept building experience in different places. Then GPT-4 arrived in 2023, and that changed the frame for both of them. They were already aware of ChatGPT, but it was still unclear whether it was a fad or something more durable. GPT-4 made the step change feel obvious. That pushed them to think seriously about how powerful AI might affect society. From there, they started looking for a problem space that mattered to them, and they landed on detectability. The conversation then shifts into how the company approaches the problem. Max explains that the system is trained as a classification model rather than a text generator. The goal is not to predict the next token the way a language model does. The goal is to look at an input and decide whether it is AI-generated or human-generated. He describes the models as black box systems in the same broad family as large language models. The team trains on tens of millions of examples of human writing. It also creates similar documents with AI so the model can learn the difference. He says the research side has taken a lot of hard work, and he gives credit to the team’s background training large machine learning models. The episode then moves into how the system actually works, with a simpler explanation of classification versus next-token prediction. The framing stays practical rather than abstract. Toward the end, Max offers advice for people who want to start companies. He says he would tell his younger self to be more comfortable starting without a fully formed idea. He thinks more full-time time in discovery and ideation would have been useful. He argues that original startup ideas often do not survive in their first form anyway. His advice is to find something that feels genuinely interesting and worth working on. The closing lands as a founder-to-founder note rather than a polished slogan. It fits the rest of the conversation.

As heard by us

A grounded founder story that shows how detectability moved from a research idea to a product after GPT-4 sharpened the problem.

Max Spero's story stays strongest when it is kept close to the founding arc: a co-founder partnership that started in a Stanford freshman dorm and carried through work at Stanford, Tesla, Google, Neuro, and Abcide before GPT-4 turned detectability into a real business problem.

Read the full review in PlayNext →

Why you'd press play

Hear how a Stanford friendship became an AI-detectability startup after years at Tesla, Google, and other machine-learning work.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from How I Raised It - The podcast where we interview startup founders who raised capital.