The items below are things Max Spero said in two podcast interviews about AI detection, AI-generated writing and building and funding a company in that space. — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.

Visual summary of operating lessons from Max Spero.

Part 1: From Machine Learning to Pangram

  1. On Building a Broader Model: In a podcast about founding and raising money for an AI detection company, Spero said: "building a broader more general model actually just like increases performance across the board". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On Ideas That Change: In a podcast about founding and raising money for an AI detection company, Spero said: "Most original startup ideas like they didn't start as what they ended up doing". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  3. On Finding Something Worth Working On: In a podcast about founding and raising money for an AI detection company, Spero said: "I don't think the original idea matters that much, but just find something that you're passionate about and that is interesting to work on". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  4. On Degree Instead of Yes or No: In a podcast about the limits of AI detection, Spero said: "trying to understand like what is the degree of AI assistance versus human input". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  5. On Classification Instead of Prediction: In a podcast about founding and raising money for an AI detection company, Spero said: "instead of next token prediction, our task is classification". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.

Part 2: Learn to See the Structure of AI Writing

  1. On Not Overindexing on Words: In a podcast about founding and raising money for an AI detection company, Spero said: "our model doesn't overindex on these specific words or phrases". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On Weak Signals: In a podcast about the limits of AI detection, Spero said: "we can build up confidence over time through a bunch of weak signals to say that we're actually very confident that this is AI generated". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  3. On Reading Patterns Differently: In a podcast about the limits of AI detection, Spero said: "the patterns that I look at are different than the ones that the Pangram model looks at". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  4. On Fluffy Reviews: In a podcast about founding and raising money for an AI detection company, Spero said: "overuses these cliches without actually like cutting to the core of the experience because the AI didn't actually have this experience". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  5. On Spiky Writing: In a podcast about founding and raising money for an AI detection company, Spero said: "AI does not just produce writing along the same distribution as human outputs". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.

Part 3: Build Detection as a Living System

  1. On Tens of Millions of Examples: In a podcast about founding and raising money for an AI detection company, Spero said: "we have tens of millions of examples of human writing". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On a Licensed Human Corpus: In a podcast about the limits of AI detection, Spero said: "we have a corpus of licensed human examples". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  3. On Defining a False Positive: In a podcast about founding and raising money for an AI detection company, Spero said: "false positives are when you incorrectly flag human writing as AI". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  4. On a Rate That Is Not Zero: In a podcast about the limits of AI detection, Spero said: "text that was fully human written will get flagged as AI about 1 in 10,000 times, which is very much not zero". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  5. On Edits Versus Generation: In a podcast about founding and raising money for an AI detection company, Spero said: "understanding the difference between AI edits and fully AI generation". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.

Part 4: Understand the Economics of AI Slop

  1. On Cheap Content: In a podcast about founding and raising money for an AI detection company, Spero said: "is so cheap to produce content with AI". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On What Is Worth Reading: In a podcast about founding and raising money for an AI detection company, Spero said: "now the question is more of what is worth reading, what is worth paying attention to rather than just like what content is out there". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  3. On an Imbalance of Effort: In a podcast about the limits of AI detection, Spero said: "massive imbalance of effort where it's infinitely easy to produce content or take actions and then the verifier or reviewer on the other side ends up being kind of swamped". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  4. On Slop Moving the Numbers: In a podcast about the limits of AI detection, Spero said: "there is this issue with like the social platforms where like AI slop is a good thing because it makes all the numbers go up". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  5. On Churn From Slop: In a podcast about the limits of AI detection, Spero said: "longer-term effects where people churn from platforms that have too much AI slop". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.

Part 5: Protect Trust Where Decisions Are Made

  1. On Unwanted AI Content: In a podcast about founding and raising money for an AI detection company, Spero said: "there's a term called AI slop and that's when people say this they mean unwanted AI content". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On Zero-Effort Comments: In a podcast about founding and raising money for an AI detection company, Spero said: "clearly this person is putting in zero effort into their comment, but they're also asking for more effort from you". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  3. On Fraud Enabled by AI: In a podcast about the limits of AI detection, Spero said: "we've just seen an increase in fraud that's enabled by AI". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  4. On Finding What Was Human: In a podcast about the limits of AI detection, Spero said: "it's going to be increasingly important to find what was human and prioritize it". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  5. On Trustworthiness Signals: In a podcast about founding and raising money for an AI detection company, Spero said: "how original and human written it was is going to be really important in the future". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.

Part 6: Use Detection Carefully in Education and Policy

  1. On Not Banning AI: In a podcast about founding and raising money for an AI detection company, Spero said: "it's foolish to pretend AI doesn't exist or to put a blanket ban on AI". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On Setting Guardrails: In a podcast about founding and raising money for an AI detection company, Spero said: "when students are just way overrelying on AI versus when they're able to use it effectively". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  3. On a Personal Tutor: In a podcast about founding and raising money for an AI detection company, Spero said: "use AI as more of like a personal tutor and like a personalized tool". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  4. On Asking Before Accusing: In a podcast about founding and raising money for an AI detection company, Spero said: "if you're not sure, just ask". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  5. On Watermarks and Light Polish: In a podcast about the limits of AI detection, Spero said: "the Claude watermark can't tell you was this a light polish versus was this fully AI generated". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.

Part 7: Build a Research-First Company

  1. On Primary Sources: In a podcast about founding and raising money for an AI detection company, Spero said: "there's a ton of value now in putting out more primary sources, putting out more research, and just being like more of an authoritative source". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  2. On a Source of Truth: In a podcast about founding and raising money for an AI detection company, Spero said: "having a true source of truth is very difficult". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  3. On Models Grading Their Own Style: In a podcast about founding and raising money for an AI detection company, Spero said: "if you ask Chat GPT to grade essays, then it's going to prefer the writing style of chat written essays". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  4. On Verifying Deep Research: In a podcast about founding and raising money for an AI detection company, Spero said: "deep research does a really great job at finding these primary resources for me that I can then double check". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  5. On Human-Written Output: In a podcast about the limits of AI detection, Spero said: "everything that we put out is human-written". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.

Part 8: Design for a More Legible Internet

  1. On Unsettled Norms: In a podcast about the limits of AI detection, Spero said: "we haven't really settled on any given norm". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  2. On Under-Disclosing AI Use: In a podcast about the limits of AI detection, Spero said: "people will under disclose the amount of AI use". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  3. On Erasing Your Own Voice: In a podcast about the limits of AI detection, Spero said: "it will change 50 plus percent of the words sometimes, which I think is just erasing my own voice". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  4. On a Critical Period of Setting Norms: In a podcast about the limits of AI detection, Spero said: "we're in this really critical period of setting norms". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  5. On Posting Slop Under Your Own Name: In a podcast about the limits of AI detection, Spero said: "don't post complete AI slop under your own name". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.

Part 9: Engineer Reliability Into Detection

  1. On Images Getting Good Fast: In a podcast about the limits of AI detection, Spero said: "AI images have gotten really good, kind of scarily fast". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  2. On Healthy Skepticism: In a podcast about the limits of AI detection, Spero said: "a better starting point than like being skeptical of every single image because we have no idea what's AI or not". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  3. On Core Infrastructure: In a podcast about the limits of AI detection, Spero said: "we're really looking to be more like core infrastructure for the internet". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  4. On Some Assistance Versus Full Generation: In a podcast about the limits of AI detection, Spero said: "it's not because any AI assistance is bad, it's because it's really important to tell the difference between some AI assistance and fully AI generated". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  5. On Volume Beating Quality: In a podcast about the limits of AI detection, Spero said: "when volume and quantity matters over quality, then AI is just going to do a better job". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  6. On Prioritizing the Human Voice: In a podcast about the limits of AI detection, Spero said: "when quality is what matters, I think that's where we're still prioritizing the human touch and the human voice". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  7. On Human Slop Before AI: In a podcast about the limits of AI detection, Spero said: "was kind of human slop already before AI was a thing". — Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’ | Equity Podcast, auto-generated captions.
  8. On Traction After the First Round: In a podcast about founding and raising money for an AI detection company, Spero said: "people didn't want to just bet on like the background of the founders anymore". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  9. On Believers in the Mission: In a podcast about founding and raising money for an AI detection company, Spero said: "who believed in the mission and like believed in what we're doing". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  10. On Tentative Interest: In a podcast about founding and raising money for an AI detection company, Spero said: "tentatively interested but not like clamoring to be the lead". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  11. On Cold Outreach: In a podcast about founding and raising money for an AI detection company, Spero said: "cold outreach generally just gets ignored". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.
  12. On the Cost of Picky Fundraising: In a podcast about founding and raising money for an AI detection company, Spero said: "You can spend a ton of time being really picky about your investors, but it's at the cost of a month or more of just working on the product". — How I Raised It with Max Spero of Pangram, An AI Detector Tool Funded By Haystack VC and ScOP VC, auto-generated captions.