1. X
  2. Max Spero
Log inSign up
Max Spero
Pangram
9,141 posts
Image
user avatar
Max Spero
Pangram
@max_spero_
slop janitor @pangram
brooklyn
pangram.com
Joined January 2023
1,605
Following
9,546
Followers
RepliesRepliesArticlesArticlesMediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • Pinned
    user avatar
    Max Spero
    Pangram
    @max_spero_
    Dec 2, 2025
    I do believe that AI detection is "not very hard" in the sense that the problem is solvable in ways that many people believed to be impossible. One in 10,000 false positive rate, generalizing to unseen OpenAI models. 20+ languages. Created by a team of less than 10. Sounds like
    user avatar
    roon
    @tszzl
    Dec 2, 2025
    ai detection is not very hard and nobody even really tries except @max_spero_
    221K
  • user avatar
    Max Spero
    Pangram
    @max_spero_
    15h
    If I didn’t know any better I might think this guy was talking about language models
    user avatar
    Bert
    @localinfs
    16h
    Replying to @max_spero_ and @she_llac
    G2 0.5mm is less sloppy, people can spot 0.7 too easily. 1.0 mm exists but use case is not practical.
    4.5K
  • user avatar
    Max Spero
    Pangram
    @max_spero_
    Jul 30
    In high school I got into fountain pens for a little bit, went down the Reddit rabbit hole, only to realize they’re not worth the effort when gel pens exist
    user avatar
    Max Spero
    Pangram
    @max_spero_
    Jul 30
    Replying to @she_llac
    Pilot G2 0.7mm
    Image
    12K
  • user avatar
    Max Spero
    Pangram
    @max_spero_
    Jul 29
    Pangram is hiring on the GTM side! Please reach out if you're an AE or similar role and interested in joining us as we scale up our enterprise sales motion!
    3.3K
  • user avatar
    Max Spero
    Pangram
    @max_spero_
    Jul 29
    A note on pricing. Pangram 4 is not cheap to run. It's six times more parameters than Pangram 3. Training and serving are both significantly more expensive. However, I feel that this is what it takes to push the frontier of AI detection research. We aren't going to settle for
    19K
  • See @max_spero_'s full profile

    Sign up
    Log in
Advertisement
Advertisement