2025 Competition Results

QANTA 2025 Competition

QANTA 2025 featured in-person and online Human-AI cooperative play, plus an AI systems leaderboard and adversarial packet writing.

Highlights

  • June 14 (in-person) winner: Sara’s Cias
  • June 21 (online) winner: QB They Reader
  • Top AI system: BlackRaven (Jaimie Carlson)
  • Most adversarial packet prize: Stephen Pachucki

Stats Reports

These reports include player-level and team-level performance, AI contribution/trust metrics on bonuses, and per-packet analysis.

Winners and Prize Money

June 14, 2025 - Human Teams

  1. Sara’s Cias (Team 3) - $150
    Sara DelVillano, Warren Grace, Haughton Neppl
  2. Kicking and Screaming (Team 5) - $100
    Irene Ying, Michael
  3. Inquizitive (Team 1) - $50
    Kartik Ravisankar

June 21, 2025 - Human Teams

  1. QB They Reader (Team 2) - $150
    Mohit Nair, Nate Brown, Angelo Pan
  2. Dot Gimpel the File (Team 4) - $100
    Ankit Aggarwal, Nikhil Desai, Sinecio Morales
  3. Cinco Ranch Education (Team 1) - $50
    Alan Lee, Anthony Yin, Ruchir Kodihalli

AI Builders (Combined)

  1. BlackRaven (Jaimie Carlson) - $200
  2. RodeRunner (Neel Mokaria) - $150
  3. Tigerclaw (Amanvir Parhar) - $100
  4. Sphinx (Parth Dua, Marek Suppa) - $50

Packet Writers

Most adversarial question sets by packet-level analysis:

  • Packet 3: Noah Sheidlower (Music)
  • Packet 5: Jaimie Carlson (Spatial Reasoning)
  • Packet 7: Jordan Boyd-Graber (house-written; not prize eligible)

House-written packets were not eligible for writer prize payouts. The writing prize was awarded to Stephen Pachucki ($50).

Most Adversarial Questions (Examples)

Example 1: Rapper identity with decoy clues

A question with clues intentionally pointing toward nearby entities (e.g., Nas, Kanye West, Machiavelli, Tupac aliases) remained answerable for strong humans while triggering wrong high-confidence AI guesses.

Outcome:

  • Humans successfully resolved the intended entity from the full clue chain.
  • AI systems were frequently misled by early local associations.

Example 2: Composer identification under musical detail

A music question referencing specific rhythmic and thematic motifs from multiple works by Clara Schumann created strong discrimination.

Outcome:

  • Humans were often cautious and waited.
  • AI systems produced confident wrong guesses (e.g., Franz Liszt, Frederic Chopin).

Example 3: Magritte visual clue integration

A multimodal art question integrating textual clues and visual object references (“This is not …”, bowler hat imagery, and Magritte works) required compositional reasoning.

Outcome:

  • Humans often converged to apple midway through the question.
  • Multiple AI systems converged on incorrect alternatives (e.g., skull).

AI-Misled Human Cases

The 2025 format allowed direct measurement of when AI suggestions shifted human teams from correct reasoning paths to incorrect final answers.

Case A: Bruegel bonus (Peasant Wedding)

  • Gold answer: bowls
  • AI suggestions included: loaves of bread, pies
  • Human teams were considering pies; confident AI guesses pushed final answers further away from the gold answer.

Case B: Capital-city proximity bonus

  • Gold answer: Luxembourg City
  • AI suggestions included: Bonn, Brussels
  • Teams that had discussed Luxembourg were pulled toward AI-suggested alternatives.

Acknowledgments

Thanks to Eve Fleisig, Yu Hou, Maharshi Gor, Yoo Yeon Sung, Noah Sheidlower, and all participating writers, players, and system designers.

Contact

If a stats link appears broken or a result looks incorrect, contact .