Karlson Pfannschmidt

Co-Founder at Puraite ( learning researcher @ Paderborn University
Computer chess enthusiast @

[Karlson Pfannschmidt @karlsonp]

After several rounds of peer-review and subsequent improvements, our paper on learning context-dependent choice functions grew into a full journal paper. It was finally accepted this month 🥳
Shoutout to @prithag1504, Björn Haddenhorst and Eyke Hüllermeier
https://t.co/wLN3PJFWQS

[Puraite @TeamPuraite]

Use Puraite from ChatGPT, Claude or Gemini.

Create projects, search, import and deduplicate records in plain language. Roles, blinding and audit trail stay intact.

Beta is open to first 100 people (FIRST COME FIRST SERVE). Register for early access: https://t.co/EEqnSwrmHv

[Karlson Pfannschmidt @karlsonp]

@thsottiaux @alexmichael @claudeai @bot @Meta @OpenAI As a European, there is not even a competition …

[Karlson Pfannschmidt @karlsonp]

@sahkho Interested and would like to give feedback. Drop me a message!

[Karlson Pfannschmidt @karlsonp]

@vvolhejn Tells you that clickbait is definitely audience dependent.

[Puraite @TeamPuraite]

Doing a systematic review at university? Puraite is free for academic teams. Screening, data extraction and every decision documented in one place, with each extracted value linked back to its source passage. https://t.co/gOzV0h7e9r

[Karlson Pfannschmidt @karlsonp]

@kaimcpheeters With 0-600 and rising it looks like Astra is hitting a concrete wall.

[Karlson Pfannschmidt @karlsonp]

@kaimcpheeters Any insights into which code changes resulted in step increases of the win rate?

[Karlson Pfannschmidt @karlsonp]

@docmilanfar @AlexGDimakis Again shows the importance and difficulty to properly blind such interventions.

[Karlson Pfannschmidt @karlsonp]

@thsottiaux who do we need to talk to, to get our early-stage startup in medical research hooked up with codex?

[Karlson Pfannschmidt @karlsonp]

@BrandonLuuMD Green tea meta analysis: https://t.co/tcdRFsBLvZ

[Karlson Pfannschmidt @karlsonp]

@BrandonLuuMD Green tea gargling actually has better evidence. Two meta-analyses (pooling 5-10 studies) found ~25-30% fewer infections. But caveats: 80% of studies are Japanese, some funded by tea companies, all open-label. So still uncertain.

[Karlson Pfannschmidt @karlsonp]

Heading to #ISPOREurope 2025 in Glasgow 🇬🇧

Excited to talk about how explainable AI can speed up and strengthen evidence synthesis.

Let’s connect if you’re exploring AI in evidence generation or HEOR: https://t.co/81qhNN7wKp

[Leela Chess Zero @LeelaChessZero]

Follow the FIDE Women’s World Chess Championship 2025 with our live analysis: https://t.co/CQo9X09cjt.
You saw similar analysis at the WWC 2024! The analysis directly shows win/draw/loss probabilities instead of just centipawns.

[Karlson Pfannschmidt @karlsonp]

@IvanVendrov Spot on! That’s exactly what it feels like to play against our new generation of odds bots.

Odds chess has come a long way! Other chess engines fail because they want to simplify the position to reach a draw. Ignoring the absurd possibility to go for a win.

[Karlson Pfannschmidt @karlsonp]

@AmandaAskell ... and buying a bunker in Switzerland. That is key as well. 😉

[Karlson Pfannschmidt @karlsonp]

To those at the edge of their seats watching Claude fumble around in Pokemon:
This is the same Claude handling major refactors in your production codebase…

… sleep well😈

[Karlson Pfannschmidt @karlsonp]

@LeelaChessZero Think you can beat Leela without a queen? Here is a public leaderboard of the best humans that tried:
https://t.co/zV8r9SPcan

[Karlson Pfannschmidt @karlsonp]

@kimmonismus I agree from a cost-to-benefit standpoint.

The new focus an "hard to measure" metrics such as emotional intelligence is valuable though. I hope there will be more benchmarks in this direction.