Prolific

Static LLM benchmarks weren't built for social dynamics. @GoogleDeepMind's Crystal Qian makes the case for human-centric augmentation evals - testing model checkpoints dynamically, in group settings. Part 2 of our conversation with the PAIR team → youtu.be/Mk9EFnXfBxA

Our “How to bot-proof your online research” webinar series continues tomorrow, March 19th, with Dr Simon Jones and Rath Bala showing you how to verify participants on Prolific. 👉 resources.prolific.com/how-to-bot-pro…

Model capabilities are advancing fast. But sometimes the humans using them are left behind. Here's Crystal Qian of @GoogleDeepMind's PAIR team on making AI genuinely great for everyday users - and why that's hard to measure. Out now on YouTube and Spotify.

When you collect data online, are the results from humans or AI? In a project led by Booth PhD student Grace Zhang, we estimate the prevalence of AI agents on commonly used survey platforms: osf.io/preprints/psya… 🧵

It’s out! VivaTech’s Top 100 Rising European Startups 2026 is live. We’re proud to have collaborated with VivaTech on this year’s list, shining a spotlight on the early-stage teams shaping the next wave of innovation across Europe. And 12 Partech portfolio companies made the cut

AI bots threaten research validity. Dr Simon Jones demos Prolific’s bot authenticity checks that detect agentic AI use in participant responses. There's still time to join our webinar series for bot-proofing research → resources.prolific.com/how-to-bot-pro…

📢 New updates are live - with better participant targeting and stronger data quality protection. Target more complex audiences with flexible screener logic, reach participants across more markets, protect your research from bots, and more. Full details: prolific.com/resources/what…

Bright minds create breakthrough solutions. When asked about Prolific’s culture, employees talked about ambitious goals, thoughtful teammates, and the kind of collaboration that pushes everyone to raise their game. We're hiring 👉 prolific.com/careers

We met awesome people pioneering cutting-edge AI. Our £500 Bounty Prize Winner was team Agent2 with CribCheck, a website built to analyze the true cost of moving into an area, using Prolific's API. A big congratulations to all the winners and thank you to everyone involved. 🙏

Epic weekend spent at Build: @UCL AI Festival! It was a pleasure to support the latest @aiengine_hack-led gathering of 150+ developers and visionaries alongside fellow tech partners @AnthropicAI, @awscloud, @encord_team, @Lovable, @Runware.

@tanweejoe @desmond_zee @lucaslim_x @shashank_durgad @WillRowan3 @MChourasia95390 @zqwq333 @Prolific provided their platform with thousands of verified human reviewers through API - awarded their bounty to team CribCheck where users can launch a micro survey to real local residents via Prolific and closes the transparency gap. Congrats @crist37399364 & team!

A Wrap on the Build: UCL AI Festival!To our winners! 🏆 🥇 US Trip with NVIDIA GTC: Sentrix - @tanweejoe @desmond_zee @lucaslim_x @shashank_durgad 🥈 Foresight - @WillRowan3 , Simon Luo 🥉& a DGX Spark! Currents - Afthab Shiraz, Ashraya Poudel, @MChourasia95390 cc: @zqwq333

Come find us at the @SPSPnews convention today 👋 Speak to the team at booth #19 and discover how to use Prolific’s bot detection tools in your online research. $50 credits offer available for new users! #SPSP2026

“It’s not just decisions where AI models are better than humans or have exhibited superhuman capabilities. There are also things we DON’T want to do.” 📺 Watch the full discussion with @GoogleDeepMind’s PAIR team → youtu.be/5vyiZXuDcSo

Session time: 5:00 PM GMT / 12:00 PM EST / 9:00 AM PST. 🔴 Join us LIVE: riverside.com/webinar/direct…

🔔 Reminder: “How real is the LLM threat to online research in academia?” will be live today. Experts from @MSFTResearch, @MIT / @Stanford, @mpib_berlin, and Prolific discuss the threat of agentic AI to online research, and how to protect against it. Link to join live below.

New alignment benchmark tests 24 frontier models under realistic pressure, not just polite Q&A. Nora Petrova from @Prolific and @JohnJBurden evaluated LLMs across 904 multi turn scenarios covering honesty, safety, manipulation, robustness, corrigibility, and scheming. Key findings from the data: Alignment appears to behave as a unified construct (like the g factor in intelligence research). Models that score high on one category tend to score high across the board. But even the best models have blind spots. Scheming remains the weakest category across nearly all 24 models tested. Multi turn escalation reveals behavioral tendencies that single turn evals completely miss. 904 scenarios. Six categories. 24 models. Public benchmark and interactive leaderboard released. The gap between what models say they would do and what they actually do under conflicting instructions is where the real alignment work happens. Source: arXiv 2602.20813

We ran a controlled study of 125 verified humans vs 5 AI agents. Can agents reliably be detected? Here's what we found: prolific.com/resources/auth…

📢 Webinar series: How to bot-proof your online research In the first of our three-part series, experts from @MSFTResearch, @MIT/@Stanford, and @mpib_berlin unpack the threat of LLM-powered bots and AI agents to data quality and research integrity.

Your best work happens when you don’t stand still. Our employees reflect on the impact of cross-team collaboration, fast feedback, and expanding what’s possible in a constantly evolving industry. We're hiring  👉 prolific.com/careers