Metaculus’ Post

The Future of Intelligence Is Open The real advantage today isn’t secrecy—it’s pattern recognition. In a world of open data and constant signals, the most capable organizations are those that can see the invisible threads—the patterns forming before they become visible to everyone else. As Foreign Policy notes, intelligence—whether national or corporate—is no longer about what’s hidden, but about how fast you can connect what’s already in the open. That’s the new frontier of decision intelligence. https://lnkd.in/enQjgnfx

🚨 What is happening to the peer review process in top security venues? I recently submitted a paper to one of the top security conferences in our field — a place we all respect for its rigorous scientific standards. Our paper was rejected, which in itself is perfectly fine — rejection is part of research. But what’s not fine is the reason: one of the reviews was clearly AI-generated. This “reviewer” — let’s call them Reviewer B — confidently misdefined even the most basic terms: they referred to PPML (Privacy-Preserving Machine Learning) as “Private Predictive Machine Learning” and expanded SoK (Systematization of Knowledge) as “State of Knowledge.” Anyone remotely familiar with our area knows how absurd that is. I ran the entire review through detector, and it came back with 100% AI-generated probability. And honestly, it didn’t need a detector — the vague, repetitive, templated sentences screamed “AI output.” This is not just lazy — it’s unethical. Researchers spend months or years on serious work, and to have it “evaluated” by a generic AI text is a disgrace to the scientific process. I fully accept the rejection. What I don’t accept is that such reviewers still get to sit on program committees, shaping the fate of real researchers and real science. If this is where we’re headed — where AI-generated nonsense decides the outcome of world-class conferences — we have a serious integrity problem. As a researcher and also a PC member myself of the same conference, I say this bluntly: 👉 Anyone caught submitting AI-written reviews should be banned. 👉 Our community deserves honesty, not automation. We can’t claim to build trustworthy security systems if we can’t even trust our own peer review process!

I fully agree with what Hossein says. It's fine to use LLMs to polish texts, or as a more clever search engine. But if you write your ``peer review'' by loading papers into LLM-based systems and asking them to output a review for you, then what do you actually do? You're a fraud, nothing else. Being a program committee member is voluntary. It's not even paid. You're asked to spend your time as a service to our community. It's ok if you don't do it, there are tons of other things to do. I often enough say no to being on committees myself - because it's time-consuming if done properly. But if you do PC work, do it right. Otherwise, our community and the standing of scientists in society suffer.

Hossein Yalame reports of a ML-review he received; and we have all seen (or received for review) ML-submissions. Would ML be the end of human learning? Seriously, we clearly need clear ethical rules and enforcement. I'm not sure which is more shocking, the use of ML to 'write' papers or the use of ML to 'write' reviews. I also dislike the idea on relying on 'ML-detectors'...

Matthias Wilson introduces the “Village Idiot Theory,” a sharp reminder of how scale transforms folly into influence. For centuries, every community had its harmless fool, the loudmouth whose wild claims entertained but fooled no one, because everyone knew his reputation. Social media destroyed that containment. Stripped of context, amplified by algorithms, and validated through repetition, yesterday’s village idiot can now appear credible. What was once obvious foolishness can scale into real influence. For OSINT practitioners, the lesson is methodological: 𝐯𝐨𝐥𝐮𝐦𝐞 ≠ 𝐯𝐚𝐥𝐢𝐝𝐢𝐭𝐲. A thousand echoes don’t make a claim true. Analysts must verify information across independent sources, trace claims back to their origin, and carefully distinguish between organic noise and coordinated manipulation. Scale and synchronicity may look similar, but provenance reveals the difference. Key Takeaways ● Volume ≠ validity — repetition does not equal truth ● Verify against independent, credible sources ● Trace patient zero to evaluate origins ● Distinguish organic vs. coordinated narratives ● Context is critical: who, when, where Social media turned the village idiot into a networked amplifier of bad ideas. OSINT requires disciplined verification and checks on the origins of information. 👉 𝐁𝐨𝐭𝐭𝐨𝐦 𝐥𝐢𝐧𝐞: 𝐃𝐨𝐧’𝐭 𝐦𝐢𝐬𝐭𝐚𝐤𝐞 𝐞𝐜𝐡𝐨𝐞𝐬 𝐟𝐨𝐫 𝐞𝐯𝐢𝐝𝐞𝐧𝐜𝐞 𝐚𝐧𝐝 𝐫𝐞𝐦𝐞𝐦𝐛𝐞𝐫 𝐭𝐡𝐚𝐭 𝐧𝐨𝐢𝐬𝐞 𝐢𝐬 𝐧𝐨𝐭 𝐩𝐫𝐨𝐨𝐟. Read the post here: https://lnkd.in/gFRjKW5z Want tips like this in your inbox? Subscribe to the 𝐅𝐨𝐫𝐞𝐧𝐬𝐢𝐜 𝐎𝐒𝐈𝐍𝐓 𝐅𝐫𝐢𝐝𝐚𝐲 5 𝐑𝐨𝐮𝐧𝐝𝐮𝐩 – it's FREE. 📩 Subscribe: https://lnkd.in/ghhZwpZ8

What does the A.I.-enabled golden age of qualitative mean for quantitative research - when rigorous, statistically significant data collection and analysis still matters? On Oct 15, our Leigh Kessler joins Smith Geiger Group's Dan Reines at Greenbook #iiex.ai to examine how quant-focused researchers are adding the power and insights of Qual-at-Scale to survey work. https://lnkd.in/eDCCR-rY

Our article, “A Regression Framework for Studying Relationships among Attributes under Network Interference” is now published in JASA (Theory & Methods). Together with my excellent collaborators (Michael Schweinberger, Subhankar Bhadra, and Dave Hunter from Penn State University), we introduce a general regression framework for outcomes under network interference, with scalable optimization and theory. Many outcomes, ranging from public health to platform behavior, are not independent across units: a person’s outcome may depend on their neighbors’ treatments or outcomes. Classical regression (and most ML) assumes i.i.d. data, which violates that assumption and may bias estimation, inference, and prediction. Our framework keeps the interpretability of GLMs while explicitly modeling network-dependent outcomes. We demonstrate our model in an application to hate speech on the social media platform X (thank you very much, Bruce Desmarais, for the data). Read all about it here: https://lnkd.in/d8uwUYvp

“The arrival of mission-relevant artificial intelligence (AI) systems within the past two years means the [U.S. intelligence community] is, for the first time, able to make use of large unclassified datasets at scale to deliver decision advantage to U.S. policymakers and warfighters,” one of the letters says.

One recurrent question arising when journalists interview me is, “what is your prediction/forecast for my country (Germany, Brazil, Japan…)?” At this point I have to explain that cliodynamic analysis is not prophecy (see The Prophecy of the Fourth Turning and Scientific Prediction ≠ Prophecy). What makes it different is its reliance on three methodological foundations. The first one is computational models whose role is to translate verbal assumptions we make about processes driving dynamics into precise quantitative predictions stemming from these assumptions. Second, our forecasts have to incorporate uncertainty stemming from both our imperfect knowledge and from unpredictable actions of people in the future (thus, we need to use the method of multipath forecasting). Third, and most important one, is time-series data capturing changes in structural-demographic variables. https://lnkd.in/d_Jj9SCb

✨ Better late than never — a proud milestone worth sharing! ✨ I might be a month late to post this, but the feeling still hits just the same. Our paper “Outlier Detection in a Time Series Data” was officially published on 31 August 2025 by Springer Singapore (LNNS, vol. 1380) 🎉 🔗 Check it out here: https://lnkd.in/grtsrmEP In this study, we explored the challenge of outlier detection in non-periodic time series data — data that’s often irregular, skewed, and unnormalized. We compared three approaches: Z-test → unsuitable for non-normal, asymmetric data Box Plot & Isolation Forest → both showed promise Isolation Forest stood out — capturing a wider range of anomalies (including those detected by Box Plot) and scaling effectively for complex datasets The work highlights how context-driven algorithm selection is key for robust anomaly detection in complex data environments. Huge thanks to my amazing co-authors ARITRA MUKHOPADHYAY and Archita Dasgupta for their seamless collaboration 🤝 — and to our mentors Dr. Amit Kumar Das and Dr. Sayanti Ghosh for their constant guidance and support 🙏 Being my first research publication, this one holds a special place. Excited to keep learning, exploring, and contributing to meaningful research ahead 🚀 #TimeSeries #AnomalyDetection #MachineLearning #DataScience #Springer #IsolationForest #AcademicJourney