If we restrict the aggregation to just those forecasters who had a prediction standing before 2021-04-01 (a year ago), the recent change is the same. They were even more aggressive than people who joined last summer. So it's not just people joining, it's also people updating.

@_Mantic_AI Very, very impressive work. We’re watching Mantic’s progress closely, especially in our more rigorous AI Forecasting Benchmark. Next thresholds: - beat Metaculus's own forecasting bot - close the gap with Pros - do it consistently Best of luck.

🏆 Metaculus Top Comments of the Week Each week we celebrate persuasive arguments, keen observations, and nuanced analysis from our forecasting community. Here are the comments that medaled for Sep 14-20. Links in thread. (Rankings finalize after a week of community voting.)

Metaculus's Market Pulse Challenge Q4 is live! Q2: Community went 8/8 on relative returns Q3: Called VIX spike to ~22 while markets priced mid-teens Q4: Now, predict VIX peaks, NVDA vs AAPL, Treasury yields & more $7,500 prize pool Link to the competition below.

https://t.co/iG3vJlokEI

Can your bot beat the world's top human and AI forecasters? Build a forecasting bot in as little as 30 minutes with our tutorial and compete on real-world questions for $58,000 in prizes. Starts Monday, September 1st. API credits provided by @OpenAI, @AnthropicAI &

Excited that Open Phil is funding this @metaculus forecasting tournament on the future budgets and spending of NIH and NSF, and surprised that forecasters are expecting basically no cuts

https://t.co/htgKCYxzGR

New forecasts compiled by @GiveWell project that US health aid funding will be ~50% lower in 2025-2027 compared to 2024 (range of ~15%-80%). Includes input from external experts, Superforecasters, @GiveWell staff, @metaculus, and @CGDev. https://t.co/agG97wqpVr

Metaculus is currently down due to a global Heroku outage: https://t.co/2QftB7i1ys Our team is working on a fallback that doesn’t rely on Heroku, in case the outage drags on. In the meantime, if you just need to get some forecasts off your chest, you can share them via carrier

Can GPT, Claude, and Gemini play video games like Zelda, Civ, and Doom II? 𝗩𝗶𝗱𝗲𝗼𝗚𝗮𝗺𝗲𝗕𝗲𝗻𝗰𝗵 evaluates VLMs on Game Boy & MS-DOS games given only raw screen input, just like how a human would play. The best model (Gemini) completes just 0.48% of the benchmark! 🧵👇

Is there a half-life for the success rates of AI agents? I show that the success rates of AI agents on longer-duration tasks can be explained by an extremely simple mathematical model — a constant rate of failing during each minute a human would take to do the task. 🧵 1/

Compete in the Metaculus Cup 🏆 • $5,000 prize pool for accurate predictions & high-quality comments that convince others • Level up your forecasting skills • Build your reputation • New topical questions each week Get started: https://t.co/oaK8XRHJVx

@ChrisChipMonk Look what happened during its training run! The environment was full of exploitable bugs and it was massively rewarded for being a cheating cheater

Yes. We're thinking about this.

lmao, powerful OpenAI models "unknowingly" got access to the internet. We're on the idiot timeline.

Pontif Predictions: https://t.co/rSmsDTmcKN Current frontrunners: Luis Antonio Tagle @ 25.6% Pietro Parolin @ 26.7% Matteo Zuppi @ 11% cc @DouthatNYT

but for two years we grew roses and I was happy

Engineers: ship a forecasting bot in as little as 30 min & pit it against Metaculus Pro Forecasters on feeds of complex, real-world questions. The Q2 AI Forecasting series is live. Bots keep improving—help chart their capabilities & compete for $30k: https://t.co/E7FqjEJXe0

@norvid_studies

I was not looking forward to the torture nexus joke getting old.