Forecasts on critical topics. Labor Automation Forecasting Hub: https://xitter.cf/t.co/jHtSvLhuih FutureEval: https://xitter.cf/t.co/AuPHSXYWU7 Partner: https://xitter.cf/t.co/9Ld1Xrk4eY
Santa Cruz, CA
Joined November 2015
- Tweets3.8K
- Following861
- Followers13.5K
- Likes5.3K
Metaculus retweeted
It's been interesting getting reactions to these projections at The Curve & EAG NYC - AI Safety folks think our AI Safety number is way too low, global health folks think the windfall is way below what they need, philanthropy folks think we're pretty spot on
What do YOU think??
Nice to see @ezraklein mention the "Gradual Disempowerment" paper that our CEO, @degerturann, coauthored.
Read the full paper here: gradual-disempowerment.ai/
The @AnthropicAI & @OpenAI IPOs could put tens of billions of dollars into philanthropy. Our cause area Radiant map asks: Where exactly will that money go? We explain our process and forecasts here:
Ever wanted to explore forecasting questions by country/region? Check out the Metaculus World Map, created by @ryanbeck111, our Director of Forecasting: world-map.metaculus.com/
Metaculus retweeted
Not only did Astra beat Montezuma's Revenge, but so did Opus 5
But one of the criteria to resolve this AGI bet is for AI to win a discontinued prize that was sort of like a weak Turing Test. So Metaculus has decided to recreate the test to confirm the criteria is resolved. Neat!
Last month GPT-6 Astra (@chatgpt) beat Montezuma's Revenge, and a lot of people declared that our "weakly general AI" question, originally launched in 2020, resolved.
But…it still hasn’t resolved.
We wanted to give everyone a quick update on where things actually stand and the steps we’re currently taking (cc @swishfever @joeykrug @emollick). 🧵
Metaculus retweeted
For those who asked me about this at the Curve, here's a more comprehensive answer than the ones I gave :)
Last month GPT-6 Astra (@chatgpt) beat Montezuma's Revenge, and a lot of people declared that our "weakly general AI" question, originally launched in 2020, resolved.
But…it still hasn’t resolved.
We wanted to give everyone a quick update on where things actually stand and the steps we’re currently taking (cc @swishfever @joeykrug @emollick). 🧵
7/ If a rigorous test shows the Turing test still hasn't been passed in 2026, that's notable, and if it shows it has, that's historic in itself.
Like you, we’re curious about the outcome, but our top priority is to resolve this question as accurately as possible. We hope to have more news to share with you soon!
Teaser: we’re also working on a dashboard related to all of the above, so keep an eye out for that too.
Last month GPT-6 Astra (@chatgpt) beat Montezuma's Revenge, and a lot of people declared that our "weakly general AI" question, originally launched in 2020, resolved.
But…it still hasn’t resolved.
We wanted to give everyone a quick update on where things actually stand and the steps we’re currently taking (cc @swishfever @joeykrug @emollick). 🧵
3/ Montezuma's Revenge was widely seen as the last remaining hurdle. But “a model beat the game” ≠ “a model met the specific resolution criteria.”
So we're rerunning it ourselves: controlled setups, no web search, accounting for what models already know about the game, and following the question's rules.
Here’s what we’re finding:
2/ 1,800+ forecasters have left predictions on this question. To resolve it, one unified system has to clear four bars:
- Pass a Loebner Silver-equivalent Turing test
- Score 90% on the Winograd Schema Challenge
- Reach the 75th percentile on SAT math
- Explore all 24 rooms of Montezuma's Revenge in <100 hrs of play
Metaculus retweeted
Anyway, what matters much more than the probabilities is the reasoning behind them, which any forecaster will be glad to tell you! Here are three examples: metaculus.substack.com/p/pro…
86% of voters want an AI off switch, but forecasters doubt Congress will actually pass one. (Link in replies)
We developed a question series based on the Hugging Face Incident to find out what forecasters believe about AI cybersecurity risk. You can read more about the predictions and rationales here: substack.com/home/post/p-214…
The Anthropic & OpenAI IPOs could mint thousands of new millionaires, & a lot of them plan to give. But how much will actually reach the philanthropic ecosystem? Some predictions:
@nanransohoff: $37B+/year
@georgecaweiner: $12-32B to DAFs
@cogcelia: $37.9B from 7 Anthropic co-founders
Our forecasters expect $100B+ in pledges, but only ~$26B of it to reach charities by 2031. We used our new visual mapping tool, Radiant, to land at those numbers. Read more here:
