EY

Eliezer Yudkowsky

XLessWrong
About

Eliezer Yudkowsky is an American artificial intelligence researcher, writer, and decision theorist best known for his work on the risks and alignment problems of advanced AI systems, particularly the concept of friendly AI and AI alignment.

26
Total Predictions
18
Resolved
25.0%
Accuracy Rate
4
Correct
Prediction History
GPT-5 may be more unambiguously a general intelligence
Predicted: April 1, 2023
Due: January 1, 2025
Resolved: September 18, 2025
Incorrect

Source: Lex Fridman interview April 2023

GPT-4 level systems will not cause a catastrophe
Predicted: March 1, 2023
Due: March 1, 2024
Resolved: September 16, 2025
Correct

Source: Lex Fridman interview March 2023

Metaculus forecasters will update their AGI timelines in predictable directions that Yudkowsky anticipated 6 years earlier
Predicted: January 1, 2023
Due: January 1, 2024
Resolved: September 18, 2025
Correct

Source: https://www.facebook.com/yudkowsky/posts/10160260422389228

AGI timeline is 3-16 years as plausible, 50 as implausible
Predicted: January 1, 2023
Due: January 1, 2039
Unresolved

Source: Multiple interviews and posts in 2023

AI systems will develop internal preferences that don't align with human values during self-improvement
Predicted: January 1, 2023
Due: January 1, 2030
Resolved: September 18, 2025
Correct

Source: Multiple recent writings and interviews

An AI system will exhibit deceptive alignment before 2030
Predicted: January 1, 2023
Due: January 1, 2030
Unresolved

Source: Various alignment forum discussions

Artificial superintelligence will exist by 2030
Predicted: December 31, 2022
Due: January 1, 2030
Unresolved

Source: https://manifold.markets/EliezerYudkowsky/will-artificial-superintelligence-e

AI will wipe out humanity by 2030
Predicted: December 31, 2022
Due: January 1, 2030
Unresolved

Source: https://manifold.markets/EliezerYudkowsky/will-ai-wipe-out-humanity-by-2030-r

AI video generation will be near-indistinguishable from real footage
Predicted: August 26, 2022
Due: December 31, 2024
Resolved: September 18, 2025
Correct

Source: https://twitter.com/ESYudkowsky/status/1563282607315382273

Current AI alignment research will fail to solve the alignment problem
Predicted: January 1, 2022
Due: January 1, 2030
Unresolved

Source: AGI Ruin: A List of Lethalities and related posts

Protection against COVID infection will not decline by an order of magnitude while hospitalization protection remains unchanged
Predicted: October 6, 2021
Due: June 1, 2022
Resolved: September 16, 2025
Incorrect

Source: COVID analysis posts

COVID-19 vaccine breakthrough infections will be rare
Predicted: April 1, 2021
Due: December 31, 2021
Resolved: September 16, 2025
Partially Correct

Source: COVID prediction posts on LessWrong

The next major AI breakthrough will come from novel architectures, not scaling
Predicted: January 1, 2020
Due: January 1, 2023
Resolved: September 16, 2025
Incorrect

Source: Predictions about AI development patterns

Fast AI takeoff will occur without preceding 4-year doubling periods
Predicted: January 1, 2020
Due: January 1, 2035
Unresolved

Source: Various alignment forum posts

Large language models will not scale to general intelligence capabilities
Predicted: January 1, 2020
Due: January 1, 2023
Resolved: September 16, 2025
Incorrect

Source: Historical skepticism about LLMs referenced in interviews

World will be ended by nonaligned AI before January 1st, 2030
Predicted: January 1, 2017
Due: January 1, 2030
Unresolved

Source: https://www.econlib.org/archives/2017/01/my_end-of-the-w.html

Higgs boson will not be discovered
Predicted: January 1, 2010
Due: January 1, 2015
Resolved: September 16, 2025
Incorrect

Source: https://www.lesswrong.com/posts/ZjikPZ67CvR64YvMK/open-thread-november-2009?commentId=HCBJZAJbXDGibLedH

AI progress will be extremely discontinuous and localized
Predicted: January 1, 2008
Due: January 1, 2030
Resolved: September 16, 2025
Incorrect

Source: Hanson-Yudkowsky FOOM debate 2008

Economic output could rise by twenty orders-of-magnitude within a week during intelligence explosion
Predicted: January 1, 2008
Due: January 1, 2030
Unresolved

Source: Hanson-Yudkowsky FOOM debate 2008

A brain in a box in a basement scenario is the median development path for AGI
Predicted: January 1, 2008
Due: January 1, 2030
Resolved: September 16, 2025
Incorrect

Source: Hanson-Yudkowsky FOOM debate 2008

AI development will require very little computing power
Predicted: January 1, 2008
Due: January 1, 2025
Resolved: September 16, 2025
Incorrect

Source: Hanson-Yudkowsky FOOM debate 2008

Deep learning/neural networks will not lead to AGI
Predicted: January 1, 2008
Due: January 1, 2020
Resolved: September 16, 2025
Incorrect

Source: Historical positions referenced in critiques

Whole brain emulation will come before artificial general intelligence
Predicted: January 1, 2008
Due: January 1, 2030
Resolved: September 16, 2025
Incorrect

Source: Hanson-Yudkowsky debate context

SIAI team will build AGI between 2005 and 2020, probably around 2008 or 2010
Predicted: January 1, 2001
Due: December 31, 2020
Resolved: September 16, 2025
Incorrect

Source: Early 2000s SIAI documentation

Final stage AI will reach transhumanity sometime between 2005 and 2020
Predicted: January 1, 2001
Due: December 31, 2020
Resolved: September 16, 2025
Incorrect

Source: SIAI early documentation

Nanotechnology will arrive no later than 2010
Predicted: January 1, 1999
Due: December 31, 2010
Resolved: September 16, 2025
Incorrect

Source: Forum post from 1999, original link not found