DOOMBENCH.COM
Safety and alignment

Sutskever forecasts reasoning AI will become less predictable

At NeurIPS, Ilya Sutskever said pretraining gains were reaching limits and forecast systems with reasoning, agency, and self-awareness. He argued that systems capable of reasoning through outcomes would be less predictable than present pattern-matching models.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

Send detailed or private feedback →
CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM27confidence 60/100

Why it moved the index

The remarks add a distinct mechanism linking reasoning and agency to reduced predictability. They are a dated forecast and do not prove the forecast systems already existed.

Assessment history

  1. R1
    Toward 27 · confidence 60

    Historical people backfill adds a dated, attributable forecast about post-pretraining capability and predictability.

    14 Aug 2026

Share this page

DoomBench social sharing card for Sutskever forecasts reasoning AI will become less predictable.

  1. DoomBench assesses “Sutskever forecasts reasoning AI will become less predictable” as evidence moving toward doom, with magnitude 27 and confidence 60 out of 100 in the safety and alignment category.

  2. The DoomBench assessment of “Sutskever forecasts reasoning AI will become less predictable” is based on reporting from Reuters and records the editorial rationale, source quality, attribution, and revision history.

  3. DoomBench summarizes “Sutskever forecasts reasoning AI will become less predictable” as follows: At NeurIPS, Ilya Sutskever said pretraining gains were reaching limits and forecast systems with reasoning, agency, and self-awareness. He...

    https://www.doombench.com/news/sutskever-forecasts-reasoning-ai-will-become-less-predictable-2024-12-13