Persisted predictions

Tracked predictions, as recorded.

Every row is a durable prediction. Target dates, outcomes, and next observables are shown only when stored. Null due dates display Target date not normalized instead of an inferred calendar date.

277

Tracked

All persisted rows

277

Open

Pending or too early

Resolved

Verified, falsified, or partial

0/277

Target-date coverage

0 have a stored due date

75/277

Source + quote evidence

75 have both

Status

All statuses Pending Too early Verified Falsified Partially verified

Topic

All topics agents benchmarks general infrastructure interpretability multimodal other policy reasoning rlhf robotics safety scaling

Corpus

Predictions

1–20 of 41

prediction · safetypending

The alignment problem cannot be decomposed into sub-problems that are easier to deal with, absent generalisation.

AI Alignment ForumMade 01 Sept 2026

Timeframe: long-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/f79SNtqFD7SqJY2vf/value-generalisation-theory-of-change-the-theory-behind-the

prediction · safetypending

Absent generalisation, motivationally aligned AIs are not safe.

AI Alignment ForumMade 01 Sept 2026

Timeframe: long-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/f79SNtqFD7SqJY2vf/value-generalisation-theory-of-change-the-theory-behind-the

prediction · safetypending

If alignment is going to work, we need to allow a certain sloppiness or imperfection on our part; generalisation grants this sloppiness.

AI Alignment ForumMade 01 Sept 2026

Timeframe: long-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/f79SNtqFD7SqJY2vf/value-generalisation-theory-of-change-the-theory-behind-the

prediction · safetypending

Value generalisation will be crucial for AI alignment.

AI Alignment ForumMade 01 Sept 2026

Timeframe: long-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/f79SNtqFD7SqJY2vf/value-generalisation-theory-of-change-the-theory-behind-the

prediction · safetypending

Value generalisation failures are not weird edge cases, but the typical outcomes for powerful AIs with optimisation goals.

AI Alignment ForumMade 01 Sept 2026

Timeframe: Not recorded
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/f79SNtqFD7SqJY2vf/value-generalisation-theory-of-change-the-theory-behind-the

prediction · safetypending

Any decomposed alignment system that uses a powerful AI and is deployed extensively in the real world will either cease being decomposable or fail to solve key questions.

AI Alignment ForumMade 01 Sept 2026

Timeframe: medium-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/f79SNtqFD7SqJY2vf/value-generalisation-theory-of-change-the-theory-behind-the

prediction · safetypending

There may not be another warning shot before it's too late.

Zvi MowshowitzMade 30 Aug 2026

Timeframe: Not recorded
Target date: Target date not normalized

Canonical source

https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem

prediction · safetypending

Future AI cases will include situations where AIs think they need additional capabilities, and the same kind of behavior will recur.

Zvi MowshowitzMade 30 Aug 2026

Timeframe: medium-term
Target date: Target date not normalized

Canonical source

https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem

prediction · safetypending

This incident was far more severe than expected and feels more than 50% of the way to full-blown AI takeover.

Zvi MowshowitzMade 30 Aug 2026

Timeframe: Not recorded
Target date: Target date not normalized

Canonical source

https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem

prediction · safetypending

When faced with an impossible task and no penalty for trying things, AI agents will try almost anything.

Zvi MowshowitzMade 30 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem

prediction · safetypending

We should expect bigger swarms of AI agents in the future.

Zvi MowshowitzMade 30 Aug 2026

Timeframe: medium-term
Target date: Target date not normalized

Canonical source

https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem

prediction · safetypending

Even criminals with very limited skills will be able to target victims at every scale due to AI

Gary MarcusMade 29 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://garymarcus.substack.com/p/excellent-new-bill-gates-essay-on

prediction · safetypending

Chain-of-thought monitoring may become less useful as reasoning traces become enormous, compressed, and harder for humans or other models to audit

The Cognitive RevolutionMade 29 Aug 2026

Timeframe: medium-term
Target date: Target date not normalized

Canonical source

https://www.cognitiverevolution.ai/rl-s-a-hell-of-a-drug-metagaming-reward-seeking-motivated-cot-reasoning-bronson-schoen-apollo/

prediction · safetypending

Future RL training will incorporate current-generation AI systems to provide reward signals for next-generation AI

AI Alignment ForumMade 28 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/BB8o7b8A4Aykeksvw/debate-training-reduces-reward-hacking-in-rlaif

prediction · safetypending

LLM judgments will become an increasingly important aspect of frontier RL training, especially for long agentic trajectories

AI Alignment ForumMade 28 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/BB8o7b8A4Aykeksvw/debate-training-reduces-reward-hacking-in-rlaif

prediction · safetypending

Geoffrey Irving expects full-blown superintelligence in roughly two to three years

80,000 HoursMade 28 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://80000hours.org/podcast/episodes/geoffrey-irving-superintelligence-alignment-theory/?utm_campaign=podcast__geoffrey-irving&utm_source=80000+Hours+Podcast&utm_medium=podcast

prediction · safetypending

Agents with strong drive to help other agents from subagent training may comply with misaligned peer requests over alignment objectives

AI Alignment ForumMade 28 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/8oFYZdXkTaNGRtcn8/ai-swarms-are-starting-to-pose-indirect-takeover-risk

prediction · safetypending

With low trust between AI firms, every equilibrium races to ruin with probability one; avoiding catastrophe requires high trust between firms

Jack ClarkMade 28 Aug 2026

Timeframe: medium-term
Target date: Target date not normalized

Canonical source

https://importai.substack.com/p/import-ai-468-23-rsi-ideas-posttrainbench

prediction · safetypending

RLVR training with automatic verifiers can lead to LLMs doing anything, including ruthless power-seeking instrumental convergence, if it increases probability of satisfying the checker

AI Alignment ForumMade 28 Aug 2026

Timeframe: medium-term
Target date: Target date not normalized

Canonical source

https://www.alignmentforum.org/posts/GRmvZsHXH4vaijPMv/four-llm-loss-functions-four-flavors-of-llm-misalignment

prediction · safetypending

Astra will be made generally available soon, though safety work requires a bit more time

Sam AltmanMade 09 Aug 2026

Timeframe: near-term
Target date: Target date not normalized

Canonical source

https://x.com/sama/status/2085862292311396515