Frontiers

Unsolved frontiers

There is no invented percentage "to AGI" here. Instead, we show concrete capabilities still missing from reliable general intelligence.

Gates before AGI

Four open areas: learning, reliability, action and resources. These are development requirements, not an AGI date forecast.

Learning from experience

Agents already preserve files, notes and history between tasks. Reliably turning experience into new skills without losing earlier ones remains an open problem.

Full dossier

Reliable autonomy

An agent must detect mistakes, respect access boundaries and let people intervene. Passing a particular test does not make it reliable in every setting.

Full dossier

Acting in the physical world

A model can describe an action, but the real world requires continuous perception, physical reasoning and correction before errors become irreversible.

Full dossier

Compute and power bottleneck

AI capability is constrained by more than algorithms. Chips, memory, data centres, electrical grids and cooling determine which models can be trained and how many runs the economy can sustain.

Full dossier

AGI

There is no agreed arrival date. New model records do not automatically resolve reliability, experience transfer and generality.

Full dossier

ASI

Neither an arrival date nor a path is established. Self-improvement is one proposed scenario, not a guaranteed property; compute, reliability and control remain separate questions.

Full dossier
Forecast archive

Forecast archive

Previously collected public estimates. Historical views, not a current release schedule or scientific consensus.

Sam Altman
The Intelligence Age
No exact date

Suggests superintelligence within a few thousand days while allowing that it could take longer. An opinion, not a release calendar.

Published · 2024-09-23
Anthropic
U.S. AI Action Plan
Late 2026 — early 2027

The lab's historical forecast for powerful AI: expert capabilities and autonomous work. Neither a particular model promise nor a universal AGI definition.

Published · 2025-03-06
Daniel Kokotajlo / Eli Lifland
AI Futures
2028 / 2030 · coding automation

The two authors' Automated Coder medians are mid-2028 and mid-2030. This specific economic coding-automation threshold is not an ASI arrival date.

Published · 2026-04-02
James Landay
Stanford HAI
Not in 2026 · opinion

Landay's annual outlook did not expect AGI in 2026. This position belongs to the named author, not to all Stanford researchers.

Published · 2025-12-15
Dan Schwarz
FutureSearch
Forecast comparison

Compares updates from selected forecasters. The shift toward earlier timelines refers to January–April 2026 in this sample, not to all forecasts today.

Published · 2026-04-12
Katja Grace et al.
Thousands of AI Authors on the Future of AI
2047 · survey 50% estimate

Historical survey: an aggregate estimate for machines outperforming humans on every task. The same study places full automation of all occupations much later, in 2116; definitions matter.

Published · 2025-10-08