DoubleChecks

What was claimed

the only and one last thing between us and AGI is erasure. Every capability that the AI currently lacks (continual learning, infinite context, long term work, etc.) will emerge naturally from a well designed erasure centric train

Our verdict

Inaccurate

Sources on AGI describe multiple missing ingredients, including continual learning, long context, working memory, and robust decision-making, not erasure alone. The claim is a strong exclusivity statement that conflicts with this broader research consensus. Research consensus indicates AGI requires flexible cross-domain reasoning, autonomous learning, and self-improvement capabilities—none of which are addressed by erasure techniques, which focus on removing specific information from models.

2 of 3 AI systems agree20 sources citedChecked Jul 31, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

A well-designed erasure-centric training approach is sufficient for AGI

Incorrect90%
1 AI checked

The only missing piece between current AI and AGI is erasure

Incorrect94%
All 3 AIs agree

"Every capability that the AI currently lacks (continual learning, infinite context, long term work, etc.) will emerge naturally from a well designed erasure centric train."

Can’t verify86%
1 of 3 AIs agree·Claude: Incorrect, ChatGPT: Misleading

Current AI still lacks capabilities such as continual learning and very long context / long-horizon operation.

Verified90%
1 AI checked

Current AI systems lack continual learning, infinite context, and long-term work capabilities

Verified95%
1 AI checked

Machine unlearning / erasure is an active and important research area.

Verified98%
1 AI checked

Verified against 20 sources

sciencedirect.comopenaccess.thecvf.comarxiv.orgopenreview.netalignmentforum.orgarxiv.orgcomputer.orgijcai.orgarxiv.orgarxiv.orgopenreview.netyoutube.comarxiv.orgopenreview.netarxiv.orgarxiv.orglesswrong.comsalesforce.comai-frontiers.orggarymarcus.substack.com

Detailed Analysis

The response contains a mix of grounded and overstated claims. Current sources do support that erasure/unlearning is an active research area and that AGI remains unfinished, but they do not support the strong claim that "the only and one last thing" between current AI and AGI is erasure. The statement also overreaches by saying missing capabilities will "emerge naturally" from an erasure-centric training scheme, which is not established by the sources.

Why this verdict

  • The response contains a mix of grounded and overstated claims.
  • Current sources do support that erasure/unlearning is an active research area and that AGI remains unfinished, but they do not support the strong claim that "the only and one last thing" between current AI and AGI is erasure.
  • The statement also overreaches by saying missing capabilities will "emerge naturally" from an erasure-centric training scheme, which is not established by the sources.

Claims checked

"The only and one last thing between us and AGI is erasure."

Incorrect94%
Sources on AGI describe multiple missing ingredients, including continual learning, long context, working memory, and robust decision-making, not erasure alone.[6] [18] [20] The claim is a strong exclusivity statement that conflicts with this broader research consensus.

This claim may mislead readers without added context.

"Every capability that the AI currently lacks (continual learning, infinite context, long term work, etc.) will emerge naturally from a well designed erasure centric train."

Can’t verify86%
The sources show progress on concept erasure and machine unlearning, but they do not demonstrate that an erasure-centric training approach will automatically produce continual learning, unlimited context, or long-horizon work abilities.[1] [7] [8] [11] [15] This is a speculative prediction rather than a verified fact.

Current AI still lacks capabilities such as continual learning and very long context / long-horizon operation.

Verified90%
The AGI survey source states that the field is still working on missing ingredients such as continual learning and near-unlimited context, and notes current models struggle with robust decision-making in interactive environments.[6] [18] [20]

What other AI think

The statement conflates machine unlearning/erasure (a privacy/safety technique) with a fundamental architectural approach to AGI. Research shows continual learning, infinite context, and long-term reasoning require distinct technical solutions—not capabilities that naturally emerge from erasure-centric training. The absolutist claim about erasure being "the only" missing piece contradicts the scientific consensus that AGI requires advances across multiple fronts.

Why this verdict

  • The statement conflates machine unlearning/erasure (a privacy/safety technique) with a fundamental architectural approach to AGI.
  • Research shows continual learning, infinite context, and long-term reasoning require distinct technical solutions—not capabilities that naturally emerge from erasure-centric training.
  • The absolutist claim about erasure being "the only" missing piece contradicts the scientific consensus that AGI requires advances across multiple fronts.

Claims checked

Continual learning, infinite context, and long-term work capabilities will emerge naturally from erasure-centric training

Incorrect88%
Current research addresses these limitations through distinct approaches: modular memory systems, synthetic continued pretraining, in-context learning, and RAG. No literature supports that erasure-based training would produce these capabilities.

This claim may mislead readers without added context.

A well-designed erasure-centric training approach is sufficient for AGI

Incorrect90%
Research consensus indicates AGI requires flexible cross-domain reasoning, autonomous learning, and self-improvement capabilities—none of which are addressed by erasure techniques, which focus on removing specific information from models.

This claim may mislead readers without added context.

The only missing piece between current AI and AGI is erasure

Incorrect92%
Research shows AGI requires advances across multiple domains including flexible reasoning, continual learning mechanisms, and architectural innovations. Erasure refers to machine unlearning/concept removal—a privacy/safety technique, not a core path to AGI capabilities.

This claim may mislead readers without added context.

The statement asserts a single, definitive cause for achieving AGI. That claim is contradicted by current scientific understanding and is overly certain without evidence.

Why this verdict

  • The statement asserts a single, definitive cause for achieving AGI.
  • That claim is contradicted by current scientific understanding and is overly certain without evidence.

Claims checked

the only and one last thing between us and AGI is erasure.

Incorrect95%
No consensus in AI research supports that a single factor ("erasure") is the sole remaining barrier to AGI. AGI is widely considered to involve many unresolved technical, theoretical, and conceptual challenges.

This claim may mislead readers without added context.

Every capability that the AI currently lacks (continual learning, infinite context, long term work, etc.) will emerge naturally from a well designed erasure centric train

Misleading85%
There is no empirical evidence showing that an "erasure‑centric" training method alone will reliably produce all those capabilities; emergence of complex abilities is uncertain and depends on architectures, objectives, data, and evaluation, not a single training focus.

This claim may mislead readers without added context.

Share this result

Try DoubleChecks