When in doubt, try stealing!

Overview

The video is a satirical-but-accusatory commentary arguing that major AI labs increasingly “steal” intellectual property while avoiding consequences. It uses multiple legal and research examples—especially involving Anthropic and OpenAI—to illustrate this pattern.

Key Arguments and Reported Cases

“Stealing” as a business model

The speaker claims AI companies benefit from others doing the hard creative work—such as:

• books
• music
• art
• code

Then AI labs supposedly take those outputs for training or commercialization, later trying to “step in” at the end to capture rewards with minimal harm or accountability.

Anthropic alleged piracy of text libraries / training data

The video cites an Anthropic $1.5B settlement related to pirating digital libraries. It also claims an internal project (“Panama”) involved destructively scanning millions of physical books to train models while avoiding licensing issues.

Copyright and publicity disputes across creative industries

The video references disputes in multiple creative fields:

• Musicians vs. Suno
  • A class action is referenced alleging right-to-publicity violations and voice cloning.
• Music publishers vs. Anthropic
  • Publishers are said to have sued over alleged misuse of tens of thousands of copyrighted songs to train models.
• Visual artists vs. multiple AI image tools
  • Tools named include Stability AI, Midjourney, DeviantArt, and Runway AI.
  • The video claims courts allowed key copyright claims to proceed toward trial.

Open-source coding: “copyright is laughable”

The speaker argues coding models were trained on large quantities of open-source code without permission or payment, including code under licenses that require copying/redistribution conditions.

A specific example is given where developers allegedly tried to resist the use of content (e.g., by editing Stack Overflow answers). The video claims the platform restored original content and suspended accounts—leaving contributors feeling their work was used anyway.

The “Millennium Prize” Controversy (Central Focus)

Core claim

The video claims OpenAI is trying to claim credit for a second Millennium Prize math problem (from the Clay Mathematics Institute, $1M per problem), framing this as potentially improper given how it allegedly connects to work by outside researchers.

Provided context

• Only one Millennium problem is said to be solved so far, citing Grigory Perelman, who reportedly declined the prize/credit.
• The speaker frames an “audience hype cycle,” contrasting an earlier OpenAI “solution” story that allegedly faded without an official true solution.

Alleged research sequence

Two researchers are described:

• Tristan Buckmaster (NYU)
• Levon Alpoj (working at Anthropic, per the video)

They are said to work for months on the Navier–Stokes existence and smoothness problem via a “smooth forcing” approach.

The video claims:

• Promising results were obtained around Aug 15
• By Aug 22, one result was verified
• They planned to publish after confirmation

Allegation against OpenAI

According to the video:

• Once information about Buckmaster/Alpoj’s progress circulated, it allegedly reached OpenAI.
• On Sep 3, Buckmaster reportedly contacted an OpenAI mathematician to clarify the work was personal and intended for publication.
• Soon after, Buckmaster allegedly attended a call where OpenAI stated an internal model produced a roughly 100-page proof using the same smooth forcing approach.

Why it’s presented as suspicious

The video argues the smooth-forcing route was highly specific and not broadly pursued. It suggests OpenAI’s proof effort involved:

• a large internal team
• “insane” compute
• starting only after hearing of the researchers’ progress

It also claims:

• Buckmaster/Alpoj were sharing drafts using code tools (“codecs/code”)
• when Buckmaster asked whether their private material was used in model training, OpenAI did not provide an answer

The video emphasizes that Buckmaster/Alpoj were not claiming certainty the proof was stolen, but it suggests the circumstances imply “IP theft” and unfair advantage.

Overall Tone / Conclusion

The speaker repeatedly frames the message as:

If AI companies don’t steal and avoid consequences, they’re “missing out.”

It ends with a call for questions in the comments and standard like/subscribe prompts.

Presenters / Contributors (As Mentioned)

• Tristan Buckmaster (NYU)
• Levon Alpoj / Alpo(j) (Anthropic)
• Diego Córdoba
• Luis Martinez
• Gregory Perelman (Grigory Perelman)
• OpenAI internal mathematician (named as “a prominent mathematician working at OpenAI,” but not identified)
• The video narrator/speaker (not named in the subtitles)
• Claude and Codex (mentioned AI models used in the research; not presenters)
• Chad GPT (mentioned colloquially as ChatGPT; not a presenter)