Your documents are on trial

Hi, and happy Tuesday.

Elon Musk wants $134 billion. He's accusing Sam Altman of stealing a charity.

(For a deep dive on the courtroom activity, check out my LinkedIn post)

But, before we laugh at the billionaires, we should probably look at our own SharePoints - because there’s a key lesson from this in how we organize our documents.

Here's what's happening in the Elon Musk v OpenAI trail.

The witness list reads like a Davos seating chart - Satya Nadella, Ilya Sutskever, Mira Murati. OpenAI’s $1 trillion IPO hangs in the balance, with Microsoft's ~27% investment in OpenAI twisting in the wind, and two of the most powerful men in tech publicly accusing each other of betraying humanity's future.

It's the trial of the decade.

But strip away the zeros and the egos, and what's actually on trial is something far more mundane - and far more relevant to all of our own everyday work…

Old documents that tell an inconsistent story.

Lawyers are reconstructing a decade of intent from scattered artifacts: diary entries, email threads, board memos from 2017 - and now testimony. Each is from a different person, and each captures a different version of what was "agreed."

For example, the “smoking gun” that triggered this trial was a diary journal by OpenAI President Greg Brockman in 2017 suggesting he and Sam Altman were secretly planning to turn OpenAI into a for-profit organization, even while assuring Elon Musk that his $38M in donations were for a "nonprofit for the benefit of humanity."

Damning, right?

But OpenAI's lawyers are presenting evidence that even Elon Musk suggested that OpenAI should convert to a for-profit - except with him in charge, through OpenAI becoming part of Tesla.

So we now have documents with two versions of the truth. 

The judge's job is to figure out what the real story is.

The entire case - with billions of dollars, and the future of the most important AI company on Earth, on the line - hinges on document archaeology.

And it's a story we see playing out in some shape or other in every organization we work with.

The dream

You’ve been pitched the dream: connect Copilot to all your documents and it'll answer anything. 

Plug it into SharePoint, Google Drive, Confluence, and the shared drive nobody's cleaned since 2019. Let the AI sort it out and get ready for a revolution in productivity! 

Here are examples of what we’ve found in the real repositories of actual teams we've worked with:

• Three versions of the same policy in two locations.

• A folder called "FINAL" that contains seven drafts.

• A document titled "FINAL_v2_actually_use_this_one" that everyone ignores in favor of the PDF sitting in email inboxes since in 2022.

• Strategy documents that were quietly abandoned but never deleted. 

• Templates that were "temporary" 7 years ago. 

And then we hand these all over to an AI and ask it to be the judge.

The result?

The same inconsistent narratives we see on trial in Elon Musk vs OpenAI. 

But the judge presiding over these documents - Copilot or whatever AI you use - will not carefully consider the evidence to come to a conclusion.

The AI will read whichever document it finds first, decide that's the truth, and give you an answer with absolute confidence. It won't hedge. It won't ask which version is current. It will simply pick a side and commit.

The model isn't hallucinating. It's doing exactly what you asked. Handed a contradictory archive and trained to provide a single answer, it gives you exactly that.

But you have an advantage Musk and Altman don't. 

They can't go back and clean up 2017. You can clean up now, before the AI puts your documents on trial.

Here are three simple techniques we’ve found to be surprisingly effective:

1. Curate the documents given to the AI. For every important topic - policies, processes, strategies, templates - put them in a specific drive and give the AI access to only that drive. The goal is to make the content unambiguous. It may sound tedious, but it is a lot easier than people imagine - just add the documents you need to the drive when you need it, rather than turning it into a special chore.

2. Give the AI context. Save and use prompts that give the AI the same information you would give a new starter, e.g. “always use documents in folder xxx and only use other documents if you don’t find it there.” Treating it less like a search engine and more like an intern helps drive it towards more accurate behavior.

3. Let the AI maintain what the AI depends on. Once you've established a clean baseline, you can set up scheduled tasks for the AI to sweep through your documents to flag duplicates, surface version conflicts, catch stale documents, and route new content to the right home. 

The verdict

When the verdict in the Musk vs OpenAI trial lands and the headlines fade, remember what you actually saw: two of the smartest people in technology arguing over inconsistent narratives.

These same inconsistent narratives live in our SharePoint repositories and are probably already driving “meh” results when your Copilot attempts to make sense of them.

Fortunately, we have the benefit of being able to clean up these narratives.

Best,

Dino

ps. For a deeper dive on what’s going on in the courtroom, check out my LinkedIn post.

https://www.linkedin.com/posts/dinogane_theres-no-live-broadcast-of-elon-musks-activity-7456571096147021824-swoc