AI That Knows You

I use Google Docs, sheets, and Drive a lot. I was recently using Bard to chat with a collection of my docs. They support the mention of @Drive in a prompt and that allows Bard to sweep up documents relevant to the prompt. It’s quite powerful. It also supports your Gmail history, so I can do stuff like this.

I decided to marry a folder of my Obsidian documents with my Google Drive docs by synching the folder into Drive. This allows me to use Bard (or the Gemini-Pro API) with select content from my Obsidian vault in AI conversations. I can also perform generative AI summations of Obsidian notes + email threads about those topics, plus related spreadsheets or documents.

I think when the LLM model knows all that you know and all that you once knew and may have forgotten, it becomes a superpower. In seconds I can create a briefing for a meeting that is generated based on all data related to that meeting topic.

Being quite new to Obsidian, but rapidly becoming a fan, I wonder – is there a better way to push or sync MD files into Google Drive?

1 Like

Sorry for being off-topic, but don’t you have any concerns allowing the AI access to all of that private data?! I’d be terrified having all of that information, allowing those engines access to it to do whatever they want with it? And no, I don’t trust them not to sell either of or all of the metadata of content onwards to other companies.

3 Likes

Not without understanding how Google uses your Gmail and Google Drive. If you stick to those realms, the ToS is clear - it’s private unless you openly publish it, and even then, they don’t have free reign. Google, unlike OpenAI thought about copyrights in AI models a decade ago to avoid exactly the lawsuit NYT has brought against them.

Think about it - if Google was able to use your Gmail archive as it pleased, it would have done this long ago. The reality is they can’t and won’t because it would collapse the global accounts for their 1.8b users on Workspaces.

But there are other nuances as well. If you use Bard and you paste content into a prompt. Google says straight up, they will use that information for training future LLMs. ChatGPT is the same. There are conditions that allow you to opt out and still use ChatGPT, but they disable a lot of features if you do.

What no one seems to know is that any introduction of your private data into a Bard prompt through an extension - i.e., @Drive, or @Gmail - is off limits to Google. They will not copy that data. They will only use it in the prompt to compel the LLM to use data included by @ reference.

How and why does Google do this?

While the world believes all data used with generative AI is up for grabs, this could not be more inaccurate. All interchanges between OpenAI, Google, and Microsoft that flow through APIs are opt-out conversations. So, if you use Coda AI, those LLM conversations travese the AI. Imagine if all third-party apps and co-pilots like Coda created the risks you describe. It would have been front-page news a year ago.

Yeah, that would be terrifying for all those who are misinformed.

There’s no shortage of breaches and intentional bad actors. But, if Google wants that $15b/month for Workspaces and another $30b/month for Duet and other AI products, its in their best interest to guard your data with precision.

1 Like

perhaps a simple sync tool …

e.g. on Android I use X-Plore File Manager’s File Sync

[free: ads +1 folder pair | one-time $5: no ads + unlimited folder pairs + scheduling]

I don’t use this as my main Obsidian syncing tool - however I do have it sync my entire Obsidian vault every 8 hrs both with Google Drive & Dropbox as a backup especially with those services having version history.

1 Like

Better than what? How are you syncing them now?

Also, better how? Does your current setup lack in some way?

I’m sorry, I’m currently not able to read Markdown files. I can only read text files in the formats .txt, .html, .doc, .docx, .pdf, .rtf, and .epub.

Are you actually seeing results from your md files?

I’m using Google Drive (the MacOS app that maintains a sync in Drive. This is a heavy-handed app that consumes memory and is known for issues that range from missing files to deleting local files under some circumstances.

I was hoping there might be some webhook magic in Obsidian that could push changed files into a web service.

Yes, but only because I developed an Apps Script trigger that transforms the synced MD files into DOC files in Drive. This is a pain in the but, but it works.

I’ve also merged some MD files into a single doc based on folders. This is also working for me and provides some interesting generative AI outcomes.

This is why I asked about a better way to move content into drive from Obsidian in the first place. Ideally, I’d like an event webhook to send them to an Apps Script service which would transform them into doc files, pre-fetch some generative AI summations, etc., and then drop them into drive ready to go.

There is a community plugin called Remotely Save, but last I knew it doesn’t support Drive yet.

You might look into third-party apps. If Drive’s app is that bad, someone may have made something to replace it. I know there are some apps like this on Android (at least 1 for Dropbox and 1 for various).

1 Like

I wonder if exporting to html would work. It was one of the formats that the error message said it could read. I know there are several ways to export to html in Obsidian. Just need a way to automate it.

Good idea - I’ll try that. In Google Apps Script it’s very easy to convert a DOC into HTML, PDF, etc.

I think there’s a pathway to full-on Obsidian → Gemini-Pro AI that will emerge in the next two weeks or so.

The Gemini-Ultra release is set for the end of Jan and with it, there is speculation that custom plugins for Bard will be announced. Essentially, the @Drive and @Gmail features in Bard are extensions. It may be possible [soon] to create an @MD extension, thus sweeping up synced Markdown files in Google Drive or from arbitrary web services.