Cross-Language Fact-Checking Is Pretty Cool

One nice thing with SIFT Toolbox compared to search is that it inherits the language capabilities of the underlying models

So Sunday morning this went viral:

It clearly implies that this “white object” and its removal by Macron is potentially evidence of a drug scandal, perhaps a bag of cocaine that was quickly moved out of a press picture.

You can watch the video if you want, but I’m not going to link it. Framed like this though it does look suspicious. Macron is about to have picture taken, sees the white object on the table, and smoothly grabs it and pulls it out of frame before pocketing it. Scandal!

Seeing this at about 11 a.m. my time that morning, I was pretty sure it’s bullshit, but you know me. I can’t help myself. I’m a checker. So I throw some terms into Google.

The top result from that search is from a person hawking a prepper guide in their pinned tweet. All the other results have nothing to do with the incident at all. (The following screenshot was actually taken a bit after I ran the query, at about noon my time, after I decided to write this up).

Keep in mind when I’m doing this it’s 11 a.m. PDT, about four hours after the video went viral. It’s got six million views. And it looks like there are no fact checks yet. I play around with search terms a bit but there’s nothing.

So I put the SIFT Toolbox prompt into Gemini and upload the screenshot. You can view the entire output here, but here’s the corrections summary portion.

Clicking through, the source that it might be a handkerchief is Greek…

But Google Translate makes short work of that:

Drawing off of work done by a French site, the article shows a detail of a high resolution photo from the meeting, where the artifact is revealed to be a tissue:

Yep, the big secret here is that Macron didn’t want his dirty tissue memorialized in the press photos. Scandal!

I used to search across different languages in fact-checks, but it was never easy. Here it’s transparent to the user. I do feel this is another reason that LLMs can be better at search than traditional search, and will ultimately be at the core of most search and fact-checking.

Update #1: Community Note

So I did not community note this item, but it occurred to me that I could show you all how to generate a community note using SIFT Toolbox.

Simply upload the photo and go through the process above (when it asks you about the searches just say “all of them”).

Once it’s done with that just type “cnote” and it will generate something called a “context report” followed by this:

Run through it and understand it. Edit it, check the links. Make it your own. Etc. But if you want to community note anything it will get you started if you’re looking at a blank page.

Update #2: The Fact-Checks Have Arrived

It’s now three and a half hours after I did my SIFT Toolbox search, and the English-language fact-checks have arrived on the Google Search page.

The tweet currently stands at 11 million views.