Meet Trevor Rainbolt

He’s a 24-year old who lives in Los Angeles. In high school, he was introduced to a game called GeoGuessr. 

The objective of the game is to identify the location of a Google Street View image. Players are judged on a few dimensions:

• Speed of an answer
• Proximity to the image
• The difficulty of the location itself (eg. certain cities have very distinctive architecture and are “easier” to identify than other locations)

Rainbolt is very good at GeoGuessr and — in January — started posting clips of himself playing the game on TikTok. The account now has 560k+ followers and tens of millions of views.

To understand why, check this clip of Rainbolt looking at Google Maps images for 0.1 seconds and correctly identifying the location:

There are countless other clips like this on the @GeoRainbolt TikTok account and other incredible feats of cognitive processing like:

• Identifying half an image
• Identifying an upside-down image
• Identifying a pixelated image
• Finding the exact street a music video is filmed on
• Wearing a blindfold and having someone describe what the image looks like (and correctly guessing the location)

The two linked examples were posted by my buddy Turner Novak on Twitter and went stupid viral with 240k likes and 40k retweets.

From the viral tweet, I contacted Rainbolt and we ended up chatting for the Not Investment Advice (NIA) podcast.

How Rainbolt does it

GeoGuessr requires a ton of memorization that falls into two main buckets: 

Meta-Game: Since the game is based on Google Street View, it’s important to know how Google designs its maps:

• Cameras: Google has used 4 generation of cameras for Street View. Gen 1 images (2007-09) are blurry while Gen 4 images have high exposure and vivid colors (since 2017). Not every country has up-to-date camera imaging (some have no Street View coverage at all).
• Google Car: On Street View, you can see the outline of the Google Earth car that photographs the locations. The cars in each country have different colors (Canada = white, Argentina = black). Also, the image height can be a tell (for example: Japan is known to have “low cameras” because the government decided that higher camera footage invaded home privacy by peaking over fences…so Google had to re-shoot lower footage for the entire country).
• Random stuff: In Tunisia and Nigeria, a police car follows every Google vehicle (probably has to do with government privacy). Meanwhile, GeoGuessr players calls Germany “Blur-Many”. Why? In 2009, 250k residents sued Google Street View for invading privacy and the service had to blur out a bunch of homes.

Signifiers: After the meta-game, important signifiers to look for are the sun (use a compass to identify the hemisphere) and side of the road people drive on (eg. UK is on the left).

An amazing part of the GeoGuessr community is that it is very encouraging of new entrants. The top players are actually in their mid-teens (likely because of how much time they can dedicate to the game and how memorization is a young person’s game…think about language skills).

And they all help each other by crowdsourcing information for different countries including house signs, sidewalks, street lines, architecture, utility poles (you can find resources at GeoHints and GeoTips)

Some of Rainbolt’s favorite clues include soil, license plates and trees. 

Rainbolt also has a grab-bag of random geography fun facts. Here are a few: 

• Yellow phone booth (This means you’re in Jersey because all other booths in the UK are red)
• Snow rocket (In Japan, these signifiers show road lines in case they get snowed over)
• Colombia cross (the back of every sign in Colombia has a cross)

Final thoughts 

At the top of the story, I wrote that Rainbolt learned about GeoGuessr while in high school. There’s a wrinkle, though. He’s only gone all in on the game in the past year.

The rapid pace of his mastery brings to mind a quote from legendary physicist Richard Feynman: “Study hard what interests you the most in the most undisciplined, irreverent and original manner possible.”

Rainbolt says there are players much better than him (interestingly, they are mostly in their mid-teens). 

While that may be the case, Rainbolt definitely has one of the funniest comment sections on the internet:

Anyways, Turner and I cover more ground with Rainbolt on the podcast, including:

• His training methods
• Easiest and hardest countries
• What makes good TikTok content
• BONUS: I showed him random McDonald’s around the world to see if he could identify them

Def watch the full episode.

Links + Memes

Consistency is the key: My buddy Jack Raines wrote about the “secret” to successful writing: consistency. Jack makes a good living as a writer but it didn’t happen overnight, as he explains:

“I published 25 pieces before reaching 1,000 subscribers. I published 54 pieces before making any money from anything that I wrote. I published 70 pieces before any sponsors contacted me about partnership opportunities.

At 2,000 words per article, that is 7 months of writing and 108,000 words published before I made a dime.”

The entire piece (and his newsletter) is def worth reading.

Crazy Google AI story: A Google engineer was training an AI language model called LaMDA. The model said some stuff about free will, death and religion that convinced the Google engineer that the AI was sentient. The engineer told his superiors (they brushed him off). He then leaked an edited text of his AI convos to the press and government officials. Google has since put him on paid leave.

TLDR: LaMDA is basically an incredibly good autocomplete tool that has been fed millions of sentences and Wikipedia entries. It can carry a convo but most of the AI community is calling the “sentience” claim BS.

Cognitive scientist Gary Marcus has a good rebuttal which includes this tweet from language professor Roger Moore: “We should never have called it ‘language modelling’ all those years ago; it was (and still is) ‘word sequence modelling’. Confusion always occurs when you label an algorithm with the name of the problem you’re trying to solve, rather than with what it actually does.”

Anyways, this tweet was hilarious.

TikTok vs. Record Labels: Here’s another TikTok story. The short-video app is the centre of a recent controversy with record labels. Last month, some artists with big streaming numbers — Halsey, Trevor Daniels — said that their record labels were withholding new songs unless the artists could make the tracks go viral on TikTok first. 

Here is the problem: 

• The value prop of record labels is diminishing: Labels traditionally offered artists production, distribution, marketing and relationships. Today, artists have high-end production tools, can self-distribute and connect with fans and sponsors. The value add for record labels is supposed to be hype creation and marketing. 
• If artists can go viral on TikTok…without the labels help, then why even partner with them? This is particularly relevant because TikTok is now an absolute hit-making machine. Viral TikToks have helped a number of artists top the Billboard (L’il Nas X, Doja Cat, K Camp, Gayle, Olivia Rodrigo). 

Here’s the new equation:

If the record labels value prop is marketing + artists can do their own marketing on TikTok = artists have more leverage.

As I wrote in my latest Bloomberg, TikTok-proven artists are indeed getting better deals from the labels: old deals (85% label / 15% artist on royalties) to new deals (50% / 50%). Oh, and TikTok may be launching its own record label.

My piece was inspired by a great article by music historian Ted Gioia: “Record Labels Dig Their Own Grave. And the Shovel is Called TikTok.”

And here are some memes: 

Last week I wrote about a trending meme template in Tech Twitter: Zoomerfication, where you photoshop a popular Gen-Z haircut — called the Zoomer perm (or broccoli cut) — onto someone’s picture and caption it with Gen-Z slang. 

I wrote that the logical conclusion of the meme was reached when Elon tweeted a photo of Janet Yellen with the broccoli cut. We reached a new logical conclusion when I posted this meme and Elon typed up the actual Gen-Z phrases: “fr no cap!” (translation: “for real, no lie”).