🍪 Gemini is here

Happy Thursday, friends. Remember me?? 

I know, it was a bad time to get sick. 

Sam getting fired from OpenAI, CZ pleading guilty to money laundering charges, Mark Cuban leaving Shark Tank, Thanksgiving with my 7 aunts - all the damn drama happened while I was gone!!

I’m sorry, I left ya in the dust. But don’t worry, I was hoarding all the good memes for ya from bed. 

So let’s get to the good stuff.

In the oven this week:

• 🤖 Gemini is finally here 

• 🗳️ Poll of the week: choose your AI

• 💬 Tweet of the week: wrapping up the year 

• 🍪 Cookie crumbs: 6 bite-sized headlines 

• 🍫 Chocolate chips: our 4 favorite finds this week

• 😂 Snickerdoodles: memes on memes for you today

FRESHLY BAKED

GEMINI IS HERE

Well ladies and gentlemen - it’s finally here. 

No, not the early bird special at Arby’s. 

Google finally released Gemini. 

They’re calling it their “most capable AI model” and there are 3 versions of it (size ascending): 

1. Gemini Nano, which is now running on device on Pixel 8 Pro devices (mostly used for stuff like smart replies or summarization) 

2. Gemini Pro, which is baked into Bard now, and available to everyone

3. Gemini Ultra, which is being tested with select Bard Advanced users and will roll out more broadly next year

Sounds like they took some naming tips from Apple… what’s next, the Gemini Pro Max?? 

Gemini’s a multimodal model, which means that it understands text, images, and videos - Google has a demo of it, which is actually pretty cool:

Alright alright, but how does it fare against other models? 

The main thing to know is that the Ultra version beats all existing AI models. 

Yup, it’s the new big boss in town.

It scored 90% on an industry-standard benchmark called MMLU (massive multitask language understanding). This benchmark comes from a set of tests that measure the performance of models on tasks involving logical fallacies, moral problems in everyday scenarios, medical issues, economics and geography, even multiple choice questions on physics.

This is the first time AI has beaten humans at the test, and is the highest score for any existing model. Here’s how everyone else did on the same test: 

• GPT-4: 87%

• LLAMA-2: 68% 

• Claude 2: 78.5%  

Interestingly, it’s not a huge leap over GPT-4. And the mid-range Pro version of Gemini beats OpenAI’s GPT-3.5, but not GPT-4.

Impressive? 

Sure. 

Does it live up to the hype? 

Ehhhhh.. 

After it got postponed a bunch of times and even got Sergey to come back to work on it, we expected a huge breakthrough, not just a slightly better version of GPT-4. 

Seems like investors felt the same way, since Google stock has stayed flat since the announcement.

Guess we’ll all have to keep waiting…

FROM THE COOKIE JAR

POLL OF THE WEEK: CHOOSE YOUR AI

Alright let’s hear it friends, are you team Bard or team OpenAI? What are you using these days?

THE BIG PIZOOKIE

TWEET OF THE WEEK

Harvest those losses, friends. It’s our time to shine.

COOKIE CRUMBS

BITE SIZED HEADLINES

CHOCOLATE CHIPS

OUR FAVORITE FINDS

A little sprinkle sprinkle: 

• Taylor Swift’s website (pre-superstar version)

• Questions that spark conversations on a deeper level

• Time’s 200 best inventions of 2023

• Apple awards for the best apps and games of the year

SNICKERDOODLES

THURSDAY MEMES

 😂 

What'd you think about this email?

• 🍪🍪🍪 Top notch!

• 🍪🍪 Meh... could use more milk

• 🍪 Undercooked, do better