It’s here.

What every AI product builds on.

Without further ado.

🤖 The GPT-4 Issue

🚀 OpenAI finally announces GPT-4

The future is now.

OpenAI finally released GPT-4, after months of speculation.

GPT, for those unaware, is the technology that powers any text-based AI tool you've seen that made you say 'wow."

This is a game-changing upgrade.

How?

For one, it’s multimodal, which is a word we definitely all know.

But for those who don't...

Multimodal means it can accept different types of data for inputs and outputs.

GPT-3.5 is text input and output. Send prompt in, get text out.

GPT-4 accepts text AND images as input. Still text as output.

What cool stuff can we do with THAT?

In a demo, an OpenAI executive showed how he sketched a drawing of a webpage. GPT-4 then turned that sketch into working code.

That feature would be an entire company before today.

But you're looking for data, right?

🤯 Unsurprisingly, a MAJOR upgrade

GPT-4 can handle over 25,000 words of text for a prompt.

Still not impressed?

OpenAI’s got book smarts too.

“For example, GPT-4 passes a simulated bar exam with a score around the top 10% of test takers; in contrast, GPT-3.5’s score was around the bottom 10%.”

See for yourself here. High marks across the board for almost all standardized testing.

Like, GPT-4 could GO TO COLLEGE.

There were two places GPT-4 fell flat.

GPT-4 still struggles with creative writing.

Same, GPT-4. Same.

It only scored a 2 for AP English Language & Composition. Out of 5.

It also struggles with Leetcode, which is the coding extension of creative writing.

So, there’s a few of us that will still have jobs.

Every other profession, though, watch out.

Here’s what's really blowing us away.

Remember how we mentioned that GPT-4 can accept images as input?

It understands memes now.

Send GPT-4 a meme, and it…gets it.

On the lame side. it improved against users trying to get around its rules.

“We’ve decreased the model’s tendency to respond to requests for disallowed content by 82% compared to GPT-3.5, and GPT-4 responds to sensitive requests (e.g., medical advice and self-harm) in accordance with our policies 29% more often.”

Buzzkills.

👀 Sick products already using GPT-4

It wouldn’t be an OpenAI announcement without customer reveals.

Here are just a few products already running on GPT-4.

DoNotPay - One click lawsuits from robocallers. It wasn’t possible in GPT-3.5, and now it is.

Diagram AI - AI designer pal, telling you about your elements and suggesting new design ideas.

Intercom AI - ChatGPT for customer service. Need I say more?

Milo AI - Solving childcare.

😎 EVEN MORE GPT-4

1. Pong in 60 seconds.

2. GPT-4 is too powerful for your puny tests.

3. GPT ate my homework.

4. Tell DoorDash what you want.

PHEW.

As always, more to come.

🤖 Got feedback for us?

Head over to our Twitter, @RobotRockDaily, and let us know what you’d like to see more or less of.