What's up? You're reading Inclined AI. If you slept in today, we’ll chalk that up to your post-Prime Day comatose. Your boss will forgive you for it, don’t worry.

Here's what we’re buying:

• A leak of GPT-4 details took the internet by storm

• Anthropic releases Claude to the general public

• That “No Moat” memo was real

• China clarifies their AI content rules

Be a sponsor of our newsletter.

THE FACTS ABOUT GPT-4 ARE OUT

When OpenAI announced GPT-4, we all waited with eager ears to hear how big the model was. Then OpenAI started closing some of their research since the competition was heating up.

Today we can confidently tell you that it’s approximately 1.8 trillion parameters and cost around $63 million to train. But this info is all from a leak getting torn from the internet faster than Barry Allen on an espresso binge.

We’re risking a lot writing about this news.

Every source I try to hyperlink here gets removed for “copyright claims” before I finish writing. In other words, OpenAI does not want people to know the dirty details.

But one term stuck out and is worth discussing since it’s not a secret concept. It’s the Mixture of Experts (MoE) model that OpenAI researchers infused into GPT-4.

So, what is MoE, and why should you care?

If you’ve tried GPT-4, you might notice that it sometimes feels like the AI is shifting from persona to persona based on the input. We can’t confirm that it’s because of the 16 experts they put into this model, but that feels like the case.

MoE is like “Ben Ten,” if you remember the show. If you don’t, a teenage boy turns into different aliens using a slick watch, typical cartoon shenanigans.

Each alien has special powers that Ben uses for unique situations.

That’s what GPT-4 is, with its 16 experts comprising billions of parameters and unique knowledge. AI developers use MoE to increase model capacity without increasing computational energy at the same rate. 

The odds are GPT-4 has so many parameters because they stuffed it to the gills with these experts, and future model versions could include even more of these network segments.

Expect open-source developers to take a cue from this news and try their own versions of this architecture.

Digital Asset Investor → learn to invest in online businesses & domains from an expert

The Sweaty Startup → for anyone interested in making money & becoming better at managing people

A Smart Bear → a unicorn founder discusses strategy & decision-making for building

Life-Changing Concepts → gives you an unfair advantage in business using mental models

Master ChatGPT → an e-book to take you from zero to hero

ANTHROPIC FINALLY ARRIVES

Claude 2 is out! It took a second, but Anthropic decided that the world waited long enough to test their ChatGPT competitor.

And by the world, I mean people in the US and UK. Sorry guys, that’s the only places with access right now. Don’t worry; access will roll out over time.

What’s the gist?

Claude is fun. It’s not connected to the internet, but neither is GPT-4. So you can toss that comparison out the window.

Overall, Claude is unique but not a giant slayer.

For example, I always test models reasoning with a simple riddle. I ask them which month has 28 days and tell them the question is a riddle.

The answer is all of them, but most people (and AI language models) will say February. Claude was no exception. GPT-4 is the only one who’s nailed that question.

The part that blew my mind was how Claude answered. It was the most authentic, human-like response I’ve seen from an AI model. It didn’t feel like a dumber version of ChatGPT.

Claude is sharp and dynamic. There’s a freshness to talking with it.

But that novelty wears thin.

The true benefit to Claude is the context window. You can give it long PDFs, and that sucker will handle it like a champ. But most AI models take files now, too, even if they don’t have that same context window.

The other part I enjoyed was the UX. Anthropic did a great job making their chat experience intuitive while standing out.

Bing Chat feels impossible to get to. Bard compresses you into a tiny window. ChatGPT feels like a website from the 90s.

Claude feels fluid and bright. It’s modern and easy to grasp.

So kudos to the team at Anthropic. It’s not a monumental shift to the world of Transformer models, but it’s a big step for them.

Quora’s Poe app will see a drop in users now that people can use Claude without their interface. That’s the most immediate impact after this news dropped.

Now we wait to see who makes the next move.

Quick Nuggets

5️⃣ 5 things businesses need to consider before adopting generative AI

📝 GPT-based tools are for more than just bland text generation now

🪖 AI has launched a new global arms race, according to this TED Talk by Alexander Wang

📄 The No Moat Memo is real, but Google’s higher-ups disagree with the position

🎙️ Ezra Klein interviewed DeepMind CEO Demis Hassabis for his New York Times podcast

🎞️ A new piece in Defector argues that the real enemy is not AI but the execs who try to abuse users with it

🗞️ Newsrooms need to lean into AI to ensure their survival during this flood of misinformation

🇨🇳 Alibaba dropped a sweet AI image generator, and now the Chinese government is releasing clear rules on the legal limits of AI content

🔥 Fresh Products

• PhantomBuster - your leads in one place, supercharged by GPT (link)

• Wisely - AI assistant to help you shop wisely on Amazon (link)

• Comicify - transforming dull text into comic adventures (link)

• Swimm - smartest way to document code (link)

• Paul Graham Bot - navigate & search Paul Graham essays (link)

• Error Hound - hunt down & fix website errors w/ AI power (link)

• GREMI - one-click AI-powered SEO campaigns (link)

• Beamcast - AI assistant on any website (link)

Good Content, Office GTA

The stapler in Jello prank goes a lot differently if Dwight is a Grand Theft Auto character. I can promise you that.

❓ Test Your Knowledge

Put your AI knowledge to the test with our quick quiz!

Reply with your answer, and check out the next issue for the correct response!

Yesterday’s Answer: C) Teeny

_{- That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.}

Check Out Our Sunday Edition

You can dive into more AI news and topics with us every week by subscribing to our premium edition.

We’ve written about the following:

• Here Is The Work AI Will Replace Soon

• 8 Role Prompts that Still Work on ChatGPT

• 10 AI Tools You’ll Actually Use

If you’re not already subscribed, that’s okay. We’re offering a free 7-day trial so that you can read this one. That’s how excited I am to post it.

Claim the offer below to support our writing 🙂