😺Our HONEST take on GPT-4.5

Welcome, humans.

We read a spicy headline yesterday that McDonalds will be using more AI in its 43K locations to speed up its service. That said, this is NOT what we were expecting…

All jokes aside, that video was apparently made with Kling, which is considered one of the best video AI for realistic, smooth movements.

As for what McDonalds is actually doing with AI? It’ll roll out internet-connected kitchen equipment, AI-enabled drive-throughs, and AI-powered management systems to improve customer and employee experiences. Ba da ba ba baaah, bots everywhere!

Here’s what you need to know about AI today:

• We share our read of GPT-4.5.
• Christie's first AI art auction earned $700K.
• Google co-founder Larry Page started an AI company.
• AI chipmaker AMD released some open-source models.

Our HONEST take on GPT-4.5…

ChatGPT’s newest AI model, GPT-4.5 came out about a week ago as a research preview, and the initial reviews were mixed. We decided not to cover it in depth because we thought it wouldn't be worth it if so few people could actually use it.

That’s changed. OpenAI just rolled out GPT-4.5 so anyone with a Plus subscription can use it up to ~50 times per week.

Here’s how we tried it out: First, we asked GPT-4.5 to write our entire newsletter for us, with no context, using only our style guide and GPT-Search.

To be fair, this is NOT how we write the newsletter; we write section by section, outlining and writing draft v1, then use AI to improve and edit our work as necessary.

So this was a task GPT-4.5 was destined to fail. But TBH, we're not mad at the result. Check it out:

Obviously we got today’s blurb idea from GPT-4.5!

Next, we compared GPT-4.5 against Claude 3.7, having them both help us write this main story via the same prompt.

Below are our two favorite paragraphs, where each AI model tells YOU when and what to use GPT-4.5 for.

Vote for which one you think is better; on Tuesday, we'll reveal which was which Model!

Who wrote better?

Now, we skipped any math and coding tests, because GPT-4.5 isn’t made for that.

But it is interesting to note that when humans directly compare outputs in blind tests, GPT-4.5 consistently ranks #1 in the LM Arena (or tied with with Grok-3) across multiple categories. That includes coding, math, and hard problems—things it wasn’t supposed to be good at.

Here's our HONEST TAKE: GPT-4.5 might not be a better model, but it’s certainly more fun to talk to. From that standpoint, this model could actually have a lot of staying power for die-hard GPT lovers. Put another way—the “vibes” are real.

Here’s an example where we asked it for its genuine take on the universe and everything in it—we’ll let the conversation speak for itself.

In this way, GPT-4.5 is actually a little bit more like Claude.

Amanda Askell, Anthropic’s resident philosopher and “Claude whisperer,” has deliberately shaped Claude’s conversational personality to prioritize traits like curiosity, openness, thoughtfulness, and nuance through a method known as “character training.”

People are already (and controversially) semi-addicted to chatting with AI. If GPT-4.5 makes it easier to keep chatting, some of the same magic that makes Claude users extremely brand loyal could rub off on casual GPT users.

And yes, if you're wondering, GPT-4.5 absolutely wrote this article. (Just kidding… maybe.)

FROM OUR PARTNERS

💼 Want to become an AI Consultant?

This stat is WILD: The AI consulting market is projected to explode from $6.9 billion to $54.7 billion by 2032. Want to capture this opportunity?

So how do you, an AI enthusiast, become a handsomely compensated AI consultant?

Well, The AI Consultancy Project from Innovating with AI has already helped 300+ students build successful consulting businesses.

Current students love:

• Tools and frameworks to find and serve clients.
• 6-month roadmap to building a 6-figure business.
• Methods to land your first client in as little as 3 days.

As a Neuron reader, you get early access to the next enrollment cycle!

Take advantage of the AI boom—request early access to The AI Consultancy Project ASAP!

Prompt Tip of the Day

Want to see what AI model your peers say is best suited for your specific-use case? Try this easy to use Prompt to Leaderboard tool on LM Arena.

Here’s how:

1. Visit lmarena.ai.
2. Click on the “Prompt-to-Leaderboard” tab at the top navigation menu.
3. Select the “P2L Explorer” sub-tab to view the interactive leaderboard.
4. Choose a main category of prompt you want to explore from the color wheel.
5. Take your pick from a series of prompt sub-categories.

For each sub-category, you’ll see detailed rankings of which AI was voted best for that task, along with example prompts of what was tested to get that result. Pretty awesome!

Treats To Try.

1. *Train your whole team on AI with ProfAI, a personalized, AI-powered coach. Upskill teams, train on personalized use cases, and grant AI certification. See a live demo on 3/18.
2. Agentforce now comes with an AgentExchange, a marketplace with hundreds of pre-built AI agent components to quickly automate your business.
3. You.com is an AI search engine like Perplexity that aggregates top AI models in one platform where you can switch between GPT-4.5 & Claude 3.7.
4. olmOCR transforms your PDFs into clean text while keeping tables, equations, and handwriting intact at a fraction of the cost of other solutions (more).
   1. Related: If you’re technical, Mistral has a new API for a similar thing.
5. OLMoE is a new open-source iOS app that can run AI2’s OLMo 2 and Tülu 3 AI models locally on your iPhone 15 Pro (so completely private).
   1. Related: AI2 also released Jamba 1.6, another family of high-quality open models you can run anywhere with good performance on long-context tasks.
6. PublicView helps you search through, analyze, and chat with public SEC filings.
7. Ataraxis, Valar Labs, and Manas AI use AI to improve cancer outcomes through treatment personalization, response prediction, and faster drug development respectively (all raised ~$20M).
   1. Related: If you’re in healthcare, check out Knowtex, which turns patient convos into automated clinical notes and billing codes, saving you hours of paperwork.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

• Christie’s first solely genAI art auction brought in $700K—a lot more than expected, even with protests.
• Google co-founder Larry Page has a new startup that’ll use AI to design and produce “highly optimized” products.
• AMD, the AI chipmaker rival to NVIDIA, launched its own family of small open models called Instella.
• A16z published a new report on the top 100 GenAI consumer apps by traffic and mobile downloads.
• FYI: We’ll be at HumanX Conference in Vegas next week as lots of big names in AI will be there—look for Grant in the Neuron Cat hat if you’re attending (and say hi)!

FROM OUR PARTNERS

This Smart Home Company Grew 200% Year-Over-Year…

With $10M+ in revenue and 127 Best Buy locations, RYSE is transforming window shades with patented retrofit tech—no costly replacements. Their public offering is live at $1.90/share.

Invest Today and Receive 25% in Bonus Shares

Intelligent Insights

• Watch this ~1 hour convo on artificial general intelligence (where AI can do everything a human can) and its many geopolitical implications.
• A new survey of AI researchers found that most are skeptical that simply scaling up current neural networks will lead to AGI.
• HuggingFace’s Thomas Wolf shared an insightful essay arguing that current AI systems are becoming “yes-men on servers” incapable of the creative, outside-the-box thinking needed for true scientific breakthroughs.
• OpenAI chairman Bret Taylor made a bold prediction that AI agents will become most brands’ main digital experience within 5-10 years.
• Here’s exactly why we probably need a way to create “verified video.”
• Ethan Mollick says ChatGPT’s Deep Research tool is largely seen as cheaper and faster to fact check than to have a human create in the first place.

A Cat's Commentary.