😺 Why AI SKILL matters more than AI model

PLUS: The world's most famous prompt list...

Welcome, humans.

This is cool: Did you know vision models (AI that can “see” images) actually process the images in multiple stages? They start from high-level pattern recognition, then break it down to detailed feature analysis. Example:

This is called hierarchical visual processing—fancy talk for how we humans go from broad categorization (“it's a hand”) to detailed feature analysis (“wait, there are actually 6 fingers!”).

So in the above example, when Claude first looks at the hand emoji, it activates neural patterns for “hand” which strongly correlate with 5 fingers.

But when prompted to “look closer,” the model examines specific regions and counts the distinct finger-like shapes, overriding its initial assumption.

Hey, at least it didn't try to point its nightmarish sixth finger at anyone else for its mistake!

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

We dive into why AI skill may matter more than the model you use.
Meta hired Cruise CEO to build home robots.
xAI pursued a $10B raise amid a Dell server deal.
DeepSeek’s founder to meet President Xi.

Advertise in The Neuron here.

The REAL AI gap isn’t between models—it’s between users…

Between o1 Pro and o3-mini, DeepSeek R1, and Gemini 2.0 Flash Thinking, there’s been a lot of new AI tools launched in the last few months.

Sure, some perform better here or there on certain benchmarks, but generally speaking, there’s been a leveling of the playing field between the top providers. And with Claude’s new model coming soon, we think that’s likely to remain the case.

That’s why we think it's worth double-clicking on this piece from The Algorithmic Bridge that argues the performance gap between AI experts and novices is growing faster than the gap between AI models themselves.

Here’s the key points:

The performance gap between good and bad AI users is growing with each new model.
Basic prompting gets basic results, even from advanced models.
But expert prompting can squeeze 2-10x more value from the same AI.
Your ability to ask good questions will soon matter more than the model you use.

We found this part REALLY relatable—these days, most of us hit an “intelligence ceiling” where we can't even think of hard enough questions for new AI models to test.

Think of it like this: Twenty years ago, everyone got access to Google. But the people who really learned how to search, filter, and verify information had a massive advantage (we still think this is true, TBH. That’s why using Perplexity, ChatGPT search, and Deep Research are huge resources—IF you know how to verify their results).

Now, the same thing is happening with AI—except this time, the gap between experts and novices is even bigger.

The good news? There's still time to get ahead of the curve. Most people are still using AI like a basic calculator when it’s actually a supercomputer. Don't be one of them.

Now here's the key insight: Ethan Mollick argues that while everyone's waiting for artificial general intelligence (AI as smart as humans in all domains), narrow AI is quietly revolutionizing specific fields.

For professionals, this means the opportunity isn't in becoming an “AI generalist”—it's in mastering how AI can enhance your specific domain expertise. The most successful AI users aren't trying to automate everything; they're becoming dramatically better at specific, high-value tasks.

What this means for you:

Learn Prompt Engineering Now:
1. Start with basic templates and frameworks.
  1. Here’s some templates for Claude and tips for ChatGPT.
  2. We’ve got an intro to ChatGPT course that we’ll update soon.
2. Practice giving your AI clear context and instructions.
3. Learn what techniques work best for each model, and what model is best for your specific needs.
  1. For instance, GPT 4o and Claude Sonnet still need chain-of-thought and step-by-step prompting, where o1 and other reasoning models don’t.
Focus on Your Domain:
1. Build prompting skills specific to your field.
2. Consider all of your repeated tasks. Which ones could AI handle?
3. Combine AI outputs with your professional judgment.
Track What Works:
1. Keep a “prompt library” of your best AI interactions—you can use ChatGPT Projects or Claude Projects to do this.
2. Note which approaches work best for different tasks.
3. Share techniques with colleagues—even if you’re using AI in secret.

FROM OUR PARTNERS

Enterprise AI without the overhead

Reduce time to value with one platform for building no-code AI agents integrated with all your company's apps.

Sana AI allows you to:

Deploy custom AI agents for any team or role without writing code.
Consolidate AI software applications and unify the UX for AI across your organization.
Integrate AI seamlessly into your existing tech stack.

Check it out here.

Prompt Tip of the Day

In case you missed it the last time we shared this, Awesome-ChatGPT-Prompts is a curated library of prompts to help enhance your AI conversations.

Here's what makes it special: each prompt is battle-tested by the community and designed to put AI in expert mode instantly. Plus, it works across models—from ChatGPT to Claude to Gemini.

Our advice? Just skim through it and Ctrl/Cmd F search for the category you’re looking for (software, business, marketing, etc).

Treats To Try.

Here are 7 Chinese AI to watch besides DeepSeek (Heads up, you might need to use Google translate for some of these sites if you don’t speak Chinese):
1. Stepfun turns your text, images, and audio into GPT-4 level responses ($2B+).
2. ModelBest runs powerful AI right on your phone, no internet needed (raised tens of millions).
3. Zhipu creates chatbots and Sora-like video generation tools ($2B+).
4. Infinigence AI makes different computer chips work together to run AI models faster (raised $140M).
5. Baichuan builds AI doctors for the Chinese healthcare market ($2B+).
6. Moonshot, who made Kimi, a DeepSeek competitor.
7. MiniMax, who has the video app Hailuo and audio app Talkie, which is a really popular competitor to Character.AI.
Ariadna chats with you about your career goals and connects you with perfect-fit collaborators and opportunities.
Crustdata delivers you real-time company and people data so you can build accurate contact lists that automatically update when someone changes jobs.
Roamr helps you earn money by staying with friends instead of hotels during work trips, splitting your company's savings between you and your host.
Letterpal helps you create engaging newsletters by finding trending news in your industry.
LLM Stats, Live Bench, Artificial Analysis, Vellum, and the Big Benchmarks collection are all tools to help track which AI models are best.

See our top 51 AI Tools for Business here!

Around the Horn.

Apple plans to launch AI in China as soon as May 2025, and it will use a combination of its own AI alongside Alibaba to censor the output for Chinese audiences.
Meta formed a new team inside Reality Labs led by the CEO of self-driving car company Cruise to work on humanoid chore-bots.
Elon’s AI company xAI wants to raise $10B at a $75B valuation as the company decides whether or not to buy $5B in servers from Dell.
DeepSeek’s founder Liang Wenfeng got a bullish signal from President Xi Jinping, as the two are set to meet up at a summit this week.

FROM OUR PARTNERS

Attention is all you need.

We (Noah & Grant) use ChatGPT + Claude every week, but we ALSO use a killer product called Attention.

It’s a must-have for all sales teams, so if you sell literally anything, drop what you’re doing and go book a demo with Attention ASAP.

Under the Hood

Firecrawl, an AI startup that helps turns any website into organized data, went viral after it posted a job offering for an autonomous agent that can research trending AI models and build sample apps with them with a $15K salary.
Prototype helps you set up Django web applications instantly with Docker and OpenAI, by converting a text description into a working codebase (learn more here).
Nia helps you understand and navigate your codebase faster through smart search and Slack-based answers.
Kokoro turns your text into natural-sounding speech in 8 languages using a lightweight 82M parameter model that you can run anywhere.