😸 Prompt Claude like a pro

PLUS: Meet NEO, your new robot butler

Welcome, humans.

A new humanoid robot called NEO is trending right now:

Developed by 1X Technologies, and standing at a modest 5 and 1/2 feet tall, NEO can jog, climb stairs, lift 40+ pounds, understand language, and “navigate physical environments” (y’know, to “do your laundry”, not “hunt you down”…)

…All with battery life of 2-4 hours—the ideal work-life balance!

All jokes aside, NEO is awesome, and has been in development since 2014… check out this 13 min video to learn all about the project.

Oh, and if you’re interested… 1X plans to launch NEO as early as next year…

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

  • We break down the top 9 tips for prompting with Claude.

  • Luma Dream Machine introduced easier prompting for camera moves.

  • VW will begin the roll-out its in-car ChatGPT assistant in the U.S.

  • Microsoft and LinkedIn found 71% of companies want “AI expertise.”

Here’s everything you need to know about prompting Claude.

While Perplexity captured the top spot in yesterday’s poll about your favorite AI productivity apps, Claude secured a close second. 

There's good reasons for that: Claude is a powerful tool for tackling all kinds of complex tasks. But to actually get the most out of Claude, you need to know how to prompt it effectively. 

Luckily, Anthropic has tons of useful resources on its website to help craft prompts that will maximize Claude’s performance.

To begin, outline:

  1. Your success criteria (e.g., “less than 0.1% of outputs fail”).

  2. How you will evaluate success (e.g., have Claude write 100 test cases for the prompt to pass before considered “successful”).

Example test flow.

Next, use these 9 strategies, in this order, trying each one until your prompt succeeds:

  1. Generate a prompt: Anthropic's Console has a prompt generator tool to create initial prompt templates (here’s the Google Colab link with the code to run yourself). To use either, you’ll need an API key.

    1. P.S.—Claude also has a Prompt library. 

  2. Be clear and direct: Provide explicit instructions, using clear and concise language, and break complex tasks into steps.

  3. Use examples: Include 3-5 distinct examples to improve accuracy and consistency, especially for specific formats or styles.

  4. Let Claude think: Allow Claude to reason step-by-step for complex problems.

    1. The simple method: Literally ask Claude to break its answer down “step-by-step.”

    2. The complex method: Tell it to first “think” in <thinking> tags to outline its reasoning, then write the result in <answer> tags (example).

Example.

More advanced prompting techniques (for use with the API + coding):

  1. Apply XML tags: Employ tags like <instructions>, <example>, and <formatting> to separate prompt components (examples).

  2. Role prompting: For specialized tasks, use the system parameter to give Claude a specific role.

  3. Response Pre-filling: Direct Claude’s output by pre-filling your initial text.

  4. Chain Complex Prompts: Break intricate tasks into smaller, sequential prompts for better results via step-by-step reasoning.

  5. For 200k token prompts:

    1. Place long-form data at the top, and your prompt at the bottom.

    2. When sharing multiple documents, use XML tags like <document> with subtags like <content> or <source>.

    3. Ask Claude to quote relevant info from long documents before carrying out its task.

Heads up: Anthropic also offers interactive tutorials for prompting in Google Sheets and Github.

FROM OUR PARTNERS

If you want to use the right model for your next AI app, you need this eBook from AWS.

This eBook helps you evaluate all the different AI models out there, and unlock benefits across your entire company.

You’ll learn how to quickly analyze any model and decide if it fits your project, based on: 

  • Speed

  • Accuracy

  • Transparency 

  • Security

  • Cost

Plus, you’ll gain pro tips on defining tasks, input and output prompts, and key success metrics.

If you’re building AI apps for any use-case, you need this ebook.

Around the Horn.

  • Luma Dream Machine 1.6 now lets you select specific camera motions from a dropdown with the keyword “camera” (example here).

  • VW will begin roll-out of its in-car ChatGPT voice assistants in the U.S. on September 6th, starting with the 2024 ID.4 EVs and the 2025 Jetta models (ID.4’s get 3 years free apparently).

  • Microsoft and LinkedIn’s 2024 Annual Work Trend Index surveyed 31K execs from 30+ countries, and 71% said they’d rather hire applicants with “AI expertise” over those with “experience.”

Treats To Try.

  1. *Discover MIT's latest Generative AI course (designed by world-renowned professors) at this FREE webinar on September 17th. You’ll learn about the course content, hands-on projects, and enrollment benefits. Limited spots—reserve yours today!

  2. Spotter (creator funding platform) launched Spotter Studio to help YouTube creators brainstorm, research, and plan videos to improve their performance and increase views.

  3. AnswerGrid helps B2B companies find and qualify relevant sales leads using automated web research.

  4. Cohere improved its Command R model series (targeted at enterprises), with better performance in coding, math, reasoning, and latency. Try them here or via API.

  5. Spellar Pro enhances meeting productivity by providing real-time speech feedback, automated summaries, and summaries to improve speaking skills.

  6. Repaint simplifies website creation by combining design and development tools in one intuitive, canvas-based editor for professionals and creators.

  7. Jamboss turns any idea into customized songs across different genres and styles (good as a birthday gift for a coworker—also, is ex. #2 a legit proposal from the founder?! We asked if she said yes…)

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

What gives it away that this is AI?

One watch, everyone knows the rules:

What gives it away that this video is AI?

Login or Subscribe to participate in polls.

Here’s what #Neuron readers had to say about last week’s videos…

  • T.J: “Cars merging into one another, driving backwards, incomplete and disappearing vehicles”

  • K.S: “Too many to count! Between cars going backwards, driving straight into traffic (going the wrong way), and the giant crystals - this is a no-brainer.”

  • J.H: “What doesn't give it away???”

  • L.T: “The way everything jiggles lol - doesn’t seem natural.”

  • P.K: “Hands down it’s the rice ball never actually striking the drums, only the cymbals. You haven’t get those simple details if anyone is going to believe this is real.”

  • I.B: “The little guy playing the drums in the middle of the food.”

Where do you #Neuron?!

Submit where you #Neuron here for a chance to be featured in our newsletter next week!

Rob from Telluride, Colorado: Best views in the world!

Amy (and Harley) from Chicago, Illinois: All eyes on the Cat!

Köksal from Akyaka, Muğla, Turkey: A dose of Neuron at an amazing Gökova bay view from the Woods. Love from Turkey for all Neurons.

A Cat's Commentary.

That’s all for today, for more AI treats, check out our website.

The best way to support us is by checking out our sponsors—today’s are AWS and MIT.

See you cool cats on Twitter: @nonmayorpete & @noahedelman02

What'd you think of today's email?

Login or Subscribe to participate in polls.