What’s in your toolbox?

Take a second to think back to May 30, 2022. What were you working on? What did your day look like? And did it occur to you that, in less than a year, you’d start using AI to ideate, strategize, and execute?

The world changes fast. There are plenty of people who will teach you how to prompt GPT-3 or GPT-4, but pretty soon we’ll be at GPT-10 and its successors. If you only learn the mechanics of one tool, you’ll have a limited toolbox. Right now, there are a couple of recognizable AI apps that people will tell you to master. You should definitely utilize them today, but our point is that we’re going to be marveling at new capabilities in another year’s time.

So personally, we’re challenging ourselves to pursue not what to think, but how to think. As we create something new for The Multidisciplined each week, we encourage process over conclusion, independent judgments, skill transferability, and adaptation.

So what are we looking at this week?

We’re checking out Pickaxe, an awesome no-code tool that builds you custom GPT-4 web apps. We’ll share our experience, use case ideas, and BTS insight from co-founder Nathaniel Mahowald. If you’re a team leader or cross-collaborator, you need to try out Pickaxe.

And as usual, we’ve got a list of cool jobs in gen AI at the bottom.

We’d love to know - what do you wish you could automate? Reply back ✍️

Go automate 1 thing today

Pickaxe is a no-code tool that builds you custom GPT-4 web apps. It’s free to start and the professional use cases are endless.

What it is:

• Opportunity to quality-control the work that you’re delegating

• Great UI that facilitates input from your teammates who don’t really get the ChatGPT interface

• Method for expediting your repetitive, discrete busy work if you know the workflow really well

What it’s not:

• A unique LLM (yet!)

• Specifically tuned for a function (like JasperAI for copy)

• A pre-packaged solution - you still need to be a good prompter

For example:

• You’re a Marketing Manager who relies on input from the User Researcher to craft 3 distinct versions of an email campaign, each targeting a different customer segment. To facilitate this process, you make an email copy template in Pickaxe, allowing the User Researcher to input traits and values specific to each of the three segments. You get 3 on-brand emails but skip the info-gathering, synthesis, and rewriting.

For example:

• You’re a Head of Operations who onboards a couple of people a week. The new hires read the onboarding packet and the company overview, but people always have questions. You can upload relevant PDFs to Pickaxe and create an FAQ app to embed on the company’s internal portal. New hires get questions answered on their own time, and you don’t get constant pings on Slack.

Honestly, no-code GPT-enabled web apps would have been wildly helpful to us in our previous professional lives. We would’ve saved hours (maybe days?) by automating tasks and redirecting that time to deep work.

Pickaxe is running a prompt engineering contest through June 11th with $1500 on the line. Try it out for yourself! Here’s a list of use cases we brainstormed:

• Customer service bot, trained on your protocols and products

• Ask the CEO anything - great for new hires

• Patient triage - a key in the current physician shortage

• Test prep tutor, trained on your history textbook

• Instagram caption writer, trained on your brand voice

• Short story writer for 2nd grade teachers, using the vocab of the week

• Beginner Spanish conversation partner

• Custom pitch email generator

• Sales coach that handles objections in real-time

• What would Bobby Axelrod say-bot

Behind-the-Scenes with Nathaniel, Pickaxe Co-Founder

Thanks to Nathaniel for sharing his story and his take on the horizons of generative AI 😀

Q: Can you tell us about the origin story of Pickaxe?

I was previously building a search engine for PE firms using GPT and saw the opportunity in building a GPT wrapper. The idea behind it is that in the gold rush, you want to be selling pickaxes. My co-founders and I wanted to build the tool that people use to create value. We initially put it together for fun and posted it on Reddit, and got thousands of people coming to use our cover letter generator Pickaxe.

Q: Where does the design inspiration come from? It’s refreshing and different from the hyper-futuristic themes we see on other AI websites.

My co-founder Ian is a designer, and he polished the look before sending it out, which was a good idea. Ian generates all the images that you see on the Pickaxe site using Stable Diffusion. The style is really inspired by Robinhood and crypto, a fantastical and fun place to be. And it really is - this technology is so exciting.

Q: Help us understand - what should people go to Pickaxe for?

I can tell you what people commonly use it for, and what people haven’t used it for but should.

Most of what we see in the community is that people automate things they don’t want to read or write. We’ve seen folks working on HR reporting and processing survey and text data that would take way too long to read. A lot of our enterprise work is about partnering with orgs to automate busy work.

What I wish would happen more organically is character creation. We’ve been able to create really good characters on Pickaxe and have a few on the site, like the Greek Philosopher one and the Hemingway bot. It would be fun to see people bringing back George Washington and other historical figures.

Q: What is a super memorable Pickaxe from the community?

Interesting story - some of our first users were in the Iraqi-Kurd community. The Kurdish language is illegal in Turkey, and discriminated against in Iraq and Iran. The Kurdish people don’t have a state for themselves. The language is hard to access, and GPT-3 is actually really good at Kurdish - better than Google Translate. We have hundreds of people using our Kurdish language bot to write essays, practice the language, and translate. It’s an incredible thing that came out of nowhere.

Q: When you started Pickaxe, was building on top of GPT-3 (versus other LLMs) a conscious decision?

We started with GPT-3 because Open AI models are just the highest quality. GPT-4 still is, currently. But what we’re expecting to happen in 4-5 months is that we’ll be able to self-host custom models that perform as well as GPT-3, maybe GPT-4.

Pretty recently, researchers at Stanford were able to build something called Alpaca, a model that yielded results similar to GPT-3.5. They only spent $600 to train it. They did it by using ChatGPT to generate the training data itself, and the realization is that a lot of people will be getting value out of self-hosted models, including Pickaxe.

Q: For us less technical people, can you shed some light on what exactly creates the gap between these different LLMs?

My understanding is that GPT-4 is just much larger than anything else, since OpenAI has spent hundreds of millions and many years doing it. It’s not that Google can’t do it for a skill reason - they just seem to be figuring out structural things. OpenAI had the right institutional knowledge this kind of model.

If you’re hyped about the generative AI industry specifically, here are some of the coolest roles we’ve seen this week:

• Director of People @ AssemblyAI

• Product Marketing Manager @ Soundful

• Customer Success Lead @ Uizard

• Marketing Director @ InvisibleAI

• Senior Manager, Revenue Ops @ SeamlessAI

• Talent Sourcer @ SnorkelAI

We want to feature an AI productivity hack from the community each week. If you have something to share with the class, hit us at [email protected] 😄

That’s all for this week. See you next Tuesday!

Lorel & Reily