Always be prompting

April 1, 2023 ndowd

Welcome to Startups Weekly, a nuanced take on this week’s startup news and trends by Senior Reporter and Equity co-host Natasha Mascarenhas. To get this in your inbox, subscribe here.

Reporter’s note: Before we get into this, thank you to the Startups Weekly readers who pointed out that last week’s link to my column was broken. If you want to know about the state of solo GPs — which oh boy, is it rocky — please read my TC+ story here: Are Solo GPs screwed? Ok, now into this week’s newsletter!

“AI whisperer” jobs are roles made for people who are savvy prompt engineers and able to manipulate ChatGPT or other generative AI tools to get what they want from it. As Bloomberg reported earlier this week, these jobs can land up to $335,000 in annual salary. Cha-ching.

It’s an impressive price tag for a job role that would’ve blown minds just one year ago. It’s also a reminder that generative AI’s boom will create a whole new set of questions about how we ask questions. By just spending a few minutes playing around on ChatGPT, I learned that you can elicit a diversity of AI-generated answers to questions just by changing a few select words and demanding bullet points versus three sentences.

Before we see a whole wave of AI prompt engineering consultancy practices for hire — just kidding, we already are — I figured it would be helpful to aggregate some of the top tips I’m hearing from the community on how to better interact with AI to get it to give you the answers you want. We already know it’s imperfect, it can’t be trusted with every single fact, and it sometimes is left speechless. What can we do with what we’ve got, though?

Many techies weighed in on Twitter — you can see the full thread here — but here are the top five tips:

1. Overdescribe. As one person put it, don’t be afraid to be extra when prompting artificial intelligence. There’s a difference between asking for a banana bread recipe and asking for a simple banana bread recipe that incorporates cardamom, takes less than one hour to bake and has a surprise ingredient. Or put differently, don’t just keep it high level, get specific about the tone and complexity you prefer in answers. Over explanation and adjective use was probably the most common tip I received when asking for advice on how to be a better prompter.

2. Ask for help. It’s not always easy to know how specific to get when you’re prompting the AI — so why not ask for help? One entrepreneur told me to ask ChatGPT how to prompt it based on what I want, so ask something like, what information do you need from me to create an itinerary for a San Francisco hiking trip? The AI will tell you what it needs — I wish humans were like this — and as a result, you’ll get a more tailored answer.

3. Give feedback. Let’s say I ask for: “a chatty but informative three-sentence explanation about the impact of generative AI on education, incorporating at least one metaphor.” ChatGPT had a pretty good response, but I didn’t love the response. Tell it that! Feedback is helpful to train the AI on the types of answers you like and takes a split second. Here’s how that process went for me:

Natasha: Give me a chatty but informative three-sentence explanation about the impact of generative AI on education, incorporating at least one metaphor.

ChatGPT: Generative AI has the potential to transform education by serving as a virtual Sherpa, leading students on personalized learning expeditions. Just as a Sherpa uses experience to guide climbers through treacherous terrain, AI can harness vast data to provide a customized path to knowledge. With the help of generative AI, education can become a more accessible and tailored journey, empowering students to reach new heights.

Natasha: I don’t like the virtual sherpa metaphor. Also, can you be more specific on which new heights students can reach?

ChatGPT: Generative AI can be thought of as a personal tutor, providing students with a tailor-made curriculum that meets their individual needs and interests. Like a skilled coach, AI can identify students’ strengths and weaknesses, and provide them with targeted guidance to help them excel. With the help of generative AI, students can reach new heights by unlocking their full potential and gaining a deeper understanding of complex topics.

4. Ask for nuance. ChatGPT only answers what it is asked, so why not ask for nuance? A founder said that they ask for pros and cons on every recommendation that they prompt ChatGPT to give.

5. Be a little cringe. Is it self-indulgent to tell AI to speak like me, specifically nudging it to be voicey, charismatic and informative? Maybe, but it helps. One of the biggest hacks I’ve learned is to throw my Twitter handle in when asking for a deep dive on a topic. ChatGPT responds better to “Make an argument for why Cauliflower is the best vegetable in the tone of @nmasc_” than “Make an argument for why Cauliflower is the best vegetable in the tone of Natasha Mascarenhas.” And because I know you’re curious about the former, I’ll leave you with ChatGPT’s argument on cauliflower supremacy:

Alright, buckle up folks, because I’m about to make a case for why cauliflower is THE BEST vegetable out there. First of all, let’s talk versatility. This baby can be roasted, grilled, mashed, pureed, riced, and even turned into buffalo wings. Name one other vegetable that can do all that, I’ll wait.

And don’t even get me started on the health benefits. Cauliflower is packed with vitamins and minerals like vitamin C, vitamin K, and potassium, not to mention it’s high in fiber and low in calories. Plus, it’s a great source of antioxidants, which help fight off disease and keep your body in tip-top shape.

But let’s be real, the best thing about cauliflower is how dang delicious it is. Whether you’re enjoying it as a creamy soup, a spicy stir-fry, or a crispy pizza crust, cauliflower has a flavor and texture that can’t be beat. And let’s not forget about its ability to take on different flavors and spices, making it the ultimate canvas for creative cooking.

So, there you have it folks. Cauliflower is the MVP of vegetables, and I challenge anyone to prove me wrong.

Those are some starter tips I’m incorporating but I’d still love to know what else I’m missing so please do throw your top tips in this thread. Also, shout out to Fundrise CMO Jon Carden for the headline inspiration!

In the rest of this newsletter we’re talking about Cerebral Valley and Substack. As always, you can follow me on Twitter or Instagram to continue the conversation. And if you feel like supporting me extra, subscribe to my personal (and free!) Substack.

Overheard in Cerebral Valley

I attended the Cerebral Valley AI conference on Wednesday, which brought together the top 200 minds in artificial intelligence for a one day deep dive. The event, put on by Newcomer and Volley, was immersive to say the least. I spoke to half a dozen YC founders who are building AI companies, bullish and surprisingly bearish VCs, and, briefly, Stability CEO and founder Emad Mostaque, about keeping it real.

Here are some great one liners:

“The pause for six months seems destructive and probably exorbitantly expensive to consider,” said Kleiner Perkins partner Bucky Moore in response to the recent open letter.
“Safety research is one of the areas were best known for … I don’t necessarily think that’s un-synonymous with a technology company, a company that is also thinking about how to take that research and turn it into something that is practical,” said Daniela Amodei, co-founder of Anthropic.
“You’re building for GPT-4, but you want to be building for GPT-10,” said Lisha Li, CEO of Rosebud AI.
“Of course we’re not in a bubble, this is bigger than 5G and self-driving customers,” Emad Mostaque, CEO and founder of Stability AI, said about AI’s hype. “When founders come to me, I say build good products and solve problems … most of the stuff is still surface level.”

And finally: If you’re still looking for more Stability, listen to my podcast interview with Credo AI founder Navrina Singh, who spoke about responsible AI, governance and fear as a not-so-great motivator.

Image Credits: Getty Images

Substack’s newsy week

Blogging and media platform Substack opened up a community fundraising round this week, seeking capital from its writers, readers both accredited and unaccredited. As of Friday morning, over 6,000 investors have pledged around $6.9 million in funding.

We spoke about the crowdfunding campaign on our podcast Equity this week, too, which was spicier than I expected. To me, if Substack successfully raises money from its community after disclosing its financials — the equity crowdfunding world will get a much needed reputation and public perception boost.

Here’s what to know, via my colleague Alex Wilhelm:

Ironically, Substack’s users could be giving professional money managers in the company a bit of a breather. The venture investors that once put a mountain of capital into Substack don’t have to invest more, the incoming dilution is effectively zero and the company gets more total funding to pursue its long-term goals. For Substack, this is a win.

We won’t see this sort of conundrum too often. Most startups could not manage this sort of crowdfund because their customers pay them, not the other way around. In contrast, Substack users make part or all of their living from the company, so they are more invested in it sticking around. This makes the Substack crowdfund unique.