"The Sequel Was Even Better"

Author

Matthew J. Sánchez
December 12, 2022

So good, so nice, had to read it twice

Some things are even better the second time around: Shrek 2, Godfather Part II, Rush Hour 2, Drake’s sophomore album ‘Take Care’. Welcome to the second issue of AI In The Middle. I am Matthew J. Sánchez and it’s my pleasure to serve a warm dose of Gen AI and invite everyone to this table.

Missed our first issue? Don’t worry, I got you. You can read the first dose of awesome right here. Much like broadn’s CEO & Co-Founder, Calin Drimbau, who wrote our first article, I’m energized by all things Gen AI. A little about me: I am currently a venture capital fellow interested in AI/ML, SpaceTech, Deep Tech, and AR/VR, and I am excited to join VC full-time in the near future.

What do we have for y’all today?

It seems like we can’t escape prompt engineering! It’s about to be the hottest job and sought-after skill in 2023 to improve the generative tech. We also talk about how you can win in the generative tech space right now. Founders, investors, builders, lend me your ears (eyes?).

All We Do Is Win, Win, Win, No Matter What 🏅

How can you win in Gen AI right now? James Currier from NFX recommends that we consider these 5 points:

I. Product speed

Launch features quickly and let the model learn over time! Don't spend too much time building the “perfect model” if it comes at the expense of other layers in the stack (i.e. your application, API/OS layer, etc.). Iterate, iterate, iterate.

II. Sales speed

Aggressive sales and marketing will be important, especially in 2023–2024, for embedding your product in customers, building network effects, and expanding into other categories.

III. Network effects

Figure out how to build network effects, specifically at the API/OS and App layers, which will surely help with defensibility, among other advantages.

IV. Embedding

If the bread-and-butter AI models trend towards commoditization, focus on how your applications and APIs/OS (the latter two layers of the generative tech stack) can help you retain customers by embedding them in their workflows or daily lives.

V. Founder-Investor Fit

Founders, find an investor who will sprint with you: you need an investor who’s willing to move quickly and take risks with you. For investors, assess whether the founders are nimble and creative enough to pivot and if the venture can be further commodified.

Prompt Engineering Is the New Black 😎

Wait, what is feature engineering? When you use feature engineering in machine learning, you are trying to see which factors you can choose, remove, tweak to develop a good predictive model. For many use cases, it has been a great approach to verify model responses. But thanks to LLM and its ability to work with natural language processing (NLP), language prompts would be the logical milestone. If you can understand prompts, and be able to build them with Gen-AI enabled models, you would be engaging in prompt engineering. This approach could help you build for new use cases, like building a thinking-out-loud model for your AI, or reasoning. Miguel Ballesteros, Principal Applied Scientist at AWS AI Labs, opines in this article that prompt engineering will be the new feature engineering for the AI/ML sector.

In the meantime, it’s been incredibly fun to ask ChatGPT things like “Can you please summarize this article like an eight-year-old?” and “Could you explain differential equations like a comedy standup.”

Loading: Prompt Engineers Wanted

We need a lot of high quality data to feed the Gen AI beast, coupled with computing chops. We would also need the model workflow to be optimized, as currently most models are under-trained and expensive. How do we do that? Enter startups like Humanloop. Humanloop is aiming to make a better GPT-3 model, to make prompt engineering reliable and optimized. We would spare you the technobabble, but if you are so inclined, they use techniques like reinforcement learning from human feedback or RLHF (phew, that’s a mouthful).

Molly Welch of Radical Ventures discusses these opportunities and gaps in the Gen AI space, with her commentary on prompt engineering (hey, there’s that phrase again!), fine-tuning, and evaluation, when it comes to LLMs.

I can see it now, a slew of fresh undergrads from the class of 2027 with minors and internships in Prompt Engineering. Matter of fact…let me get off here and enroll in a course myself 🤪

Gen AI Deals that make your eyes (and mouth) water 💰

What caught our eye? 👀

Twelve Labs cashed in on the Gen AI hype, scoring a whopping $12M in Seed extension, led by Radical Ventures with participation by a consortium of others like Index Ventures. Twelve Labs is developing technology that makes sense of the endless barrage of videos we watch daily across YouTube, TikTok, and more. They’re creating a “video understanding infrastructure” to enable search within videos to be incredibly easy and intelligent thanks to scene context comprehension.

We’re well on our way to jumping straight to the best part of 3 hour movies without pressing fast forward x8 or using chapter skips…and going too far, so you gotta rewind by 30 seconds 🙄

New developments to sound suave at the water cooler ⛲

Things to learn when you need a raise

Please subscribe to AI In The Middle to stay in the know. 

Don't forget to share the newsletter with your own network and give us a follow (@getbroadn).

Before you go

Despite launching first with AvatarAI.me, @levelsio shares lessons learned as his AI-generated Avatars platform gets outperformed by viral Lensa AI, which is raking in $1M+ in daily revenue 🤯

There’s an adage that goes, it doesn’t matter who did it first, it matters who did it right. Keep this in mind founders and investors. That's it for today!

— Matthew J. Sánchez (@matthewjsanchez)