AI Frontiers: Helena Merk (OpenAI DevDay)

Thank you. So excited to be here. Before I kick things off and dive into climate and AI, could talk about this for hours, I want to do a quick pulse check for the room to understand people's knowledge of climate and how it overlaps with AI. If you don't mind raising your hand if you feel you have a sense for the types of problems we need to overcome in climate to avoid a catastrophe, even just a few.

Okay. Awesome. Keep your hand up if you have a good sense for how software can accelerate the climate transition. Okay. What about how AI specifically? Awesome. I hope by the end of this talk all of you can have your hands raised and have the answers to these questions. Climate change is one of the biggest problems facing the world right now, and as of last year, it also became one of the biggest economic opportunities. Just to put context around these numbers, last year around $4. 5 billion was invested in generative AI startups from VCs. About 100 times that amount was dedicated from the government into climate. That's $400 billion through one act alone, the Inflation Reduction Act. This spurred a wave of policies around the world where various countries are fighting to have the best policies to spur innovation around climate. If there was ever a time to work in climate, it's now, and these are six of the big categories we need to catalyze. As you can tell, these are all physical, real world problems.

Climate change is a physical, real world problem, and software can help us move faster, and we need to move faster. We're pretty much behind on every single one of those goals. In the next few slides, I'm going to walk through some really exciting applications of more traditional AI and ML and then I'll get to what you're super excited about, which is where do language models and things like ChatGPT and GPT-4 fall into line.

The first thing I'm going to talk about is soil. Turns out dirt is a really good place to store carbon dioxide, and to prevent a global catastrophe, we need to pull CO2 out of the air, store it in the ground. If you farm in a different way than we do today, as in going back to more traditional ways, with regenerative agriculture, we're able to pull CO2 into the soil. However, that takes longer and costs more time. If you're looking at farmers, their margins are thin, no one's going to do that unless you're incentivizing them. Currently, we incentivize regenerative agriculture, but testing whether or not they're following best practices is really, really time-consuming. People literally go out to fields, collect samples of dirt, bring it back to a lab, and do this many times over in order to certify a farm to be regenerative. What if instead, we could stream data all the time from a series of sensors and satellite imagery? A startup perennial is doing this today.

Something else we need in the climate transition is more renewable power. We've all heard of wind energy but that doesn't mean innovation has to stop there. Windlift is building drones that fly in this figure-eight shape and are able to capture a lot of energy in a far more modular design than a typical windmill might. They tether this device to the ground and they're able to relocate this to adjust for wind patterns or bird patterns.

Speaking of wind, there's another way to optimize a system. If you don't know too much about wind, I'll give you a 101 real quick. It's a chaotic, turbulent flow of energy. Wind turbines today don't really react to that, they're just there spinning. If we could in real time adapt the positioning of the blades, we could capture more energy from the turbulence. A startup called Windscape is doing this today by using neural networks. This means farms are able to have a higher ROI and that means cheaper renewable energy.

Once we have more clean energy, we need to electrify all of our homes. Ideally, we could just build a bunch of new houses, and they would all be perfect and not leak any energy and be fully electrified from day one, but that's not the reality we're in. We need to retrofit old houses. The bulk of the homes we have are outdated, and we need to add more insulation, we need to upgrade the stovetops. Doing that today is filled with soft costs. You're sending contractors to homes, they're literally filling up the walls, figuring out where are things leaking, they're using sensors, and they do that many times because customers want various quotes. A company called Kestrix is reducing a huge amount of this cost by creating models around where heat is leaking.

The final example I'll walk through is this company called Chemix. They're able to run simulations about various experiments they should be running. This means they're able to iterate on battery design far faster. If we apply this more innovative approach to R& D across the board, we could have way more breakthroughs in research. Now that I breezed through this like a 1,000 foot view, I'm going to get to how do LLMs fit into all of this. I personally spent all of 2022 diving into climate. I spoke with hundreds of founders and engineers and brilliant people across the board, trying to understand what was preventing them from scaling their climate solutions. What frustrated me so deeply was that every single conversation ended up in talking about permitting or grant writing or reading corporate disclosures, and everything was just filled with paperwork and money. Everyone needs more funding. For the most part, it blew my mind that everybody was stuck in these processes that felt very outdated. Turns out, LLMs help with paperwork-heavy processes. At Streamline, our mission is to accelerate the climate transition. We help companies unlock access to critical funding from the government. As I mentioned at the beginning of this talk, last year, $400 billion was unlocked. However, it's stuck behind this wall of PDFs. One thing I'll add is companies that are working in these hard technologies are dependent on this grant funding. This is not money that VCs are giving to companies because it's very risky. They've always depended on this grant money. Now it's a bigger problem than ever.

Just to paint this a little bit deeper, if you are a climate company today or any other company seeking grant funding, you're pretty much running into hurdles in every step of the process. It starts before you even start applying. As you start to look for a grant opportunity, you're looking at federal opportunities, state opportunities, every single different local government. Hours later, you find one that you think you qualify for. Then you get to page 37 of a 150 page PDF, and you don't qualify after all. You repeat this process, and eventually you find an opportunity you're a good fit for. You spend 100 or 200 hours polishing everything, and you submit it, and you get disqualified because you had an empty page at the end of the application, and that's not illegible. That's the number one reason, tied with probably applying for things that you were never meant to apply for, because it wasn't a good fit for your technology. Eventually, you win a grant, hurrah, but then you realize you have to report on how you're spending every single dollar to the T, and if you mess that up, you don't get your money back. We looked at this problem, and we realized language models are actually really, really beautifully designed to help solve all of these problems. At Streamline, we're building grant management software to help companies with discovery, writing, and reporting to save climate tech companies up to 80% of their time in seeking this government funding. We've been working with ChatGPT and the whole GPT family since day one, and it's been amazing in helping us accelerate all of these workflows.

I'm Helena. I'm the CEO and co-founder of Streamline. Feel free to reach out. I'll be here all day. Thank you so much.