Mind the Storage Gap


Co-authored with Mark Khalil

Deployment of renewables has been accelerating globally. In 2022, around 12% of electricity came from wind and solar, compared to less than 3% a decade ago. These resources offer access to low-cost clean energy – but only intermittently. Energy storage systems can capture excess renewable energy in times of abundance and discharge energy when sun and wind are scarce. Unfortunately, the development of storage assets has not kept pace with renewables, creating a massive storage gap. In this post, we will explore the storage gap and discuss three hypotheses on how it may get filled.  

At the end of last year, the global supply of wind and solar stood at nearly 2 TW, while standalone batteries reached just over 50 GW (0.05 TW). Pumped-hydro offered another 175 GW (0.175 TW) of storage, though because of its required scale and siting restrictions, traditional pumped-hydro is limited in where and how quickly it can expand.

One way to understand the importance of closing the storage gap is to look at the disparity between power demand and supply of renewables throughout the day – the so-called “duck curve” (look at the yellow line to see the duck):

During daylight hours, there is often an excess of renewable energy, but when the sun sets, non-renewable sources have to quickly ramp up generation to meet demand. The sudden fluctuation poses a risk of grid instability if non-renewables cannot adjust in time. And the glut of power in the “belly of the duck” is so acute that there are increasingly frequent periods of negative electricity prices – even while prices remain high at other times. Energy storage offers a temporal bridge between times of abundance and scarcity of renewable energy, smoothing supply and demand and facilitating gains from trade across time.

The severity of the storage gap has created an opportunity to build transformational startups that will fundamentally change the shape of energy generation and distribution. As we look to invest in the space, we are exploring three hypotheses: 

First, individual asset owners will aggregate their assets into networks instead of participating in storage markets themselves. Beyond storing energy for direct use, storage systems can create value by providing flexibility and ancillary services to the grid. However, most asset owners won’t have the expertise, resources, or risk tolerance to effectively take advantage of these capabilities on their own. Instead, they might opt to join a storage network where experts can virtually direct their assets and share the incremental value. Connecting storage assets could also unlock new services that no individual asset could offer, like access to different geographies depending on where the best opportunities are at a given time. Moreover, these virtualized networks can be agnostic with respect to the details of the underlying storage systems, enabling portfolios of batteries with different chemistries and non-stationary storage like EVs.

Second, storage has the potential to grow most quickly in the commercial and industrial (C&I) market as a result of the strong value proposition to C&I customers and fewer barriers to entry. Downtime in industrial processes can be extremely costly, making on-site energy storage an attractive backup for when the grid is down. And as C&I companies increasingly add behind-the-meter assets like solar and EV charging infrastructure, synergies between storage and those other applications make storage all the more valuable. Furthermore, given their scale, C&I customers are well positioned to invest in the bi-directional interconnects that enable storage assets to interact with the grid  – especially as compared to residential customers who have to amortize the upfront cost of an interconnect over a smaller amount of storage. C&I also has the advantage of generally being located where there is more space available to safely site batteries, which could pose a fire risk in other settings.

Finally, lithium-ion batteries will remain hard to compete with directly, suggesting the best storage startups will either leverage lithium-ion technology or counterposition with features that lithium-ion structurally cannot offer. The cost curve for lithium-ion looks similar to solar, dropping ~97% in the last 30 years and ~80% in just the last 10 years. As we continue to manufacture lithium-ion batteries at larger scales, our production processes will become even more efficient, further driving down costs. Because lithium-ion is already on such a steep learning curve, competing approaches cannot just be 10% cheaper or more efficient – they will have to be radically better along some other dimension. Such an approach might look like thermal storage for industrial plants with large heat needs or safer indoor-friendly batteries for urban areas with higher fire risks.

Addressing the storage gap represents an opportunity to supercharge the value of renewables and benefit both individual energy users and the grid itself. We are excited about innovative storage solutions that aggregate decentralized assets into networks, focus on C&I, and either leverage lithium-ion or offer unique features that lithium-ion cannot. If you’re working on building this future, we would love to hear from you.

The post Mind the Storage Gap appeared first on Union Square Ventures.

AI Reliability and Building Trusted Brands


tl;dr: For increasingly capable AIs to gain widespread adoption, they will have to earn our trust by reliably doing what we want in the ways we want. Four promising areas for ensuring reliability and corrigibility in AI products are: refining AI behavior, interpreting models, evaluating models, and building systemic guardrails to mitigate the harms of misbehaving AIs.

Recent progress in AI is extremely exciting and has the potential to change the world for the better. However, in order for AI products to gain deep traction, they will have to earn our trust in the domains that matter the most to us. Andy summarized many of our questions around trust and AI in a recent post. As he points out, different business models, data policies, security systems, and use-cases will demand different levels of trust from the end-user.

One area I’ve been thinking more deeply about is how we’ll come to trust that AI products will do what we want in the way we want. Given their incredible capabilities, we will increasingly cede responsibility to AIs. But as black boxes, there’s no guarantee that they’ll act exactly how we’d like. How can we get sufficiently comfortable with an AI’s “character” before empowering it to be an agent in the world?

As AIs are deployed into consumer products and enterprise tools, companies will risk losing their customers’ trust if they can’t avoid especially aberrant behavior. Parents won’t want their children exposed to inappropriate content or manipulated for nefarious purposes; businesses won’t want artificial agents spending resources inefficiently or undermining their brand. In order for AI to scale to its full potential, we will need better technologies for ensuring AI reliability and corrigibility. 

There are four areas that seem especially promising in this regard: refining AI behavior, interpreting models, evaluating models prior to deployment, and building systemic guardrails to mitigate the harms of misbehaving AIs. The first three of these are products that could be sold to AI companies looking to build or preserve trust, and the fourth is something that non-AI-native companies will need to deliver a great experience to their users in a world full of AI spamming, hacking, etc.

Behavioral Refinement

It’s useful to think of AIs as, in the extreme, akin to random number generators. In that analogy, if you’re hooking them to an important process, you’d want to constrain the set of possible outcomes as much as possible without sacrificing usefulness. Without any further specification, a sales bot trained on the set of language on the internet might find itself cursing out potential customers. Instead, you’d want to refine its vocabulary and tone to that of an effective sales agent (without making it worse at understanding language or otherwise selling products). 

One example of this approach is Reinforcement Learning from Human Feedback (RLHF), which incentivizes a model to shape its outputs based on judgements from human evaluators. RLHF drove ChatGPT’s viability as a consumer product by forming the model outputs into conversational responses and by helping avoid potentially controversial content. This is probably the most developed of the four areas, and we’re excited about further progress in behavioral refinement techniques like Anthropic’s Constitutional AI and enabling technologies like Scale AI’s RLHF labeling tools.

RLHF makes ChatGPT more chatbot-like and restricts the range of content it will discuss

Though techniques like RLHF can help make an AI behave more reliably, the goals and behaviors an AI develops will always be an imperfect approximation for how we want it to act: it only receives finite data and feedback, after all. Since we can’t perfectly specify how it should act in any given situation without losing generality of capabilities, it would be helpful if we could “read the AI’s mind” to understand how it represents our goals for it. 

You could imagine an AI doctor trained to output diagnoses based on symptoms. This AI doctor might get a diagnosis wrong like any ordinary doctor might, which is to say it might follow the prescripts of accepted medical reasoning to an incorrect conclusion. However, the AI doctor might also get a diagnosis wrong because it simply “hallucinates” a connection between symptoms and a disease that no doctor – or human – would suggest. In such a situation, it would be helpful to contextualize a diagnosis by “reading the AI’s mind” to verify if its reasoning resembles that of a competent doctor. More broadly, companies might be better empowered to preempt problematic AIs if they can check how the AIs are reasoning about the tasks they face. 

The field of interpretability is still nascent, but recent advances with smaller-scale models makes us excited about the potential for commercial applications. Our hunch is that the minimum viable interpretability product is much simpler than a complete “brain scan”. A tool as narrow as detecting whether an AI thinks it’s being truthful – without otherwise understanding its “brain” state – could help companies avoid manipulative behavior.

Model evaluations

It’s hard for companies – let alone consumers – to vet the models they’re relying on for themselves. With everything that could go wrong with giving a black box the power to act on your behalf, customers of AI products will most likely want some validation that the product won’t end up in a predictable failure mode. We’re already seeing this dynamic with the Alignment Research Center (ARC) red-teaming OpenAI’s models prior to public release and thereby lending OpenAI a stamp of approval they can reference at launch. Though ARC is doing interesting work, they operate in a fairly bespoke way that doesn’t obviously scale to a world with many orders of magnitude more AI products being deployed each year. We’re interested in what a more scalable version of model evaluations might look like.

GPT-4 Technical Report
Systemic resilience

The first three areas I mentioned are all tools that AI companies can use to build a trusted brand. In a world with misbehaving AI agents running around, however, other companies will have to prove resilient to AI manipulation in order to maintain their brands. In the same way that Cloudflare helped websites develop a reputation for reliability by making them robust to DDoS attacks, we think there will be new tools to promote resilience to malicious AI behavior. New AIs might be able to get around CAPTCHAs (perhaps by asking for help on TaskRabbit), develop better phishing and spoofing strategies, and so much more. The internet’s infrastructure is not yet prepared for a world of AI agents interacting with websites on humans’ behalf. 

Of course merely making AI products do what we want in the way we want isn’t sufficient for earning our trust. We hope to dive into the other issues around AI and trust in more detail in future blog posts.

The post AI Reliability and Building Trusted Brands appeared first on Union Square Ventures.

Joining USV – Matt


I’m thrilled to be starting at USV as an analyst on our investment team. I’m joining USV from AeroFarms, a startup building vertical farms around the world. During my time there, I wore a bunch of different hats, ranging from working with executives as chief of staff to developing our expansion strategy and forming strategic partnerships as a member of the corporate development team.

Before AeroFarms, I studied Philosophy and Computer Science. One of my favorite philosophers, Christine Korsgaard, once wrote that philosophy is just ordinary reasoning rendered persistent. Her sentiment succinctly captures what drew me to the discipline: we all occasionally wonder about life’s big questions, and philosophers are just the people who stubbornly dedicate themselves to searching for ever better answers. I have found that the startup world is defined by a similar ethos, attracting people who notice the same problems we all encounter but who feel compelled to dig deeper and to try to solve them. At AeroFarms, I loved being surrounded by people whose ordinary interest in issues in agriculture became a persistent passion for wielding science and engineering to make food production more sustainable and secure. Now, at USV, I’m looking forward to meeting founders who refuse to accept easy answers about why the world is the way it is and who run towards society’s most pressing problems.

In that spirit, I’m excited to be working on our climate fund with founders who are innovating in critical industries like energy and agriculture and building out new areas like carbon markets and carbon removal. I’m particularly interested in deep tech climate companies transforming the hardware our society runs on. Beyond climate, I will be working on our Thesis 3.0, investing in trusted networks that broaden access to knowledge, capital, and well-being. Given the wide variety of companies that fall into our theses, it’s no accident that the USV team is committed to a generalist approach. As I dive into this new role, I’m equally keen on reading white papers, touring innovative production facilities, and playing guinea pig for new consumer apps. 

If you’re building something that aligns with our theses, shoot me a note at [mm at usv dot com] or connect with me on Twitter @matthewjmandel!

The post Joining USV – Matt appeared first on Union Square Ventures.