Gorillas, hype cycles, and my MIT EmTech talk
Ignoring the herd and focusing on what matters
Prefer narration? Listen to the audio version of this post.
Yesterday, I was the opening act at MIT Tech Review’s EmTech conference, where I spoke about how day-to-day progress in AI makes it easy for us to miss some of the step changes we’ve seen in the past few years:
People go from questioning whether reinforcement learning could have applications outside gaming to seeing it adjust the magnetic core of nuclear reactors and unlock chat behaviors like ChatGPT;
Grainy images and shonky video generation to cinematic quality available on demand;
LLMs going from generating bad fan fiction to creating protein sequences and genome editors;
Systems struggling with object permanence to having the ability to perceive and interact with the world stably in rich detail.
Coincidentally, a few weeks ago, the Turkish 15-minute grocery delivery start-up Getir announced that it was pulling out of Europe and the US.
At first glance, it may not be obvious why these two events seem related. What has AI progress got to do with a loss-making delivery service?
Back in 2021, Air Street led a $6M round in Allcyte, an Austrian spinout company that extracted patient’s tumor cells and then used microscopy and computer vision to measure drug potency and tailor treatment selection. On the same day, Gorillas, an on-demand grocery delivery start-up, raised $290M and surpassed a $1B valuation.
At the time, the contrast between cancer and convenient food delivery clearly upset me on some level:
Ten weeks later, Allcyte was acquired by Exscientia, before the company went public on the Nasdaq in the largest European biotech IPO ever. Meanwhile, following its acquisition by Getir, Gorillas no longer operates outside Turkey.
On the one hand, the delivery start-up boom and bust doesn’t really matter. It didn’t have systemic repercussions. But at the same time, it incinerated capital that could have been invested productively and squandered the technical talent of a large number of people.
It’s also not the first time this has happened. Money has been thrown at electric scooters, meal kit delivery, co-working spaces and any number of sectors in the illusory pursuit of economies of scale. Many of the characters that brought us the 15 minute delivery boom are now discovering a long-held passion for military drones or diving headfirst into the race for ever larger foundation models. As we’ve argued before, there’s a good chance that much of this money will end up being wasted on producing undifferentiated, economically unsound leviathans. Sound familiar?
What should we take away from this?
Firstly, VC is hard. It’s easy to point and laugh at what seem like obvious mistakes. But only a small number of firms generate great returns and there’s no rapid feedback on your decisions. Pass on one too many good opportunities and it may be difficult to stay in business. If all the smart people you know seem excited about a space, it’s counterintuitive to disagree. In a game of weak signals, it’s hard to ignore the strong ones.
That’s why it’s important to have a clear strategy from day one and stick to it. Even when it’s incredibly tempting not to. Back in 2015, years before I started Air Street, I made the case for investing in AI-first companies. When people ask me for my ‘thesis’, I still point back to that piece, because the fundamentals hold true. Namely, that AI is a powerful force-multiplier on technological progress, because everything around us is the result of intelligence. That means the next wave of category-leading companies will be AI-first.
Finally, ignore the herd. If an idea didn’t make sense before it became fashionable, unless there’s been a real shift in fundamentals, it probably still doesn’t. 15-minute delivery had in fact been tried before, most famously by Webvan, during the dotcom bubble. Nothing about the logistics, margins, or the unit economics had changed in the intervening years. Charging people less than it costs to fulfil their order is just a bad business model, no matter how much cachet it suddenly obtains.
Prestige can serve as a reverse indicator. Often complex and, frankly, unsexy work can go on to produce the strongest businesses.
PolyAI, which recently hit a $500M valuation, started life in Nikola Mrkšić’s PhD thesis, an incredibly detailed dive into how models understand language. When I first met Nikola back in 2017, he ran through a deck walking through in-depth research on NLP and reinforcement learning dialogue work. Similarly, Profluent, which recently released the world’s first open source AI gene editor, was a biorXiv preprint just a few years ago.
Meanwhile, V7, which has gained traction with leading global enterprises, initially targeted all of its energies at data annotation - a tedious but critical part of any AI workflow and has now expanded into AI-first automation with V7 Go.
No one could accuse Hedera Dx of ‘herding’ given that liquid biopsies aren’t yet used in Europe, despite their huge potential to improve the accuracy of cancer care and treatment, while reducing pain for patients. Whereas, when Dimitrios Kottas left Apple’s Special Projects Group in the Bay Area to return to Greece to start Lambda Automata, many of his friends probably thought him mad.
This is a trend that’s been repeated throughout the history of venture and the technology sector. It’s best to be non-consensus at first, believing that you will one day be voted consensus by the market.
You could easily add the likes of Anduril to the list, with Katherine Boyle investing when defense AI was more likely to attract protests than talent. Or back when I met Alex Kendall, when he was starting Wayve and looking for investment, he was breaking with the approach set by Waymo and Cruise by throwing out rules-based systems, detailed maps, and extensive sensor suites.
In short, pursuing real problems that excite you isn’t just more fulfilling, it’s often better business too. Sadly, having a good idea or being a gifted researcher is only a small part of the story - execution is key.
We’re creating the playbook for AI-first founders, whether it’s through our writing or via events such as the forthcoming Research and Applied AI Summit, which brings together voices from Google DeepMind, Wayve, Synthesia, Vercel, and others. We will discuss all of these topics there and across our freely available writing over the coming months.