It’s a WONDERful life with ChatGPT
It’s well understood that there’s a class of search queries that we all used to perform in Google that we now perform in ChatGPT. It occurs to me lately, however, that not only is ChatGPT taking Google marketshare, it’s actually expanding the TAM of the already enormous search market.
There’s a class of search queries that I’d broadly categorize as “wonder” that are not utilitarian in any way. Particularly on mobile, “wonder queries” are low enough value that if the answer is not going to be easily accessible, it historically hasn’t been worth the time to prosecute them at all.
“Is there regulation that prevents buildings from keeping scaffolding up in NYC for long periods of time?” I wonder that every time I walk by the annoyingly persistent scaffolding on my block, but I can visualize the government website I’d have to navigate on my phone to satisfy that wonder, and repeatedly the trade doesn’t clear to Google it. I wonder…but I don’t NEED to know.
Lately, I’ve realized that my mental model around wonder has shifted, and all of the sudden nearly every wonder I have is actionable. I’m generating net-new queries that never would have happened before because the value exchange didn’t clear in a world where Google’s response asked even the slightest effort from me (i.e. clicking through, or parsing an article). At a business level, I find it fascinating that a market I perceived to be fully mature is starting to expand again, but at a user level I’m even more excited that my brain and behavior has been rewired to prosecute any and all wonder before it dissipates. Wonder is ephemeral…it’s so easy to let it go. But with ChatGPT at my side I’m actually wondering more and more, pulling inklings of thought into the conscious, because ChatGPT has created a reward function for my brain to wonder.
Read Full Post | Make a Comment ( None so far )Current Ideas and investment wishlist
Every so often I send an email to friends I’ve picked up along the way with what’s on my mind and where I’m looking to invest. Figured I’d post it here as well. As always, I’d be glad to collaborate and discuss around any of this stuff and welcome connections to your friends or portfolio companies building along these themes. jordan@pacecapital.com
1) Always on mic as a primitive for AI applications: This is not a new theme for me, but it’s really top of mind right now. We’ve made a hardware investment along this theme in a company called Friend, who aspires to deliver social and emotional value against the context captured by an always listening wearable device, but I’m on the look out for pure software applications that leverage both passively captured audio, as well as passively captured visual information by virtue of screenshots, screen recording, or even cameras. Early instantiations of an always listening UX in products like Amazon Echo and even Siri had to be gated by a “wake word UX” because the value that Amazon and Apple could deliver against a string of spoken natural language did not outweigh the perceived loss of privacy that such experiences requested. That’s changed with the advent of large language models. Now that the value deliverable against a string of natural language has grown by a few orders of magnitude, we are seeing the trade of privacy for value clear across a number of new use cases and verticals. The presence of an always listening meeting bot in a zoom has gone from bizarre in 2019 to completely normalized in 2023. I, and my physician, are now fine with my doctor appointment being recorded because it enhances the care that I receive. I’ve seen pure social applications utilizing the microphone + AI to capture and summarize my lived experience as an input into social posts for my friends. I think this primitive is new and important and will present in many more places. If you are your friends are challenging the norms around privacy by passively capturing boat loads of context as an input into AI powered applications and services, I’d love to jam with you.
2) Expanding token limits within embedding models: I think of embedding models, in some sense, as a data compression method, where we are able to condense a diverse set of unstructured data into a discrete, legible, and standardized output which preserves relational properties key to personalization and recommendation in an AI first world. Currently we, as users, can at best be represented in vector space by a constellation of embeddings as opposed to one singular embedding because we are constrained by how many tokens can be passed through an embedding model in a single shot. I think the max token volume you can pass into embedding models is currently around 9K in a single input, but I’m very interested in people working to expand that number without losing precision or fidelity. Of course in LLMs, we’ve seen the impact of growing context windows to hundreds of thousands or even millions of tokens as an input to inference, but embedding models are lagging behind pretty materially. This feels important to me and I’d love to jam with people on methods and approaches that could blow through this limit.
3) A calculated world: I read recently that data centers represented 4.4% of US power consumption in 2023, and that in the next two and a half years that number is projected to approach 13.2%. This was totally insane to me. I have been paying attention to how capex is flowing into this type of infrastructure, of course, but for some reason the power consumption anticipated on such a near term time horizon really changed my lens and perspective on what this buildout means. I’d read that OpenAI or Facebook or whoever are plowing all this money in, and sort of mentally chocked that up to hyperscalers pulling forward investment that would set them up for the next decade or whatever of demand for AI products and services, but to see that in 2.5 years the pure volume of calculations being executed in the world will dwarf that of the last 3-4 years combined makes me think more philosophically about the decline of ambiguity or uncertainty in society, and other similarly fundamental changes that will sneak up on us way faster than I appreciated. I see a ton of obvious first order implications of this type of growth, none of which are lost on private equity firms allocating around this space (energy production, data center build out, supply chain, blah blah), but I’m really trying to grasp at the second order stuff and am at the beginning of thinking in this way if you’d like to join me.
4) AI-native Jira: It’s pretty clear to me that work orchestration ticketing systems like Jira, Trello, Shortcut, etc…were not built to contemplate AI labor coexisting alongside human software engineers and designers. It’s also pretty clear to me that the an increasing proportion of the tickets flowing through these systems are being completed with assistance from AI, and in an increasing number of cases purely through agentic code generation. Brian Armstrong recently tweeted:
There is a ticketing system to be built that considers AI agents working alongside human engineering and design teams as a first class citizen. I’ll take it a step further and posit that one day soon people external to the org will be writing tickets alongside internal teams and that software development and CX will merge and be orchestrated in a single environment. As soon as the line between support tickets and feature requests blurs (i.e. when a company can cheaply and ~programmatically fulfill support tickets that require software development to satisfy the request), companies will be able to establish a much more direct line between end customers and product development and the ticketing system that enables this shift will be extremely valuable. I’d also consider incubating something in this space with the right founders.
Bonus:We’re hiring an early career investor to join our 4 person investment team at Pace. Looking for someone who is high context in our ecosystem, spiky, creative, ambitious, and interested in developing into a career investor. Ideal person has less than 5 years of experience, but has already made their way fully into startupland, either through personal curiosity, previous investment role, operating role, etc…(i.e. 2 years at Bain Consulting and curious about VC does not fit the bill).
Read Full Post | Make a Comment ( None so far )
Jordan Cooper's Blog: startups, venture capital, etc... 