Sheila Gulati, Managing Director
Some days are bigger than others, and this is one of those days! GPT-4 is here, and we are thrilled to welcome it to the world!
During my lifetime, I have been witness to technology creating incredible new paradigms. I remember before the iPhone, how unimaginable it would have been to think of getting a ride through my phone to the airport, without having to speak to a person. Now I don’t own a car!
We at Tola were so deeply inspired by the advent of cloud computing that we founded a venture firm to invest in that transformation. Today we are even more bullish on the depth and speed at which Large Language Models (LLMs) and AI will impact our lives, ultimately for good. The cloud was the catalyst for incredible value creation over the past decade, and AI will be even bigger. The incredible innovation we see today with GPT-4 was enabled by both the transformer revolution and the power and scale of the cloud, and the training opportunity afforded by massive compute resources.
The launch of GPT-4 is another major milestone in the rapid evolution of AI. While there is much work to be done to ensure that widespread AI adoption is safe, secure and reliable, we believe that today is a sea change moment that will lead to further profound shifts. AI hype exists for a good reason – we believe that AI will truly transform how the world works!
What we’re excited about in GPT-4
- Increased Intelligence: GPT-4 is much more intelligent than prior versions. Akin to human understanding, where one gains experience with inputs over the course of a life, GPT-4 has been trained on massively more data and parameters, which is enabled by the deep partnership with Microsoft Azure.
- Multimodal Comprehension: Until now, GPT has used a language model – text goes in, text comes out, and the system learns from text. GPT-4 uses a multimodal language model, meaning that, in addition to text, it has the ability to understand and interpret image inputs, then output that understanding in text. We imagine models will continue on the multimodal path.
- Improved Reasoning: With GPT-4’s enhanced accuracy and reasoning capabilities, it is possible to provide enterprises with more capable and accurate AI right out of the box. Of course, there are still some limitations, and end users will be responsible for fine-tuning the model for their specific use case.
- Memory / Higher Token Limits: You are now able to feed multiple data points into the model in a single request. This keep state makes it possible to remember and maintain information about current state or context over time, as a series of instructions are executed, or tasks are performed. In other words, the system will remember when I ask for a more casual writing style or that I love exclamation points!
In the VC world, we talk about 10x in financial terms, but, at Tola, we think of 10x in human potential. What would it look like if we didn’t have to do menial tasks anymore? What if we didn’t have to search for the right words, but had solid starting points to jump from? If we don’t need to spend time churning on every routine task, can we increase output by 10x? (Could I realize my dream of logging into my inbox with draft replies to all messages, waiting for me?) We see AI as a way to supercharge our time!
AI is also a way for workers to efficiently get what they need to realize the best outcomes for themselves, for customers and for their business. Consider these applications:
- Finance: AI can drive the ability to automate the understanding of all documents used to create financial statements. This would improve process efficiency by removing the manual review process and reduce risk exposure to human error.
- Software development: Faster testing coupled with more frequent and more efficient code reviews, supported by AI, mean applications and systems are more secure, more private, and ultimately benefit consumers.
- Sales: Sales teams have exactly the right pitch at exactly the right time to deliver maximum long-term success, as enabled by a system that has learned what pitch works best and provides salespeople with the presentations and email content that they need.
- Marketing: Mass personalization is now possible with AI that learns from the past and predicts probable outcomes. This enables real-time marketing campaigns targeted at micro-audiences. Every experience a customer has with a brand can be personalized to them.
AI has the ability to help transform humanity – to make our lives and our jobs better – but we must proceed with caution and consider the known – and many unknown – challenges along this journey. (More on this at another time…)
There is valid concern that AI will replace people, and while we predict that there will be displacement, we believe that quality of life will improve for a great number of people. In addition to working more efficiently and on less menial tasks, new job descriptions will be created, and a fresh crop of positions will be realized. The ideal scenario is that we build solutions where humans complete less rote work and can instead spend their time on more interesting and meaningful undertakings. There will be much change, so this real disruption must be carefully managed as it affects people’s livelihoods, but the ability to offer more interesting and purposeful roles creates huge opportunities for us to make life better for all.
We see AI as a life before / life after moment. And we are hopeful, because this is an opportunity to take what we have learned from other technology paradigm shifts – the good, the bad, and the ugly – and use these findings to ensure we are thoughtful about how we leverage AI for the good of humans, and that we work towards building the society that we want.
AI is not just for the tech mega-caps who stand to benefit because they are building these new LLM systems. There is a huge opportunity for all companies – from start-ups to enterprises – to build AI capabilities and integrate AI features into their products, benefiting customers, workers, and the bottom line.
The pace of enterprise adoption will be gated by the scaffolding needed to make AI work for businesses. The level of innovation that we are seeing in the market is truly inspiring, and we look forward to more value being delivered upstack, by software companies, as they leverage the awesome powers of GPT-4 and other LLMs.
We will be talking more about AI in the coming months – data governance, responsible AI, 10Xing of specific roles, accuracy, and more – but today, we’re celebrating the beginning of a new and exciting era!
Happy Pi Day; what an exciting day!