I attended Celesta Capital’s TechSurge Summit on February 13, 2023 at the Computer History Museum. In this piece I will talk about interview with Nic Brathwaite Founder and Managing Partner of Celesta Capital as well as Sriram Viswanathan (Founding General Manager of Celesta and heavily involved in venture investments in India), and a panel discussion by John Hennessy (Chairman of Alphabet).
In a companion article I will talk about my interview with John Hennessy, Chairman of Alphabet (Google’s parent company) and Vint Cerf, also with Google, during the TechSurge Summit. In addition, I will also discuss some notes related to a semiconductor panel at the Summit that included Sanjay Mehrotra (CEO of Micron).
Celesta Capital is a Venture Capital (VC) investment company and they put on the Summit to explore the future and state of deep tech investing – ranging from investment trends in deep tech to industry trends increasing demand for new tech advancements, and why investments need to focus on technologies that enable mass adoption of emerging applications. I will try to give some insights related to this theme, especially regarding the growth of data and digital storage trends.
First from my interview with Nic Brathwaite and Sriram Viswanathan, I first asked them about his experience with Flex and what he sees as the future of hardware. Nic indicated that the growth of emerging technologies and applications in the deep tech space there is phenomenal growth ahead. For instance, 5G enables IoT and autonomous vehicles, which requires lots of sensors and data storage as well as communication. Each technology will have additional hardware requirements. On the manufacturing side you have the ODM side and hardware products with innovative OEM products.
I also asked about potential VC investments driven by new storage and memory technology such as NVMe-oF and CXL. Sriram answered and said that he came from Intel and was involved in some of these topics. He said that company investments change over time and there has been a paradigm shift so companies such as Apple are building their own M1 and M2 chips at TSMC (which they would never have anticipated back in 2005). There is a much higher need for tighter integration of hardware bound tightly to software, whether at the data center or at the edge. Unless you do this, you can’t deliver the power and performance needed for many applications.
The next stage is beyond the integration of core and software and includes integration of memory, storage and communications. The interconnect is the major bottleneck. They invested in innovium (sold to Marvel). Software at the edge and the cloud need tighter integration with security as well. You can fit distributed legers (blockchain) to avoid compromising infrastructure. He said that they are investing in another company in the Kubernetes space, Platform 9, that focuses on the data rather than the data transport. Enterprise databases will not be entirely in the cloud or multiple clouds or on-premise and this creates new investment opportunities.
Nic was briefly acting CEO of a company in the autonomous driving space. They are delivering their first peta-OPS machine that consumes 20 watts of power on a 5nm chip. The automotive industry is changing a lot, but if you look in the generative AI space, the current data center infrastructure is woefully inadequate. You need power and performance at a price point that makes sense. He said that Azure was 20% overcapacity, especially with generative AI. They have investments in companies that are working with unstructured data rather than databases and deal with discontinuities in hardware and software requirements.
There was discussion in one of the panels on using generative AI for particular applications. Nic said that in order to provide the performance and capabilities emerging that you need special purpose semiconductor devices because general purpose devices are powerful enough and tuned for those applications. For a systems company to create differentiation need the right combination of software and hardware. Differentiation must be done at the component level.
John Hennessey was in a kick off session interviewed by Kia Kokalitcheva, Axios. She asked him what he thought about Chat GBT. John said that he was impressed with the quality of the natural language capability and said that at least at a superficial level that it often gets things right but that it is always confident that it has the right answer. He talked a bit about the history of AI going back to John McCarthy and IBM’s Deep Blue. He said that Deep Blue used brute force computing but Alpha Go did not. He said that AI development is moving faster than Moore’s Law and the number of papers on this topic is accelerating.
He was asked to comment on when Chat GBT gets things wrong—he commented that there was at least one case where Chat GBT made up the references. He thought that it might be better to train such an AI on more useful data rather than a bunch of random websites. He suggested that e.g. Wikipedia would be a good source and that it will be several more years before such generative AI becomes something we can be proud of. Kia then asked him what he thought was the place of generative AI in applications. He thought that it could be used to do a good summary of a topic and perhaps help to make a beautiful PowerPoint presentation and it could potentially revolutionize legal briefs.
He also said that the current cost of inference is too high and that Chat GBT is too often busy. He thought that there were opportunities to build AI systems trained and focused on particular uses, which would lead to smaller models and they would be more practical. He thought we are 1-2 years away from useful products, particularly in business intelligence. He also said that the use of AI allows us to program with data rather than lots of lines of code. Google was hesitant to produce something like Chat GBT, they didn’t want the system to say wrong or toxic things. He said that the tech industry needs to be more careful to encourage a civil society and that many tools, such as the Internet, were not anticipated to be used to do evil things.
John said that AI can be an amplifier of human intelligence. It could be used to help teach kids in a classroom with customized instruction to match their rate and type of learning. He said that the chance of making a true general AI is much more likely than it was in the past. He also made comments on defensive technologies, blockchain, fighting climate change, the future of semiconductor technology in the US and medical innovations.
Celesta’s TechSurge Summit covered investment trends in deep technology and included insights on data growth and demand. John Hennessy, CEO of Alphabet, covered many topics, including how AI can be an amplifier of human intelligence.
Original post: https://www.forbes.com/sites/tomcoughlin/2023/02/27/ai-an-amplifier-of-human-intelligence/