Once again the world of AI is buzzing with the news of OpenAI’s latest release, DALL-E3. By improving the DALLE algorithm and integrating ChatGPT it will soon become easier than ever to create amazing AI art that is more nuanced and detailed than before. Whilst in itself a very impressive advance, it is perhaps a sneak preview of what multi-modal foundational models will be able to achieve in the near future.

Once again somewhat overshadowed by OpenAI’s announcement, Google Brain Research released a really cool paper on Generative Image Dynamics. We’ve seen GANs before that allow you to manipulate points and positions of an existing image but the addition of dynamics is truly extraordinary. You can try it for yourself on there website, but here is a little sneak peak:

There are now a plethora of generative AI tools out there, some more useful than others, and we can expect consolidation as the industry develops (This is a great recent paper exploring competition and regulatory dynamics in foundational AI models). However, one tool I have only just recently discovered is Perplexity. Whilst effectively it is “just” a wrapper for GPT-4 or Claude-2, it is focused on knowledge discovery and I have found it really useful for background research on some of the topics in today’s newsletter.

Signals From the Frontier

Covering Deep Tech signals that indicate the future may be different from the present

  • BoschMobix and Peaq Network revealed a p2p parking and charging scheme that permits autonomous transactions between vehicles and connected infrastructure at IAA Mobility. The system uses “moveIDs” that act as self-sovereign identities built on the blockchain. The project had backing from the German government and is part of the larger EU Gaia-X project to create a federated, secure, data infrastructure.
  • Microsoft has said it is ending support for the TLS (Transport Layer Security) versions 1.0 and 1.2 in a belated move that was recommended by the NSA back in 2021. TLS provides the security protocol for encrypting web pages and VPNs. The latest version is 1.3 which was released in summer of 2018 and helped reduce overall network latency as well as making the “handshakes” between computers more secure.
  • ASU and NASA are researching space based microelectronics manufacturing. In principle, manufacturing in microgravity should result in a faster and cheaper process because it eliminates the need for etching and allows for a thinner film layer. Testing is scheduled to take place on Zero-G flights in October.
  • Pulsar Fusion has started building the world’s largest fusion rocket engine. The rocket is a so called Direct Fusion Drive in which the charged particles are used to generate thrust directly. If it works, the technology promises faster space travel and lower fuel payloads.
  • Intel has revealed its glass substrate technology. To be clear this is not a move away from silicon chips, glass is after all formed from Silica, but instead this is a replacement of the previously organic substrates on which the Silicon dies sit. Compared to the previous (primarily plastic) technology, glass substrates are more rigid, enable more efficient power and higher speed signaling and, somewhat ironically given Intel’s reputation for developing power hungry heat sinks, higher temperature tolerance. Whilst probably not available until the end of this decade glass substrates are likely to be one of the keys to the next generation of high-performance processors.
  • China has launched its Comprehensive Research Facility for Fusion Technology (CRAFT). The facility, nicknamed “Kuafu” after a mythological giant that tried to capture the sun, is due to be completed by 2025 and is an important milestone in China’s plan to achieve large-power generation by 2035 with its CFETR project.
  • The cyborgs are coming! Nasa has teamed up with Apptronik to develop the capabilities of its Apollo humanoid robot to become an astronaut’s assistant or a remote controlled “avatar” on other worlds. Meanwhile Agility Robotics has announced its ramping up development of its RoboFab facility which can build more than 10,000 of its Digit robot per year.

Deeptech Reads

Articles covering emerging technologies in more depth:

  • This article from Nature on different ways to asses machine intelligence includes some great interactive tests to see whether you can tell AI and human apart and how you perform against the machines. I think one of the main challenges is that we don’t have a very good definition for what human intelligence is…
  • Researchers often make groundbreaking discoveries that can unlock huge opportunities. But the chances of that research turning into a commercial success can often revolve around what IP protections were put in place. Two patent lawyers from Knobbe Martens provide some tips for academics in this article on how to identify and protect patentable inventions. As they say, be proactive, not reactive!
  • Iain Gillespie, Principal and Vice-Chancellor of the University of Dundee (which was recently named as the UK’s top university for producing spinouts by Octopus Ventures) makes the case for more coordination and more support to help university spinouts.
  • Last year I read both the masterful Technological Revolutions and Financial Capital, which presents I think the most robust model of technological change albeit one that seems be less relevant today, and the fascinating Technology Trap, that distinguishes between short- or long-term impacts of technological change as well as replacing and enabling technologies. So I am excited to see a new book, Blood in the Machine, which I have added to my reading list. The book focuses on the historical Luddite movement and parallels with todays rise of gig work and hyper-automation. As George Santayana said:

“Those who cannot remember the past are condemned to repeat it.”

George Santayana
  • With that in mind, a couple of items exploring the implications of all this technology are first an article on the impact of robotics by Adnan Masood and secondly this conversation between Yuval Noah Harari and Mustafa Suleyman on AI:

Deeptech Deals

A round up of deep tech deals and new programs:

  • The US Solar Energy Technologies Office has opened up its Advancing US Thin-Film Solar Photovoltaics funding opportunity by making $36M available for research, development and demonstration of two thin-film PV technologies. The technologies being supported are early-stage Perovskite Tandem Photovoltaics and Advanced Cadmium Telluride Photovoltaics. SETO expects to make 2 to 15 awards with the funding and the deadline to apply is October 24th at 5pm ET.
  • UBQ has raised $70M to tackle residual household waste by converting it into advanced thermoplastic materials. The round was led by Eden Global Partners with participation from TPG’s Rise FundBattery Ventures and M&Gs Catalyst strategy. The funding will support global expansion for UBQ including additional facilities in Europe and North America. Meanwhile Waste Robotics has raised around $7.3M to accelerate its technology for autonomous waste sorting.
  • In a deal earlier this year, Renaissance Fusion raised €15M from Lowercarbon CapitalPositron Ventures and Norrsken VC. Renaissance will use the funds to help it reach its goal of a commercially viable 1GW fusion reactor by the 2030s. This article is a great exploration of Renaissance’s technology and development plans.
  • SCVC has announced its reached its first close on target to raise a $100M fund dedicated to investing in health and climate deep tech. The new fund will write cheques of $500k to $3M at pre-seed and seed. The fund’s first investment is VyperCore who are developing a fabless RISC-V processor.
  • TemasekNTU and NUS have launched a $55M program to support Singapore’s university spinouts. The program will target energy transition, biotechnology, computation and cognition startups.
  • ThinkMetal has raised $300K from in a pre-seed round led by 100X.VC. Founded in 2021, ThinkMetal is creating an ecosystem of hardware and software for metal 3D printing.
  • Oak Ridge National Laboratory has launched a new program to connect entrepreneurs with national laboratory developed technologies. The program, called Safari, is targeted at entrepreneurs who have previously established and sold at least one previous company. Safari program managers will then matchmake entrepreneurs with interesting technology across ORNL’s portfolio.