(Re)Share | #23 - I'll take the blue pill
Moon landings | BCI | Sentient AI | Nuclear fusion | AI Safety | EV batteries
Greetings to all the new subscribers that signed up because I may or may not but 100% did shake down all of my personal email contacts. I apologize for nothing.
Stuff worth sharing
Where no man has Goa-ne before - Just last week India became the fourth nation in history to successfully land an object on the moon. The Chandrayaan-3 touched down near the lunar south pole in a seemingly flawless mission. It’s been four years since India’s last landing attempt, during which massive overhauls were undertaken on all levels of the probe’s design. Highly recommend watching mission control’s landing reaction if you need your heart warmed.
60’s era saber-rattling - I’m currently reading One Giant Leap and in it I learned that President Kennedy was the first to install secret recording devices in the Oval office. As a result the book features an actual transcript with NASA head, James Webb, arguing about the space race with Russia vs. a more holistic orbital research. Tl;dr - JFK didn’t actually care about space or the moon, he really just wanted to embarrass communists. Respect.
Interstellar incomPutince - Just a few days before India’s glorious triumph, Russia also launched a spacecraft to the moon. To be clear, Russia launched a spacecraft into the moon. To be more clear, Russia launch a spacecraft that crashed into the surface of the moon. The unmanned Luna 25 mission hit a snag in lunar orbital calculations, which are caused issues in its descent plan. Pretty wild juxtaposition to see this happen while reading the history above.
I’ll take the blue pill - The Matrix is a story of brain-computer-interface (BCI). You think of things, which creates electrical signals, which are detected by an implant, which translates them into reality through a digital avatar. While the film paints BCI in an apocalyptic light, this is a really remarkable story of when it can go well. Ann Johnson suffered a stroke 18 years ago and has been unable to speak or move since but this astounding research group from UCSF gave her voice back. Using a paper-thin grid of 253 electrodes and machine learning trained on her own wedding video, Johnson was able to speak with her husband for the first time in nearly two decades. I highly recommend watching this 5 minute segment on the research or you can read the publication here.
ChatGPTPS Reports - Open AI unveiled an enterprise-grade product featuring enhanced security and privacy, unlimited GPT-4 access, and a significant expansion of context windows for longer inputs. Amazing news for cost conscious corporate middle managers. Terrible news for a lot of AI startups.
A really weak week - Cruise recently gained approval from California’s public utilities authority to operate at all hours in the city. An enormous win for any AV company. They subsequently threw it all away in spectacular fashion by getting into two crashes in seven days, one of which with a fire truck. Thankfully no one was seriously injured apart from Cruise’s reputation.
Invisible ink - Dwight’s not the only one with secret messages. Deepmind released SynthID a new tool to covertly watermark and identify synthetic images. Watermarking is one of the most cited new methods of authentication against increasingly ubiquitous Generative AI. SynthID embeds a digital watermark directly into the pixels of an image, making it imperceptible to the human eye, but detectable for identification. The theory being that platforms / publications / etc. could then filter out synthetic media where applicable. Loads of implementations challenges but this is an active area of investigation for me.
Sensing sentience - I really like dinner party question games and my favorite is, “What is the closest thing to a conspiracy that you actually believe?” My answer has always been the idea that we may be enslaving sentient AIs without realizing it. Hypothetically if an AI was developed that was actually self-aware, how would we know? This paper attempts to answer that through a collection of ”indicator properties” of consciousness, which collectively could detect a self-aware program. According to the researchers, there are currently no conscious AI systems at present. Unless that’s what they want us to think…
Watts up with that? - The annual RAAIS conference released recordings of the keynote talks, which featured a bunch of heavy hitters. My personal favorite was from Siddharth Khullar, Senior Director of Machine Learning R&D at Northvolt. For the uninitiated, Northvolt is a juggernaut of EV batteries and are really pushing the boundaries of what’s possible; in large part due to the incredible ml research that they undertake. I encourage a full watch but specifically the segment on electrode inspection (21:14) and how they predict / prevent the physical defects that cause chemical explosions.
Cost-effective Gogeta - A friend sent me a startup that’s building a micro reactor for nuclear fusion. While I can’t share the company’s materials I did find the scientific basis quite interesting. Nuclear fusion is a process that produces energy when two nuclei join to form a heavier nucleus. This is significantly safer than fission (splitting atoms) but it requires an insanely difficult environment of heat and magnetic force, making it (so far) unscalable. The Lattice Confinement Fusion method is a novel approach that creates the same kinetic energy but in an ambient metal solid. There is a lot more detail in the article on charged fuels and measured reactions but I’m guessing I’ve lost most readers at this point.
Portfolio Flex
Our newest investment, Martian Lawyers Club, announced their $2.2 million pre-seed round. MLC applied generative AI to video games for player personalization. Where most startups focus on visual elements (skins, items, etc.) MLC takes aim at core game mechanics (speed, gravity, etc.). An extremely ambitious mission, to be sure, but we’re super excited about it.