The Self-Driving Unicorn You Need to Meet & MicroStrategy's Bitcoin Bet | E2052

Jason and Alex kick off the show (0s)

• When new terms are created for revenue or expenses, it can be a warning sign, with a "pink flag" indicating a need to ask questions and a "red flag" meaning to stop participation (5s).
• For MicroStrategy holders who have doubled their money, the advice is to sell and buy Bitcoin directly, as they can now get more Bitcoin than before, and it's better to own the coins personally (30s).
• The concept "not your keys, not your coins" suggests that owning Bitcoin directly is preferable to holding it through a third party (41s).
• This week's guest on the show is Michael Saylor, who can explain the concept of owning Bitcoin directly (48s).
• The show "This Week in Startups" is brought to you by Cloud Devs, Vanta, and Kite, and is hosted by Jason Calacanis and Alex Wilhelm (56s).
• The show started 14 years ago as a daily interview with founders, but is now three days a week, discussing startups, technology trends, and the main character of the moment (1m54s).
• The hosts will be discussing various topics, including the main character of the moment, startups, and technology trends, with a big doc to discuss today (2m27s).

Introduction of guest Alex Kendall Co-Founder & CEO of Wayve (2m32s)

• Alex Kendall is the first guest on the show "This Week in Startups", and he is from the Wayve team, a company involved in autonomous driving (2m35s).
• Wayve is considered one of the most interesting companies in the realm of autonomous driving (2m49s).
• Wayve is slightly more under the radar in the United States compared to other companies, such as Waymo (2m53s).

Alex Kendall on Wayve's autonomous driving approach and AI technology (3m0s)

• Wave, a company backed by SoftBank, Microsoft, and Uber, has raised $1.05 billion in a Series C funding round and is part of the Twist 500, with a unique approach to autonomous driving. (3m0s)
• The company's approach to autonomy is centered around AI, specifically decision-making and complex multi-agent environments, and is focused on building embodied AI to enable self-driving to scale. (4m8s)
• Wave's AI model is a single neural network capable of driving in different countries, vehicle types, and on roads it has never been to before, using a camera-first solution and a single GPU. (5m1s)
• The company believes its technology can be brought to the mass market through machine learning, as many consumer vehicles being produced today have the necessary equipment, including a single GPU. (5m27s)
• The GPU mentioned is capable of running the whole neural network locally on the vehicle, without the need for an off-device component. (5m45s)
• Wave's approach to autonomy has shifted from writing lines of code and processing the world in real time to using neural networks and machine learning to enable self-driving. (5m55s)
• The company's AI model was trained on videos of humans driving correctly in complex environments, such as the Streets of London, to enable it to generalize situations and predict how others will behave. (6m14s)
• The company has been using an end-to-end approach since its inception and the founder's research days at the University of Cambridge, despite it being a contrarian idea 7 years ago (6m49s).
• The company's approach has led to remarkable outcomes, including training systems to understand how to drive a car and engage in language to explain actions or take instructions (7m17s).
• The models can be trained to not only drive and understand the world but also engage in natural language, allowing for a more human-like experience (7m24s).
• An example of this is a car being able to understand and respond to instructions, such as slowing down or explaining its actions while overtaking a bus (7m35s).
• The company's approach differs from Tesla's bet on using cameras alone to solve the self-driving problem, with the founder suggesting that their approach is more comprehensive (8m7s).
• Tesla's approach is based on the idea that human eyes have been taking input for 100 years, and cameras have higher fidelity and are looking in 360°, but the company disagrees with this assessment (8m13s).

LiDAR vs. camera-first solutions debate (8m29s)

• The industry is divided between LiDAR and camera-first solutions for autonomous vehicles, with some companies opting for a sensor-agnostic approach that combines multiple sensing modalities for redundancy and scalability (8m29s).
• A camera-first solution is considered more scalable long-term, as it provides the necessary information for decision-making, but incorporating other sensing modalities can help bring a system to market sooner (8m47s).
• Key considerations for sensor selection include a good supply chain, cost-effectiveness, and the ability to provide surround, redundant vision for L3 or L4 solutions (8m59s).
• Many mass-market vehicles, particularly luxury models, already come equipped with surround cameras, surround radar, and a single forward-facing LiDAR, making them a suitable platform for autonomy (9m11s).
• While some robo-taxis have more extensive sensing capabilities, camera-radar or camera-only solutions can still provide a high level of intelligent autonomous driving (9m40s).
• A company has been developing its autonomous fleet in London for seven years and recently expanded to the US, where it has been testing and learning to navigate new scenarios such as four-way stop signs and right turns at red lights (9m53s).

Cloud Devs - Visit for an unbeatable offer on hiring elite Latam talent today. (10m19s)

• Cloud Devs is the best place to hire top Latin American talent, with a market of millions of developers and over 10,000 senior engineers rigorously vetted for communication and problem-solving skills (10m39s).
• The platform can fill both technical and non-technical roles, from accounting to graphic design, with elite Latin American talent, and companies like Coinbase are choosing Cloud Devs due to its cost-effectiveness and time zone matching (10m53s).
• Cloud Devs is offering Twist listeners an exclusive "Pay What's Fair" deal, where they only pay what they feel the talent is worth, for a limited time, and can be accessed by going to clouds.com/twist (11m20s).
• A self-driving car system was tested in San Francisco, with no prior training or HD map, and was able to drive effortlessly, demonstrating generalization by applying concepts learned in different environments to new scenarios (11m40s).
• The system's ability to generalize and adapt to new environments is seen as a key factor in unlocking autonomy and bringing it to the world (12m8s).
• The system is described as generally applicable to environments it has learned from before, and can be taken to other places without relying on HD maps, making it sensor-agnostic (12m20s).

Economic viability and strategy of autonomous vehicles (12m23s)

• A different approach to bringing self-driving technology to scale is through driver assistance, which can be integrated into vehicles with the right equipment, potentially leading to global scale and millions of vehicles around the world with hands-off eyes on or hands-off eyes off L2 or L3 solutions (12m53s).
• This approach can bring more diverse data to enable an AI solution to achieve autonomy, with the key challenge being to make it a safe and economically viable solution (13m33s).
• The cost of self-driving cars by Waymo is estimated to be around $100,000 to $150,000, but this cost is expected to decrease as the technology advances and becomes more widely available (13m51s).
• Elon Musk claims that Tesla's self-driving cars can be made for $30,000, which seems reasonable as a future goal, with the vision being that every vehicle produced will be capable of hands-off eyes off driving (14m42s).
• The hardware needed for self-driving cars, including sensor packages and compute, is already available in more expensive cars, and the cost of these components is expected to decrease as they become more widely available, potentially making self-driving cars a mass market product (15m32s).
• The bill of materials for self-driving cars is estimated to be around $2,000, with the computer potentially costing $500 or less, and surround cameras and radar adding to the cost (15m39s).
• The company's plan is to build the most intelligent, safe, and trusted AI model, which requires scale, and to achieve this, they want to work with leading fleets and manufacturers around the world (16m21s).
• The company's AI model is flexible and can be distilled down into different variants to accommodate various sensor configurations and compute systems used by different manufacturers (16m37s).
• The goal is to have the AI power L2 to L5 driving, including driver assistance, safety features, hands-off eyes-off driving, and driverless vehicles, in both commercial and consumer vehicles (16m45s).
• The company sees itself in partnership with car manufacturers and fleets, including taxi services like Uber, to deploy its AI in diverse applications and make it safer (17m7s).
• The company has already integrated its AI into cars, vans, and has conducted grocery delivery trials, and believes that diverse data will make the AI safer (17m17s).
• The AI needs to be adaptable to different platforms that other manufacturers and fleets are building (17m31s).
• The company plans to have its AI technology available in a test fleet that can be accessed in the US, but the exact timeline and availability are not specified (17m36s).

Timeline for Wave's technology deployment and future market scenarios (17m49s)

• The public will be able to experience the product as part of a consumer vehicle that can be purchased, with the option to go for a ride in the Bay Area or London (17m50s).
• The timeline for the product's market release is not publicly disclosed, but it is expected to be soon, with the hardware and regulations in place, and the company working hard to ship the AI (18m0s).
• The market for self-driving technology is expected to be competitive, with various players such as Uber, Lyft, Tesla, and Toyota, each with their own approaches to winning (18m21s).
• Uber has multiple autonomy partners, including the company being discussed, and plans to be the aggregator, while Tesla wants to be the aggregator and provide the technology to other manufacturers (18m51s).
• In 5-10 years, likely scenarios for consumers include various companies providing self-driving services, with some possibly working together and others competing against each other (19m3s).
• The development of self-driving technology is compared to other technological advancements, such as the cognitive AI and language LEL space, and the competition between iPhones and Androids (19m16s).
• To successfully scale self-driving technology, it is essential to work on the hardest aspects first and ensure that the technology can be adopted seamlessly without requiring significant infrastructure changes (19m30s).

Vanta - Get $1000 off your SOC 2 (19m58s)

• Founders who want to sell to bigger customers need to clear compliance checks, including having SOC 2 (a standard that ensures companies keep their customer data safe) sorted out, in order to land large deals and operate at the highest end of the market (19m58s).
• Vanta makes it easy for companies to get and renew their SOC 2 compliance, with customers becoming compliant in just 2-4 weeks on average, and also automates compliance for GDPR, HIPPA, and more (20m30s).
• Using Vanta can save companies hundreds of hours of work and up to 85% on compliance costs, and they are offering $1,000 off SOC 2 compliance at v.com/twist (20m55s).
• The autonomy scale is expected to be driven by a trusted system that can drive in busy urban environments, merge with traffic, and interact with other intelligent behavior, which will come from an AI system (21m4s).
• There is a lot of space to run in the autonomous vehicle market, with 90 million vehicles produced each year, and the goal is to enable fleets and manufacturers to gain access to the benefits of autonomous technology (21m30s).
• The business model for autonomous technology could involve giving car producers a global license and charging a fee per vehicle produced per month, such as $10 per vehicle for up to a million vehicles per year (22m26s).

Autonomy subscription models and AI driving behavior challenges (22m34s)

• A vision for the future of robotics involves a subscription-based model for autonomy services, where consumers pay for the value of time saved, with the cost shared between the fleet or manufacturer and the AI provider (22m35s).
• Currently, the automotive industry is used to upfront purchases, but as the technology advances to Level 3 autonomy, a subscription model is likely to become more prevalent (22m58s).
• The cost of such a subscription service could be around $500-600 per month, or $6,000 per year, plus the cost of the car and insurance, although insurance costs may decrease with the use of autonomous software (23m23s).
• The value of time saved by using autonomous driving could be significant, with estimates suggesting that consumers might pay $5 per hour for the service, translating to $200 per month for 10 hours of driving per week (23m49s).
• The potential market for autonomous driving is enormous, with 90 million cars produced each year, and scaling the technology to reach a wider audience is crucial (24m16s).
• The application of embodied AI goes beyond automotive, with potential uses in other types of robotics, representing a vast market opportunity (24m29s).
• The development of autonomous driving technology is a competitive space, with companies vying to save lives and recapture wasted time, ultimately benefiting consumers and society (24m50s).
• The use of autonomous driving systems, such as Tesla's FSD, has improved significantly, providing a smooth driving experience, although some interactions, like traffic circles and left turns, can still be jittery (25m37s).
• A solution to the lack of decisiveness in self-driving systems is a self-supervised large-scale system that can learn emergent behavior, allowing the car to navigate complex scenarios like roundabouts and merging traffic without being jerky or uncomfortable for passengers (25m55s).
• This system can learn from dynamic scenes and adapt to different driving cultures, such as those found in London, where drivers often cut corners and break rules, but still manage to navigate safely (26m49s).
• To balance the training data learned from humans, who often bend or break rules, with the need to follow safety rules and regulations, a balance must be struck, possibly by going to regulators and explaining how humans actually drive (27m26s).
• Embodied AI, such as self-driving cars, has an advantage over cognitive AI in that there is a clear rule set to follow, and the goal is to create a system that can provably stay within those bounds while still providing a sense of collaboration and customization (27m46s).
• The system should be able to learn and adapt to different driving cultures, such as those found in different cities in the US, where drivers have distinct behaviors and expectations, such as in New York, LA, and Texas (28m30s).
• The ability to learn and adapt to these different cultures is crucial in avoiding road rage and ensuring a smooth and safe driving experience (28m35s).
• The development of self-driving cars may also provide a solution to AI safety as a system, as it requires the creation of a system that can provably stay within safety bounds while still providing a sense of collaboration and customization (28m4s).
• Road trade instances in Texas tend to go in a certain direction, with people being polite, whereas in New York, people tend to yell and scream at each other during interactions (29m14s).
• The possibility of AI models driving like people from different regions, such as New Yorkers or Texans, is being considered, with the idea that a car could adapt to a driver's style or be prompted to drive in a certain way (29m28s).
• A demo is being built that allows a person to drive for a couple of minutes and then ask the car to drive in their style, or prompt it to drive in a certain way, as long as it stays within safe bounds (29m48s).
• The car's AI model can be adjusted to accommodate different driving styles, such as a "Boston driving mode," which is jokingly referred to as being particularly challenging due to the reputation of Boston drivers (29m38s).
• The key aspect of the AI model is that it knows the bounds of what is safe and can adapt to different driving styles within those bounds (29m57s).

Kyte - Download the Kyte app today and use code JASON to save 10% on your first rental. (30m1s)

• Kyte is a car rental startup that delivers cars directly to customers, eliminating the need to visit a rental location and wait in lines (30m15s).
• The service allows users to manage their entire trip through the Kyte app, avoiding the need for manual data entry and extra fees (30m56s).
• Kyte operates in large cities across the United States, including Los Angeles and San Francisco, and offers a range of professionally maintained cars (31m12s).
• To use Kyte, customers can download the app or book online at ky.com, and use the code "Jason" to receive 10% off their first rental (31m17s).
• Kyte also offers a self-driving feature with customizable preferences, including an "aggressive mode" that allows for more frequent lane changes (31m39s).
• The self-driving feature aims to provide a more relaxed driving experience, allowing users to focus less on the road and more on other activities (31m51s).
• The company has received support from notable figures, including Bill Gates, who has test-driven their cars and expressed enthusiasm for the technology (32m25s).
• Kyte is working to expand its services, with plans to integrate with ride-sharing networks like Uber and Lyft, and potentially deliver food and other items (32m46s).

AI driving culture and Twist 500 update (33m0s)

• The progress made in AI driving technology is exciting and reminiscent of the early days of neural networks, with companies like Deepgram working on language translation and learning. (33m0s)
• The approach to AI driving feels similar, with the goal of making it work in various locations, regardless of the company behind it, such as Tesla, Wave, or Weo. (33m37s)
• Having multiple companies working on AI driving technology will lead to competition and more options for consumers, allowing them to compare safety records and track performance. (33m52s)
• The safety records of AI-driven cars will become a key factor in purchasing decisions, with metrics such as miles driven, interventions, accidents, and fatalities being considered. (34m35s)
• By the time the next generation is old enough to drive, AI driving technology is expected to be so advanced that it will be considered irresponsible to let humans drive, especially teenagers. (34m52s)
• A comment from the audience mentions that in countries like Holland, where guns are not prevalent, drivers are generally more polite and courteous, disputing the idea that guns are necessary for safety. (34m59s)
• A discussion about gun culture in Texas highlights the contrast with places like New York, where signs often warn of strict penalties for carrying guns, whereas in Texas, signs politely ask patrons not to bring firearms into establishments. (35m53s)
• The Twist 500, a project mentioned earlier, has been delayed due to various factors, including the addition of new shows and the company's move to Austin, but the team aims to reach 250 entries by the end of the year. (36m34s)
• The team, including Maddie from the research team, is working to get the Twist 500 project back on track, and more interviews can be expected in the future. (36m48s)

MicroStrategy's Bitcoin strategy and market concerns (37m5s)

• MicroStrategy, a company run by Michael Saylor, has been investing heavily in Bitcoin, with some people describing it as a Ponzi scheme due to its business model and Saylor's past trouble with the SEC (37m15s).
• The company's software business has been declining, but its Bitcoin business is thriving, with MicroStrategy being the number one holder of Bitcoin after Satoshi and above the United States government (38m46s).
• The company's stock price has been volatile, reaching an all-time high of $540 per share before dropping to around $371 per share, resulting in a potential loss of $150 to $100 per share for investors who bought at the peak (37m53s).
• Some people in the Bitcoin community, including Vinnie Lingham, are concerned that Michael Saylor's acquisition strategies could damage the Bitcoin ecosystem and brand (39m35s).
• The concerns include the use of convertible notes and SL bonds, which some see as a form of financial wizardry that could allow MicroStrategy to corner the market on Bitcoin (38m37s).
• Despite these concerns, the person discussing MicroStrategy is long on Bitcoin and believes it is a real and valuable asset, having bought some under $100 and made a significant profit (39m5s).
• The discussion highlights the religious intensity of people's complaints about questioning MicroStrategy's business model, with some people becoming overly defensive and personal in their responses (40m56s).
• The person discussing MicroStrategy emphasizes that their concerns are not personal and that they are happy for any entrepreneur to be successful, but they want to highlight potential red flags in the company's business model (40m22s).

Bitcoin convertible debt analysis (41m33s)

• The discussion involves MicroStrategy's convertible debt, which is a $3 billion worth of notes due in 2029, and are senior unsecured obligations that do not bear regular interest and the principal does not accrue (42m11s).
• The notes are convertible into class A common stock at a ratio of about 1.5 shares per $11,000 of principal in the notes, which means they will convert to stock at about a $672 per share rate, a 55% premium at the time of issue (42m36s).
• The notes' conversion rate is almost double now because the stock has come down, but usually in a convertible, one can also get their money back with a little bit of interest (42m55s).
• The people loaning the money could choose to get their $3 billion back plus 8% a year in interest, which could be buried in the documents, and this could be $3.5 billion or more in 5 years (43m22s).
• To pay this back, MicroStrategy would need to have cash, sell shares in the company, or sell the underlying Bitcoin, which could have a massive price impact if they have to liquidate (43m45s).
• Starting June 1st, 2028, holders of the debt have the right to require MicroStrategy to repurchase for cash all or any portion of their notes, a year before they actually reach the maturity point (44m14s).
• The repurchase price would be equal to 100% of the principal plus any accrued and unpaid special interest, although the definition of special interest is not clearly defined in the document (44m36s).
• MicroStrategy has issued convertible debt, which allows holders to convert their debt into equity at a later date, with the option to get their cash back if they don't think they'll make their money back by 2029 when the notes mature (45m6s).
• This setup could lead to a cascade where if MicroStrategy needs to repay the debt, they might need to liquidate some of their Bitcoin, causing the Bitcoin value to go down, which in turn would cause their share price to go down further (45m36s).
• The mechanics of this setup are complex, and it's possible that the SEC will require MicroStrategy to share more information about it, which could lead to consumers becoming more aware of what's going on (45m55s).
• An accounting expert could potentially build a model to understand the dynamics at play, using the number of Bitcoins MicroStrategy holds (around 200,000 or 300,000) and their cash reserves, which are not publicly disclosed (46m14s).
• MicroStrategy's CEO, Michael Saylor, has a concept of keeping the company's treasury in Bitcoin, rather than cash, which could impact their ability to repay the debt (46m30s).
• The company's plan to issue convertible debt every year could lead to a situation where they have an incentive to do bigger converts to buy more Bitcoin, which could drive up the price and create a self-reinforcing cycle (46m43s).
• However, the people providing the convertible debt may demand tighter terms, such as clawbacks or liquidation preferences, which could impact MicroStrategy's ability to issue future debt (47m30s).
• The company's stock trades at a multiple to the value of its underlying Bitcoin, which is not fully understood, and may be related to the value of its software business (48m28s).
• MicroStrategy holds a significant amount of Bitcoin, with the last reported amount being 386,000 Bitcoin. (48m38s)
• The company has a massive premium on the net asset value of its underlying Bitcoin, and the reason for this premium is unclear. (48m44s)
• Despite reading various responses and tweets, no clear answer has been found to explain the premium, with some people becoming defensive and suggesting that others should "do their own research." (48m52s)

Deciphering complex business models and historical examples (49m3s)

• When businesses are too difficult to understand, it often raises concerns, as seen in the speaker's experience with over 400 startups and 2,000 podcast episodes, where complex business models can be a red flag (49m28s).
• The phrase "do your own research" can be a warning sign, as it may indicate that the business model is too complex or not transparent, similar to the case of Do Kwon, the CEO of Terra, who was arrested after failing to explain his algorithmic stable coin (49m40s).
• Michael Saylor, the CEO of MicroStrategy, used the term "BTC yield" to describe a ratio between the total number of Bitcoin owned by MicroStrategy and each block of 1,000 shares outstanding, which does not actually refer to yield (50m31s).
• The term "BTC yield" is notable, as it may be an attempt to create a new metric, similar to "Community Adjusted EBITDA," which was used to make a company's numbers look better by adjusting the EBITDA metric (50m56s).
• The use of the term "BTC yield" raises concerns, as it may be an attempt to obscure the actual performance of MicroStrategy's Bitcoin holdings (51m1s).
• The ratio described by "BTC yield" is essentially a measure of how many Bitcoin are owned by MicroStrategy per 1,000 shares outstanding (51m12s).

Evaluating new financial terms and investor advice (51m24s)

• When evaluating new financial terms, it's essential to understand the actual meaning behind them, as they might be misleading, such as the term "yield" being used to describe a ratio, specifically the ratio of how much Bitcoin one gets per share (51m24s).
• New terms for revenue or expenses can be a warning sign, with "pink flag" indicating a need to ask questions and "red flag" signaling to stop participation (51m35s).
• For MicroStrategy holders who have doubled their investment, the advice would be to sell the shares and buy Bitcoin directly, as this would allow them to own the coins themselves and potentially get more value, following the principle of "not your keys, not your coins" (51m59s).
• Owning Bitcoin directly or through a trusted platform like Coinbase or Robinhood is recommended for those who want to invest in the cryptocurrency (52m20s).
• The importance of having control over one's own coins is emphasized, as relying on a third-party platform may not provide the same level of ownership and security (52m14s).

Final thoughts and invitation to Michael Saylor (52m29s)

• Michael Saylor is welcome to be a guest to explain his perspective, and it's not a personal issue. (52m29s)
• The root cause of MicroStrategy trading at a premium to its net asset value (NAV) remains unclear, and no satisfactory explanation has been provided. (52m38s)
• Financial coverage has failed to provide a good explanation for the premium, with the only reason being "stocks go up," which is not considered a good long-term investing position. (52m45s)
• The conversation will be continued on the next episode, scheduled for Monday. (52m58s)