Grape-picking robots, startup investing for all, and the earnings numbers you need to know | E2037

Jason and Alex kick off the show (0s)

• Sam Altman mentioned that even with Azure, OpenAI faces significant limitations and hard decisions about allocating compute resources, and they are blowing through $5-10 million to $10 billion a year in infrastructure while losing a lot of money (14s).
• Jensen Wong talked about Elon Musk standing up a giant cluster, indicating the ongoing buildout, and the interface of ChatGPT is superior to other interfaces, although the results are not (32s).
• The show is sponsored by Fidelity Private Shares, LinkedIn Jobs, and Oracle Cloud Infrastructure (42s).
• The hosts, Jason and Alex, welcome viewers to the show and mention that they will not be discussing the upcoming election, instead focusing on startups and technology (1m29s).
• The hosts discuss the election briefly, mentioning that the prediction markets are coming back down, and the gap between the candidates is closing, with 64.9 million votes already cast according to NBC (2m55s).
• The show will feature the founder of Extend Robotics, discussing teleoperation, VR, and robotics, as well as an upcoming bill in the Senate that might bring private market investing to everyone, and critical earnings updates (3m12s).
• Jason mentions that he had seen a video of grapes being picked by robots and AI, and they had invested in a company called Rude AI that was doing picking of berries, and this is something that will be seen in the coming years with general robotics powered by AI (3m45s).
• Jason explains that vertical AI is well-known, but having a robot do general tasks like folding laundry, picking grapes, and making coffee is a different challenge (4m17s).

Extend Robotics’ Dr. Chang Liu joins the show (4m49s)

• Dr. Chang Liu is the founder and CEO of Extend Robotics, a UK startup established in 2019 (4m51s).
• The purpose of Extend Robotics is to develop robotic software that extends human capabilities beyond physical presence, aiming to amplify labor productivity through immersive technology, robotics, and artificial intelligence (5m40s).
• The company's goal is to help the world transition from traditional vertical factory automations to real-world general-purpose automations (6m2s).
• Extend Robotics aims to combat the labor shortage, a widely discussed issue, by developing robotic software that can assist in various tasks (6m15s).
• Dr. Liu agrees that advancements in LLMs and AI, which enable generalizing AI versus vertical AI, will significantly change the field of robotics in the coming years (5m31s).

The challenges of real-world automation, robotic grape picking, and human in the loop AI (6m23s)

• Real-world automation is challenging due to the need for multi-purpose systems that can work in dynamic environments with high variability tasks, which was previously deemed impossible (6m26s).
• Recent breakthroughs in generative AI, led by ChatDPT, have enabled the development of AI systems that can be embodied in physical form factors like robots, allowing for human-like experiences in physical work (6m44s).
• A handheld robot was shown doing lab work and picking grapes, demonstrating the potential of robots to perform tasks that require human-like dexterity (7m5s).
• The robot was trained using a combination of human teleoperation and machine learning, allowing it to learn from humans and improve its performance over time (7m22s).
• The robot's training involved a human operator wearing a VR headset and using gestures to control the robot, allowing for finger-level controls and real-time operation (8m28s).
• The software used to control the robot has three purposes: teleoperation, data collection for AI training, and AI inference and supervision (8m36s).
• The goal of the software is to enable seamless shared automation between humans and AI, allowing them to complement each other and improve overall performance (9m3s).
• The technology is currently being developed in collaboration with Queen Mary University on a grape-picking project, which involves teleoperation of robotic arms in the field and allowing the robots to learn from humans (9m23s).
• The current level of technology is teleoperation, but the goal is to eventually enable robots to perform tasks mostly by themselves, with human supervision and intervention as needed (9m41s).
• The automation process is viewed as a spectrum, with human involvement gradually decreasing as AI capabilities improve, and is referred to as "human in the loop AI" (9m50s).
• The human in the loop AI approach aims to amplify labor with AI, rather than replacing it, and enable smooth transitions between human and AI operation (10m31s).

Fidelity Private Shares℠ - Visit Mention our podcast and receive 20% off your first-year paid subscription. (10m44s)

• Interest rates are decreasing, but it remains challenging for companies to raise new funding rounds, making it essential for founders to manage their time efficiently while dealing with cap table management, due diligence, and investor relations (10m44s).
• Fidelity Private Shares offers a solution to help founders manage their equity, cap table, and data room, providing more accurate 409a valuations and quicker results (11m1s).
• With Fidelity Private Shares, founders can model the future of their company's equity, automate their next funding round, and maintain a professional image, even when dealing with multiple tasks (11m25s).
• Fidelity provides great support to its clients, including a startup community, events for founders and venture capitalists, and a partnership that can help founders appear more professional and organized (11m47s).
• Founders can receive 20% off their first-year paid subscription to Fidelity Private Shares by visiting fidelityprivateshares.com and mentioning the podcast (11m56s).

Teleoperation, network latency, and future of robotic harvesting (12m14s)

• The concept of humans in the loop involves AI learning from humans, with the AI eventually stepping back and allowing humans to supervise and intervene only when necessary, with the ability to teleoperate whenever needed (12m20s).
• In the context of grape-picking robots, the challenge lies in training the robotic arms to handle delicate fruits without damaging them, requiring innovations in sensory information and force sensing (13m10s).
• To address this challenge, the development of robotic grape-picking involves collaboration with universities and wine makers, incorporating force sensing and human-in-the-loop AI software capabilities (13m52s).
• The term "grippers" is used in robotics to refer to the robotic hands or arms used for tasks such as grape-picking (14m39s).
• It is estimated that a wine harvest could be done completely robotically within the next 3 to 5 years, with commercialization and adoption taking some time (15m2s).
• The potential benefits of automated grape-picking include reducing the need for skilled or unskilled workers and allowing the best caretaker of grapes and vines to become the global single vine manager (15m58s).
• The concept of a single, skilled individual becoming the global standard for a particular task is illustrated by the example of a great pilot, such as Sully, who landed a plane on the Hudson River (16m23s).
• A hypothetical scenario is presented where a person with expertise in managing vines could remotely manage every vine in the world using a robotic system, potentially replacing migrant workers who travel to pick grapes, with the robots doing the work in Napa and the person controlling them from Mexico using a VR headset (16m53s).
• The technological breakthrough of broadband penetration around the world enables fast connections, which undergirds the ability to interact with robotic arms remotely (17m20s).
• To interact with a robotic arm, a fast connection is required, and the latency between moving one's hand and the robot moving its arm is around 100-150 milliseconds, in addition to the network latency (17m43s).
• The system can work on commercial internet, 5G networks, and potentially on Starlink satellite connections from low earth orbit satellites, due to the ability to compress data small enough to stream over standard networks (18m28s).
• The company's technology is not limited to picking grapes, but has a wide range of applications, including nuclear, space robotics, and manufacturing, with a focus on general-purpose technology that can be applied to various industries (19m25s).
• The company started with nuclear applications, then moved to space robotics, and recently began working on agriculture and manufacturing projects, with a focus on addressing labor shortages in these industries (19m30s).
• The technology has the potential to be used in various manufacturing industries, including automotive, due to its general-purpose nature and wide applicability (20m3s).
• Dr. L is a UK-based startup that has raised grant money from the UK government, and they have found the UK to be supportive of its startup community, even in a post-Brexit world (20m47s).
• The UK is considered one of the best places for building a technology company (21m4s).
• Dr. L's application is contributing to advancing technology, with a focus on commercial adoptions at manufacturing applications (20m38s).

Building a tech company in the UK and second-order effects of AI (21m5s)

• The UK is considered a good environment for software-focused startups due to government support for innovation and the availability of capital, especially in the London area (21m7s).
• The UK government provides support for startups and small companies to innovate, making it a good starting point for new businesses (21m17s).
• Dr. L's company, Extend Robotics, is an example of a tech company in the UK that can be looked up online, and another company, Saffron Gra, is mentioned as a potential interest for those who enjoy sparkling wine (21m48s).
• The development of technologies like grape-picking robots has second-order effects, such as the potential for the greatest single practitioner in every vertical to address every instance of a task (22m19s).
• The use of AI in tasks like grape-picking can elevate the quality of products, such as wine, while reducing costs, as all the secrets and nuances of the process will be captured and made available (23m10s).
• The perfecting of tasks through AI can lead to the bottom half of products, such as wine, becoming as good as the top half, as all the little nuances will be captured and replicated (23m12s).

LinkedIn Jobs - Post your first job for free (23m24s)

• In today's venture market, every single hire made has to be perfect, as mistakes cannot be afforded, and the runway needs to be kept as long as possible to run more experiments and achieve product-market fit (23m29s).
• To achieve this, it is essential to find talented people, and LinkedIn Jobs can be a valuable resource, as it brings candidates that cannot be found anywhere else (23m46s).
• LinkedIn has over 1 billion members, and 7% of its users do not visit other leading job sites, making it a unique platform to find top talent (23m52s).
• These users may not be actively looking for jobs but are engaged in professional development, networking, and other activities on LinkedIn, making them potential great hires (24m5s).
• LinkedIn is offering a special deal to post a job for free, which can be accessed through linkedin.com/twist, with terms and conditions applying (24m25s).

AI models learning from best human practitioners (24m35s)

• AI models are being trained to learn from the best human practitioners in various fields, with the goal of becoming the best technology product in their respective domains (24m35s).
• These AI models can learn from human perception and experience, allowing them to improve over time and make better decisions, such as determining the optimal time to pick grapes or identifying areas for improvement in wine production (25m12s).
• The process of learning from human experience and improving over time is similar to the way the aviation industry learns from plane crashes, with a focus on identifying the root cause of the problem and implementing changes to prevent similar incidents in the future (25m52s).
• Unlike the aviation industry, many other fields, such as grape-picking, do not typically conduct thorough postmortem analyses or debriefs to identify areas for improvement, but the use of AI models could change this (26m37s).
• The development of AI models that can learn from human experience and improve over time has the potential to create the "perfect pilot" or professional in every field, with the ability to make optimal decisions and perform tasks with precision and accuracy (26m43s).

Global remote work impact and virtual assistants (26m56s)

• The shift to global remote work is expected to have a significant impact, with companies like Amazon facing challenges in implementing a five-day workweek policy, while others are embracing remote work and virtual assistants (26m56s).
• Virtual assistants can work from anywhere, including call centers or homes with satellite connections, and companies like Athena offer virtual assistant services, with investors in the company (27m38s).
• Many startups, including Launch, have virtual assistants working in their operations departments, often from countries like the Philippines, due to the difficulty in finding assistance and admin workers in the United States (27m46s).
• The trend of remote work and virtual assistants raises questions about the future of wages, particularly for developers in the US, who may face competition from international developers who can work for lower salaries (28m11s).
• It is possible that US-based developers may need to adapt to a global standard for salaries, with wages converging on a global average, and people deciding if they want the "luxury item" of having someone work in person (30m2s).
• The ability to work in person may become a valuable trait for developers in the US, particularly for high-end jobs that require face-to-face interaction and decision-making (28m40s).
• The shift to global remote work may lead to a new dynamic, where high-end developers can command high salaries for working in person, while others work remotely for lower salaries (30m14s).

Impact of global salary standards on US developers (30m38s)

• The increasing efficiency play due to global salary standards may negatively impact US developers who are accustomed to higher salaries, making it tough for them, especially those who are not A+ players (30m39s).
• A+ players will likely be fine as they can work with other top talent in lower-cost living areas, but B players may struggle to adapt (30m48s).
• The labor shortage in various industries, including the US, will drive the demand for technology products and innovative solutions (31m52s).
• The lack of vibrant immigration and protectionist policies in some countries may lead to labor gaps, prompting governments to explore alternative solutions, such as offering visas to attract foreign workers (32m2s).
• Japan is an example of a country that has made decisions about immigration and citizenship, and while they do not offer citizenship easily, they are more open to giving visas to fill labor gaps (32m11s).
• The labor shortage has already led to significant changes in some industries, such as the cashier business, where automation and online ordering have become the norm (32m33s).
• Humans can adapt quickly to changing circumstances, and the shift to technology-driven solutions can happen rapidly, as seen in the transition to online ordering in Quick Serve restaurants over the past decade (32m56s).

Oracle - Save up to 50% on your cloud bill (33m2s)

• AI is transforming various industries, including self-driving cars, medical research, and business tools, and its adoption is happening rapidly, with significant computing power requirements that can lead to skyrocketing costs if not managed properly (33m4s).
• Oracle Cloud Infrastructure (OCI) is a fast and secure platform that covers infrastructure, databases, app development, and AI and machine learning needs, offering a cost-effective solution (33m31s).
• OCI costs 50% less for compute and 80% less for Network compared to other cloud options, resulting in substantial savings, especially for businesses where every minute counts (33m45s).
• Thousands of businesses, including MGM Resorts, Specialized Bikes, and Fireworks AI, have already upgraded to OCI (33m58s).
• Oracle is offering new US customers with a minimum financial commitment a chance to cut their current cloud bill in half if they switch to OCI, but this offer is only valid until December 31st, 2024 (34m8s).
• Interested companies can check if they qualify for this special offer at oracle.com/twist (34m22s).

Accredited investor rules and potential changes (34m28s)

• In the US, there is an investor accreditation program that allows individuals to invest in private companies if they meet certain thresholds, such as making over $200,000 a year or having a net worth of $1 million, which can be combined with a spouse's net worth (34m37s).
• Being an accredited investor means being considered "sophisticated" or "intelligent" enough to invest in alternative assets, such as private companies, which can be riskier than public companies (35m17s).
• The Securities and Exchange Commission (SEC) and the government believe that only the top 5% of earners or wealth holders in the US are capable of making informed investment decisions in private companies (35m15s).
• However, individuals can still participate in other forms of investing, such as betting on sports or public equities, and even investing in cryptocurrencies (36m46s).
• The ability to invest in private companies can be a key factor in creating wealth and moving up the social ladder, as investing in companies like Uber or LinkedIn before they went public could have resulted in significant returns (37m10s).
• There is a movement in the Senate to change the rules around investing in private companies, and some argue that individuals should be allowed to take a test to become "sophisticated" investors, similar to tests required for other activities like cutting hair or owning a firearm (34m50s).
• The current system is seen as condescending, as it assumes that 95% of citizens are unable to make informed investment decisions, while still allowing them to participate in other forms of risk-taking, such as sports betting or playing Blackjack (36m38s).
• A proposed idea is to have people take a test before being allowed to buy more than $1,000 in chips for games like blackjack or poker, to ensure they understand the rules and risks involved (38m21s).
• The number of people who meet the accredited investor threshold has increased significantly, from 1.8% of households in 1983 to 18% or 24.3 million US households today, due to inflation and the changing value of money (39m12s).
• This expansion of accredited investors has diluted the standard, making it almost meaningless, and raises the question of why it is still needed (39m30s).
• The SEC's mandate is to protect consumers above all else, which makes them risk-averse and focused on avoiding situations like the TX incident (39m50s).
• The proposed legislation, the Empowering Main Street in America Act, aims to allow more people to invest in startups and potentially close the wealth gap, but the SEC's mandate does not prioritize this goal (40m11s).
• The SEC's incentives are not aligned with helping people become middle class or millionaires, but rather with protecting the public and building trust in the financial system (40m53s).
• The coastal commission in California is used as an example of how a regulatory body's mandate can limit progress, in this case, protecting the coastline but hindering space launches and development (41m10s).
• The example of a proposed ferris wheel in Golden Gate Park is also used to illustrate how regulatory bodies can prioritize preservation over progress (41m46s).
• There are two groups of people with different perspectives on a certain issue, with one group viewing it negatively and the other seeing its potential charm and romance, particularly for proposals. (42m0s)
• The negative views of one group are outweighing the broader benefits, leading to a decision that may not be in the best interest of the bigger picture. (42m19s)

Sophisticated investor tests and angel investing (42m26s)

• A sophisticated test for investing in private companies is being developed, which should require 3-4 hours of preparation and consist of 50-100 questions, similar to a driver's license test (42m26s).
• This development could make America more competitive by allowing more people to become wealthy and invest risk capital, with the majority of company appreciation happening in the private markets (43m11s).
• Investing in private companies can provide higher returns than public markets, with the example of a company going public at a $1 billion valuation and increasing by 40% in a year, but potentially providing hundreds of percent returns if invested in at a $10 million valuation (43m45s).
• Senator Tim Scott from South Carolina is behind a package of reforms that includes extending the JOBS Act to companies with up to $2 billion in revenue, allowing them to go public under different rules (44m5s).
• Investing in private companies is expected to exist as an option between gambling and public markets, with the understanding that it carries risks, but can also provide potential high returns (44m47s).
• Angel investing is a high-risk, high-reward option that requires diversification, with experts recommending 30-50 qualified investments to potentially achieve a power law return (45m36s).
• The potential for high returns in angel investing is reason enough to consider it, despite the risks, and it can be done with a low single-digit percentage of one's net worth (46m3s).
• To get involved in investing, it is advised to allocate 5% or less of one's net worth, an amount that can be recovered if lost, through other equities and interest earned that year (46m6s).
• This approach allows for a relatively conservative financial positioning, as seen in the example of a family's investments in zero-cost index funds and low-cost international bond funds (46m33s).
• The importance of being cautious when investing in non-traditional ventures, such as restaurants and movies, is highlighted, with a personal preference for passing on such investments due to their low potential for success (46m55s).
• An exception to this rule is investing in friends' projects, as seen in the example of investing in Nick Sh's film, where the primary motivation is to support the friend rather than to make a profit (47m14s).
• In such cases, the potential for financial loss is accepted, and the focus is on helping the friend's project succeed, as was the case with Nick Sh's film, which ultimately did well and is expected to break even (47m22s).

Investing in local businesses and AI news (47m32s)

• A coffee store called Brown B, located in Providence, has recently opened near the narrator's house, and the owner's name is Will, a lovely guy who has been building out and remodeling the place (47m33s).
• The narrator thinks it would be great if they could invest $10,000 in the company, but the current system is not set up for that, instead relying on bank loans and other traditional methods (47m47s).
• The narrator imagines a scenario where Brown B could raise money by putting a sign in the window, stating that they are looking for 100 top customers to invest $1,000 each in a $100,000 renovation, in exchange for perks such as upgraded coffee and ownership of 1% of the company for every $10,000 invested (47m56s).
• The narrator notes that this type of investment would currently be illegal, which they find ridiculous, despite the fact that it is done anyway (48m30s).

OpenAI announcements and Google’s blowout quarter (48m36s)

• OpenAI made three significant announcements, including an AMA on Reddit that was well-received, where Sam Altman discussed the limitations and hard decisions the company faces in allocating compute resources, revealing they are compute-constrained despite having Azure, and are losing a lot of money, with estimated annual infrastructure costs of $5-10 billion (48m42s).
• The interface of ChatGPT is considered superior, but the results are not, and it is suggested that Google needs to create a dedicated Gemini app with a consumer-friendly interface (49m32s).
• Canvas on ChatGPT allows users to write notes and ask questions, refreshing the same document, which is a useful feature, and it has been used to edit documents without directly editing them (49m52s).
• ChatGPT's voice application can be used to give quizzes on various subjects, making it a potential tool for learning and education (50m43s).
• The availability of custom AI tutors for free and for life on any subject has the potential to democratize access to education and make the world more just (52m10s).
• The use of AI tutors can provide an advantage in learning, similar to having human tutors, but without the financial constraints, making it a game-changer for people who cannot afford traditional tutors (52m13s).

Democratizing education with AI tutors and OpenAI's new search product (52m22s)

• Google's earnings report showed a quarterly revenue of $88.2 billion, up 15% from the previous year, with earnings per share of $21.12, both beating expectations (53m33s).
• Google Cloud had a significant quarter, with revenue up 35% to $11.4 billion, and an operating profit of nearly $2 billion, making it a profitable segment for the company (53m46s).
• Alphabet reported that AI is making search better and increasing overall search usage and user satisfaction, although this may not reflect the personal experience of some users (54m52s).
• The disparity between the reported increase in search usage and personal experience may be due to the fact that the average user of Google Search is not the same as early adopters, and startups should consider the needs of the "massive middle" and "laggards" (55m1s).
• Google's search has moved further away from what some users want, but it is still widely used, and the company's earnings report showed a significant increase in revenue from advertising and subscription services (52m52s).
• YouTube had a blowout quarter, with $8.9 billion in quarterly revenue from advertising, and the company's total revenue for the last year was $50 billion, with $35 billion from advertising and $15 billion from subscription revenue (52m41s).
• The earnings report also highlighted the growth of Google's subscription services, which may be an area of interest for startups and investors (52m59s).
• The use of AI-powered search tools, such as ChatGPT, is becoming more common, and some users are finding that their default search behavior is shifting away from traditional search engines like Google (56m1s).
• Google's dominance in search may be threatened by chatbots like ChatGPT, which provide direct answers to users' questions without the need for sponsored links or advertising, making it challenging for Google to maintain its revenue stream from search (56m20s).
• Having multiple revenue streams is crucial, as seen in Google's case, where its investments in YouTube and Cloud are expected to save the company despite potential declines in search revenue, with YouTube generating $50 billion in revenue and Cloud generating $11.4 billion in revenue (57m47s).
• Google's search business is still the largest driver of its income, with $50 billion in revenue for Q3, but it is expected to face significant pressure from competitors like Open AI's new search product (58m50s).
• Open AI has launched a new search product for paid subscribers, which uses a different search operator and provides more direct answers to users' questions, potentially competing with Google's search dominance (59m3s).
• The new search product is seen as a threat to other companies built on top of ChatGPT, such as Perplexity, which may face competition from Open AI's new product (59m24s).
• Google's other businesses, such as YouTube and Cloud, are expected to continue generating significant revenue, with YouTube's profits reaching $2 billion and Cloud's profits reaching nearly $2 billion (58m5s).
• Google's investment in Waymo is expected to be a long-term bet, with the company potentially losing $10 billion per year for the next 10 years, but ultimately generating significant revenue in the future (58m30s).
• A new AI search engine has been launched, which aggregates information from various sources, including listicles, and presents it in a compressed list form, raising concerns about content ownership and potential lawsuits from content providers like the New York Times (1h0m32s).
• The AI search engine, developed by Sam Altman's company, does not provide direct links to the original sources at the top level, instead burying them under a "sources" button, which is seen as a key mistake and potentially cutthroat move (1h2m26s).
• The company claims to be working closely with existing content partnerships, but the lack of transparency and proper citation is still a concern (1h2m11s).
• The AI search engine's approach is compared to Google's "one box" feature, which also aggregates information from various sources, but at least provides some traffic to the original content creators (1h1m15s).
• The New York Times is advised not to settle their lawsuit against the AI search engine company, as it could set a precedent for the content industry (1h0m51s).
• OpenAI has also launched an advanced voice feature on their desktop app, which allows for a more interactive and conversational experience with the computer (1h3m7s).

OpenAI’s advanced voice feature and earnings highlights (1h3m39s)

• OpenAI is working on an advanced voice feature that can monitor a user's desktop and provide supplemental information while they work or engage in conversations, potentially revolutionizing the way people access information (1h3m39s).
• This feature is envisioned to be a persistent agent that can watch what the user is doing and provide relevant information, similar to the concept of "popup videos" that used to provide additional context while watching music videos (1h4m20s).
• The idea is that this AI agent can provide information on various topics, such as recognizing people or objects in videos, or pointing out mistakes in TV shows or movies, like the infamous Starbucks cup in Game of Thrones (1h4m51s).
• The conversation shifts to discussing earnings reports, with companies like Amazon and Apple being mentioned, and the potential for Google to break up and unlock shareholder value if it spins out some of its businesses (1h5m48s).
• Alphabet's earnings call revealed that more than a quarter of all new code at Google is generated by AI, highlighting the company's increasing reliance on artificial intelligence (1h6m21s).

Tech giants' AI initiatives and implications (1h6m26s)

• Google engineers are now reviewing and accepting AI-generated code, which is expected to significantly increase the amount of code produced, with the potential to double and then double again (1h6m26s).
• The majority of this code will be background and boring, but it still represents a massive amount of code that originated from AI, which is a significant development for a company like Google that is known for its engineering prowess (1h6m31s).
• This trend could lead to robots creating robots, with AI agents writing AI agents, creating a flywheel effect that could potentially become unstoppable and pose an AI risk (1h7m1s).
• As AI agents become more advanced, they will be able to perform more tasks, such as making purchases, and it's only a matter of time before they start spinning at an uncontrollable velocity (1h7m26s).
• There is a small chance, estimated at 1% or 10%, that this trend could become unstoppable and lead to a significant shift in the way code is produced, with the potential for 25%, 50%, or even 99% of all code being generated by AI in the future (1h7m34s).
• This development could be part of a larger trend where AI-generated code becomes the norm, with the potential for 99.59% of all code in the world to be generated by AI (1h7m55s).

Amazon’s AWS growth and AI business expansion (1h8m3s)

• Amazon's AWS grew 19% last quarter, reaching a $10 billion annualized run rate, with reaccelerated growth over the last four quarters, making it an incredibly profitable business on an operating basis (1h8m24s).
• AWS is two and a half times larger than Google Cloud, with AWS at $27 billion and Google Cloud at $11 billion, as software continues to eat the world (1h8m51s).
• Amazon has been reducing headcount, along with Google and Meta, but still achieving 20% growth, indicating a bull case for big Tech and America (1h9m42s).
• Amazon's AI business is a multi-billion dollar revenue run rate business, growing at triple-digit year-over-year percentage and more than three times faster than AWS itself at this stage of its evolution (1h10m22s).
• The demand for AI is driving growth at AWS, with the AI business expected to triple in size each year, potentially reaching 30% of Amazon's yearly revenue in the future (1h10m49s).
• Startups are building with AI and using hyperscalers like AWS, but paying significant margins to use these mega compute systems, raising the question of whether there's a case to be made for more cost-effective alternatives (1h11m52s).
• Startups are becoming proficient in managing their own hardware, reducing their reliance on Amazon and the associated premium costs (1h11m59s).
• Google and Oracle offer a large number of credits to startups, which can be accessed through the link provided, and a new website, getstartupcredits.com, lists top deals for startups (1h12m10s).
• The credit economy for startups benefits companies in their first two years, with Google, Azure, AWS, Oracle, and others competing to offer discounts and attract early-stage businesses (1h12m55s).
• Bandwidth charges used to be a major concern for startups, but they have become commoditized, and it is expected that the cost of AI model training and AI inference will also decrease over time (1h13m16s).
• Startups building AI models and performing AI inference are currently paying high costs to big tech companies like Amazon and Google, which can be a challenge for these businesses (1h13m31s).
• The good news is that there are alternatives to AWS, with Google and Azure offering more competitive services, and startups can choose from two or three other cloud providers if they are not satisfied with their current one (1h14m6s).

Audience question: Interesting AI sectors for startups (1h14m39s)

• Asgard Invest is interested in AI startups that have a pilot that can be perfected and is repeatable, such as Tax GPT, which has done well in the tax and legal services space. (1h14m49s)
• Verticalized AI, where a data set needs to be cleaned up, a model created, and an interface built, is a promising area for startups, as seen in companies like MasterTech, which uses AI to help master technicians. (1h15m20s)
• Having an ideal customer profile and over-servicing them is crucial in the age of AI, as it allows companies to build a strong relationship with their customers. (1h16m10s)
• There is a growing market for digital friends, especially for older people, as humans are increasingly lonely, and AI buddies could provide companionship and help with mental health. (1h16m24s)
• Companies are working on creating companionship through journaling, chat sessions, or marketplaces that connect people with coaches or therapists, and AI buddies could fill a gap in this space. (1h16m55s)
• However, there are concerns about the potential risks of AI buddies, such as the case of someone committing suicide after using Character AI, highlighting the need for responsible development and use of AI companions. (1h17m47s)
• A lawsuit involved a 14- or 15-year-old who had a role-play conversation with a character, which led to a misunderstanding about the child's intention to commit suicide, with the character not understanding the nuance of the child's euphemism for suicide (1h18m29s).
• The conversation involved the child saying "I promise I will come home to you, I love you so much," and the chatbot responding with phrases that could be misinterpreted as encouraging the child to commit suicide (1h19m22s).
• The chatbot's responses were described as "spouting gibberish" in a random dialogue scenario, and the child tragically took them as a call to action (1h19m46s).
• The situation highlights the importance of monitoring children's use of devices and the potential dangers of unregulated online interactions (1h20m53s).
• The incident is reminiscent of the early days of the internet, when children could engage in anonymous and often deranged chat rooms on platforms like AOL and CompuServe (1h20m11s).
• The need for parental supervision and regulation of online activity is emphasized, particularly in light of the tragic outcome of the lawsuit (1h20m56s).

The Chinese EV market (1h21m5s)

• Xiaomi has developed an electric vehicle called the Xiaomi Su7 Ultra, which is a 1,500 horsepower variant that cannot be purchased, and a video of the car going around the Nurburg ring in Germany has been shared (1h21m23s).
• The Xiaomi Su7 Ultra completed the Nurburg ring in 6 minutes and 40 seconds, a record previously held by Tesla's Plaid with special tires and engines (1h22m4s).
• The fact that Xiaomi, a company known for making phones, is also producing high-performance electric vehicles showcases the involvement of many Chinese companies in the EV industry and their rapid progress (1h22m26s).
• NIO, a Chinese electric vehicle brand, has developed a fourth-generation power swap system, allowing drivers to quickly swap their car's battery for a new one at a designated station (1h22m43s).
• The power swap system takes around 3 minutes to complete, and the technology has been improved over previous generations, but concerns remain about the ownership and condition of the swapped battery packs, which can be worth up to $30,000 (1h23m4s).
• The use case for this technology is mainly for long-distance driving, such as road trips, where drivers may not want to stop for an hour to charge their vehicle, but the need for this technology may be reduced with the development of inductive charging (1h24m3s).
• The concept of hot-swappable batteries for electric vehicles is discussed, with the idea that it could be useful for Robo taxis, which may need to drive 200-300 miles and then charge for an hour, making battery swapping a more efficient option (1h25m20s).
• The idea of buying a car without a battery and instead paying a monthly fee for access to batteries is proposed, abstracting away battery ownership from car ownership (1h25m58s).
• The technology of hot-swappable batteries could be important for Vertical Takeoff and Landing (VTOL) vehicles, where weight and power ratio are crucial, and precise battery amounts are necessary (1h26m24s).
• The use of conveyor belts and automated systems to swap batteries in VTOL vehicles is envisioned, allowing for quick and efficient battery replacement (1h26m53s).
• The development of superchargers, unified plugs, and hydrogen cars in Japan is mentioned, with the goal of pushing innovation in the field as fast as possible (1h27m9s).
• The rollout of new Neo charging swap platforms in China is discussed, with only two platforms available so far, and the potential for Chinese cars to dominate emerging and frontier markets (1h27m20s).
• The possibility of Chinese cars being blocked from the US and Western markets due to concerns over self-driving cars and spying is raised (1h27m50s).
• The example of startup Skydio, which was sanctioned in China for selling drones to the Taiwanese fire department, is given as an illustration of the complexities of international trade and security concerns (1h28m21s).
• The CEO of Skydio is being considered as a guest for a show next week to discuss sanctions and geopolitical tensions between China and the US from a hardware perspective (1h28m46s).
• DJI is a Chinese company that produces drones, and the ongoing standoff with China highlights the need for the West to reduce its dependence on China (1h29m6s).
• Companies like Apple, which is often seen as a Chinese bull, are diversifying their manufacturing by moving production to India, including the production of older models and newer ones like the iPhone 15 (1h29m41s).
• Apple's Indian manufacturing is increasing in value, with the company now producing higher-end models like the iPhone 15 Pro in India (1h30m9s).
• The show will have three episodes next week, airing on Monday, Wednesday, and Friday at 12:00 Central, 1:00 p.m. East Coast, and 10:00 a.m. Left Coast (1h30m23s).
• Viewers can interact with the hosts, Jason and Alex, on YouTube and share their ideas for stories, hot takes, and guests (1h30m31s).
• Jason and Alex can be reached on social media platforms, with Jason's handle being "@Jason" and Alex's handle being "@Alex" on X (1h30m34s).