By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
a16z2023 Big Ideas in Technology (Part 1)
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 2:37
Series Overview and Themes of Part One
Steph Smith explains that a16z partners were asked for one big idea startups could tackle in 2023, producing 40-plus ideas, and that this two-part series covers 12 of them. Part one spans consumer, AI, games, and enterprise, with recurring themes of physical-to-digital transfer, next-wave AI, fault tolerance, social commerce, and bots.
- •Partners each spotlighted one big idea for 2023, yielding 40-plus ideas
- •Part one covers consumer, AI, games, and enterprise with six partners
- •Themes include physical-to-digital, fault tolerance in AI, social commerce from the East, and bots
- •The content is informational and not investment advice
- 2:37 – 10:40
Social Commerce: Video and Live Shopping
Connie Chan presents social commerce as inevitable, then explains why Western platforms still have too much friction between inspiration and checkout. She argues live video builds buyer confidence, community, and time-limited promotions, like the infomercials that worked.
- •Western platforms break up the path from inspiration to purchase, sending people to other sites
- •Live video enables direct questions to sellers, community, and limited-time discounts
- •China built live shopping and its supporting ecosystem years ago, while Whatnot shows early US traction
- •Western shopping is search-based rather than discovery-based; Shein and Temu push discovery
- •Social, entertaining shopping like carts shared with friends has existed in China for years
- 10:40 – 16:58
Why the West Lags: Copycat Platforms and Mobile-Only Markets
Steph Smith asks why social platforms copy one another and why Eastern ideas take so long to arrive. Connie Chan cites a super app mentality, top-down organizational change with near-term P&L hits, and the US being mobile-first rather than mobile-only. She expects more experimentation in 2023 and possibly a new category of creators who are good at selling.
- •Platforms copy proven features because they already have distribution
- •Ideas from Asia usually take at least a year, generally three to five years, to catch on in the US
- •Pushing shoppable live video takes space from revenue-generating ads, so the change must come from the top
- •Developing countries are mobile-only, while the US stays PC-oriented and backwards compatible
- •In 2023 platforms and brands will experiment more, and good sellers may form a new creator category
- 16:58 – 25:44
Unlocking the Third Space: Definition and Digital-Native Advantages
Anne Lee Skates introduces Ray Oldenburg's third space and the need for tools serving online-first communities. She defines third spaces physically and psychologically, notes their decline, and describes what digital unlocks: a global pool of like-minded people and a new notion of presence, using Whatnot and Dungeons & Dragons as examples.
- •Third spaces are not home or work and let people explore identity freely and safely
- •An Atlantic stat shows people with a favorite local place dropped from two-thirds to a little over half
- •Online access to a global pool lets niche fans find each other
- •Whatnot adds gamification via live auctions and lets people jump between streams
- •Digital presence lets you change environment, mood, and appearance, as seen in RPGs
- 25:44 – 30:39
New Modes of Online Connection and Blending Digital with Physical
Steph Smith and Anne Lee Skates discuss activities beyond gaming, such as interest-based communities and generative AI art for friends and pets. They then look at transfer from digital back to physical, including VR math learning and using digital data to enrich in-person gatherings.
- •Whatnot, TikTok, Reddit, and Discord facilitate stranger-first, interest-based relationships
- •Lensa AI lets people make generative art for friends, family, and pets
- •Prisms VR classrooms saw advanced kids more eager to help struggling peers
- •Pokemon Go was the last mass-market product prompting in-person interaction
- •Digital data could enrich in-person meetups, e.g. connecting fans at concerts
- 30:39 – 37:20
Games as a Neverending Turing Test: Can We Trust AI?
Jack Soslow explains that bots in games are becoming neural-network-based and increasingly indistinguishable from humans. In discussion with Steph Smith about ChatGPT, he says the Turing test is a flawed, binary metric that tests deception rather than understanding, and explains why he mostly trusts AI while validating results.
- •Bots already exist in games like Words With Friends, Call of Duty Mobile, chess, and RuneScape
- •Startups like Inworld AI, Convai, and Charisma.ai build in-game agents with conversations and objectives
- •The Turing test tests deception, not understanding; ChatGPT can be confidently wrong, as in multiplication
- •Trust depends on training data and filtering layers, which are opaque
- •Jack trusts AI for most things, validates elsewhere, and is encouraged by open source and model improvement
- 37:20 – 44:50
Designing Games Around Bots: Companions, Personalization, and Chess
Jack Soslow and Steph Smith discuss whether bots ruin games. Jack argues games should be designed with bots in mind, distinguishes creator-made bots from player hacks, and imagines believable NPCs and personalized multiplayer with close-win bots. Steph points to chess and Fischer Random Chess as humans adapting.
- •Games are structured play, and bot bans become a neverending arms race
- •Believable NPCs and companions could make single-player worlds richer
- •Personalized multiplayer with AI opponents can deliver close wins and flow states
- •A bot is made by the game creator, while a hack is made by a player to bypass rules
- •Chess and Fischer Random Chess show humans adapting, and separate human-only servers could be offered
- 44:50 – 47:09
2023 Outlook for Games: Foundation Models and a Creation Renaissance
Jack Soslow says he is excited about 3D, audio, and video foundation models arriving in 2023 and working together, since game assets combine models, sound, and code. He and Steph Smith compare this to easy online publishing, imagining non-gamers spinning up games.
- •Image models were largely solved last year and text keeps improving
- •3D, audio, and video foundation models are expected next
- •Game assets link 3D, sound, and scripts, so connected AI could abstract away rote work
- •Like online publishing, game creation could become accessible to non-experts
- •The real skill will be knowing what a good game looks like
- 47:09 – 57:36
The Metaverse Goes Fashion Forward: Skins and Brand Partnerships
Doug McCracken argues Gen Z and Gen Alpha value digital self-expression and will reward brands that embrace digital fashion. He gives examples of skin economies in Fortnite and League of Legends, brand partnerships like Balenciaga and Louis Vuitton, and tactics like scarcity and feature differentiation.
- •Character skins are big business despite no gameplay benefit
- •Fortnite, League of Legends, Minecraft, GTA Online, Roblox, and Animal Crossing have skin or item ecosystems
- •Luxury brands like Balenciaga and Louis Vuitton partnered with games
- •Brands treat partnerships as marketing now, with scarcity and feature differentiation like DJ Sona skins
- •Fortnite earned over $2 billion in multiple years, more than half from cosmetics
- 57:36 – 1:00:38
Interoperability and 2023 Hopes for Digital Fashion
Steph Smith pushes Doug McCracken on why games remain siloed. He cites technical hurdles and game moats, but points to avatar tech like Ready Player Me and kids' expectations from Roblox, predicting interoperability will grow. He hopes to see more digital-only fashion businesses in 2023.
- •Standards and technical hurdles plus entrenched games limit interoperability
- •Ready Player Me has 6,000 developers signed up for cross-game avatars
- •Kids raised on Roblox expect to customize avatars across experiences
- •Interoperable and non-interoperable futures can coexist
- •Digital-only fashion businesses like RTFKT and DressX and digital-to-physical trends are exciting
- 1:00:38 – 1:06:36
Generative AI Beyond Text-to-Image: Workflows and Fault Tolerance
Sarah Wang presents AI advancing from images to complex workflows such as text to SQL and Excel modeling. In response to Steph Smith's fault tolerance framing, she explains that AI-assist workflows with humans in the loop, like code completion, video editing, and pitch decks, can deploy immediately, unlike high-stakes areas like autonomous vehicles.
- •AI could bring 10 to 100 times performance improvements to enterprise productivity
- •Fault-tolerant creative tasks adopted AI first, while autonomous vehicles have the highest stakes
- •AI assist keeps humans in the loop so it can be deployed now
- •Coding, Runway video editing, and pitch creation are early examples
- •Fully automated results may come later
- 1:06:36 – 1:13:57
Foundation Models, Fine-Tuning, and Differentiation by Owning Workflows
Sarah Wang discusses whether the future involves a few horizontal LLMs or many fine-tuned models, calling it use-case dependent. She argues differentiation comes from owning sticky workflows and proprietary data, as in the cloud shift, and that existing companies should add AI. She also names data workflows as a 2023 opportunity.
- •Fine-tuning with proprietary enterprise data improves accuracy for B2B
- •Capital demands may lead to an oligopolistic structure like the big clouds
- •Salesforce, Workday, and ServiceNow won the cloud shift through useful, sticky workflows
- •One-click add-AI features are novel but not sticky
- •AI in data workflows is a promising near-term area
- 1:13:57 – 1:16:19
LLMs in Fintech: Fault-Tolerant Uses and Trust
Sumeet Singh describes fintech balancing new LLM rails with customer trust. He says early uses fit fault-tolerant areas like customer engagement, with human agents as backup, while underwriting and accounting need to be 100% correct.
- •LLMs may speed dataset training and automate insurance claims and loan origination
- •Complex uses like automated investment decisions need reserves of user trust
- •Customer engagement tools have fault tolerance because human CX agents remain
- •Automated underwriting and accounting require accuracy, with IRS fines as a risk
- 1:16:19 – 1:24:01
Transparency, Regulation, Distribution, and Back-Office Automation
Steph Smith and Sumeet Singh discuss how transparent founders should be about LLMs, possible regulatory scrutiny from bodies like the CFPB, and the importance of distribution as a moat. Sumeet expects near-term gains in back-office automation, favoring automating 100% of a small slice of a workflow.
- •LLMs may be back-end, with consumers never seeing the interface
- •Regulators could act if automated loan decisioning or tenant screening uses discriminatory data
- •Product leaps will likely come from tuning large models, but distribution stays key
- •Back-office automation for SMBs is low-hanging fruit
- •Better to automate 100% of 5% than a little of 100% of a workflow
