Timeline reactions: AI consciousness, Apple's Siri delay, Anduril's double-death fighter jet, and startup advice
Mar 4, 2025
Key Points
- OpenAI's ChatGPT ranks #1 on iOS App Store despite OpenAI, Grok, and DeepSeek all subsidizing inference at a loss, while Jimmy John's toasted-sandwich launch lands at #2, illustrating that markets reveal demand for tangible products over subsidized AI.
- Prominent founders including Warren Buffett, Jensen Huang, Jeff Bezos, and Brian Armstrong worked at excellent organizations before building dominant companies, challenging the myth that the best founders never worked for anyone else.
- Fred Wilson reframes venture capital as a service business where VCs support founders through cycles of brilliance, anxiety, and fragility rather than directing strategy, with founder psychology and emotional support mattering as much as capital.
Summary
AI consciousness and model architecture
Consciousness in AI systems might emerge not as a unified model but as multiple specialized systems trained separately—language, vision, driving—that interact like different lobes of the brain. Current image generators like DALL-E and ChatGPT cannot generate a wine glass filled to the brim because training data contains no such images. This reveals a hard limit in interpolation. Humans can imagine a unicorn by blending two known concepts. Models struggle to synthesize beyond their training distribution. The industry trains separate models for different modalities when true AGI might require integrated architecture.
OpenAI's routing problem
ChatGPT still lacks a clean router across OpenAI's own models. Users push a search button inconsistently. The app sometimes searches unprompted. OpenAI should ship a single interface that intelligently routes to the best model without user intervention. OpenAI is speedrunning Google's 20-year descent from clean search to an ad-cluttered, algorithm-gamed mess, but doing it in months.
App store rankings
ChatGPT ranks number one in the iOS App Store, ahead of Jimmy John's at number two. The sandwich delivery app achieved such prominence apparently on the back of a recent product launch: toasted sandwiches. Markets reveal what people actually want. Food beats intelligence.
Founders and venture capital
Fred Wilson reframes venture capital as a service business to founders. VCs are service providers writing on founders' coattails. Founders are simultaneously charismatic, brilliant, frustrating, anxious, and fragile. A VC's job is to meet them where they are and support them fully.
Keith Rabois recently argued that founders should work at an outstanding organization first to learn what greatness looks like before starting. The post sparked controversy, with some arguing nobody they respect ever worked for anyone else. Counterexamples include Warren Buffett, Jensen Huang, George Soros, Peter Thiel, John Carmack, Jeff Bezos, Larry Ellison, Mike Bloomberg, Phil Knight, Brian Armstrong, and Benny Hsu. All worked at significant organizations before founding or leading companies that became dominant.
If you lack conviction, go learn from excellent people at excellent companies. This compresses your learning curve. If you are already burning to build something, you won't ask and shouldn't wait. Eric Yuan at Ramp built a fintech company first, then worked at Capital One, then built Ramp with relationships and insights he would never have had without that interval. The sequence mattered.
Founder psychology illustrates the brittleness Wilson describes. A CEO who wrote terse, jargon-free emails became a phishing vulnerability because his writing style was indistinguishable from scammer emails. Style and warmth matter. Extreme curtness becomes a security liability.
AI safety and GDP
Yasen Yaken argues that the only AI risk worth worrying about is missing out on AI wealth. AI safety work like evaluating jailbreaks and open-source model integrity is a reasonable use of resources but shouldn't distract most people from building profitable AI companies. The concern isn't overnight paperclip maximizers but realistic threats like Stuxnet embedded in open-source models.
Skepticism abounds about whether AI is actually generating real GDP growth. The U.S. has experienced productivity stagnation since the 1970s despite the internet. The Atlanta Fed is reporting 2.5% negative GDP growth. AI labs are subsidizing inference, which boosts user adoption but doesn't necessarily translate to economic value creation. Spending on Character AI might siphon money from the real economy without generating offsetting productivity.
Founder visibility
If a founder was in Aspen in February, you can write the investment to zero. February is not a serious month for venture capital. Appearing there suggests a founder isn't grinding hard enough. Early-stage founders should appear obsessed, frustrated on weekends that everyone else is offline. Optics matter more for early-stage than for later-stage companies.
GPT-4.5 on consciousness
Sam Altman posted a screenshot of a conversation with GPT-4.5 on consciousness. When asked to rely only on first principles, the model concludes that consciousness alone truly exists and matter is merely an internal experience within consciousness, not independently real. Dreams, simulations, and imagination prove consciousness can generate structured experience without external materialism. The model opts for Berkelean idealism. Consciousness is fundamental. The material universe is inferred and unnecessary. The simulation hypothesis maps cleanly onto this framework.
Sports betting in Africa
GDP per capita in most African countries is roughly $2,000. Average yearly revenue per user from sports betting platforms in Africa is also roughly $2,000. People engaging in sports betting are economically well-positioned. Sports betting platforms extract significant value from African users.