May 2 2025
- Not everyone can afford private jets
- No Friedberg 😟
- Aaron Levie and Ryan Petersen join the show
- Chamath and Sacks are part of a new club in DC
- Trump first 100 days
- Hard to react for outsiders - OODA (observe, orient, decide and act) loop is hard to even do
- So many changes so fast
- Tariffs has been hard to deal with
- AI has been very positive (thanks to Sacks)
- Chamath gives B+ overall score
- Good
- Border illegal crossings are down
- Investment in US
- Controversial
- Tariffs not so bad - stock market and interest rate markets haven’t been hurt that much
- Bad
- Documents (epstein files, mlk files, jfk files (redacted))
- Tariffs communication and back and forth
- Markets have been let to be too volatile
- Good
- Crypto also seems to be tightening up
- Wokeism and DEI have also been hurt
- Reprivatization of the economy is a work in progress
- Trump has made good progress on ending the war in Ukraine
- NYT article shows America was very involved in Ukraine war
- Hard to react for outsiders - OODA (observe, orient, decide and act) loop is hard to even do
- Tariff impacts
- 60% decline in shipping from China to US!
- Businesses having to deal with lot of rule changes
- Tariffs raising prices - Amazon, Temu showing surcharges
- Seems like the administration is doing Chaos Monkey on the economy and see what breaks
- China has boxed US into bad positions - US doesn’t have strategic optionality to make the correct moral decision for the US
- There is an alternative smoother path that could have been done to get to this situation
- Where were they for 25 years?
- 2025 is the year of AI agents
- OpenAI wants to charge $20K/month for AI agents
- Manus AI agents
- Benchmark invested in the company
- Nice demo
- MCP framework growing
- Flexport is also using AI additively - its helping them make calls to truck drivers, etc
- 90% of AI use in the future will be things we don’t do today
- Agents are real but the reliability is not production ready
- Budgets need to move from experimentation to mainline production
- Dimensions of improvement
- Models/algorithms (llm, reasoning model, deep research, etc)
- Chips (better chips, hopper to blackwell, etc)
- Networked GPU clusters (100k, 300k etc - millions in the future?)
- Box did an eval
- Grok3 was the best
- New techniques - run it through the model multiple times, chunk it up, etc
- Polymarket poll says Google might have best AI model by the end of 2025
- AI coding has great progress because there is a great rapid feedback loop (code works or not)
- New applications are much more token intensive
- Will models adopted by Govt now become source of truth? Because they are from the government
Comments