I Asked AI What to Do With the Next 5 Years of My Life
In a Nutshell
AI retrospective analysis of personal decisions reveals memory is consistently 20% inaccurate and startup stories become heavily sanitized over time, while AI tools now enable 20-year decision audits in hours rather than years. AI excels at refining existing ideas and cross-referencing health data but struggles generating original concepts, and its productivity gains mainly accelerate low-value work. The most valuable application is asking AI open-ended questions about five-year career and life directions with full conversation history, producing high-quality suggestions that match friend-level insights.
These notes were generated by AI and may contain inaccuracies.
Using Claude code with Ferris APIs to conduct a 20-year retrospective analysis of angel investing. The analysis examines who made what introductions, which companies were not replied to that ended up being successes, and which companies were turned down that ended up being really important. The goal is to test personal stories against actual data rather than relying on remembered narratives about why certain decisions were made.
A wearable device categorizes AI and listens 24/7, revealing that people typically know about 70% of what they think they know. The device demonstrates memory discrepancies, such as when someone requests a dark chocolate bar at 7:00 p.m. but the recording shows it was requested at 5:00 p.m. and specified as milk chocolate. This reveals that memory is consistently about 20% off from actual events.
Startup genesis stories undergo significant sanitizing and editing over time. Founders work on material for so long that they come to believe their edited version is the actual truth. This applies to both public storytelling and private self-narration, where individuals are not exempt from this memory editing process.
LLMs provide significant value in holistic health management by synthesizing data across medications, supplements, predispositions, side effects, and historical health events. The speed and depth of analysis allows for interrogation of complex health questions. The primary benefit has been avoiding potential disasters by checking contraindications between substances and determining whether symptoms A, B, or C could explain symptom D.
To minimize AI hallucinations in health-related queries, responses are fact-checked across multiple LLMs rather than relying on a single source. This cross-verification approach helps validate information while acknowledging that complete elimination of errors remains impossible.
Investing serves as an objective way to evaluate thinking and decision-making quality rather than simply amassing wealth. The scorecard approach allows fine-slicing of decisions to determine whether thinking was sound. Unlike vague questions like "Was I thinking well last month?", investment outcomes provide concrete data points for analysis.
Running counterfactual scenarios helps examine alternative outcomes: what if the opposite decision was made, what if investments weren't sold, what if different choices were pursued. This intellectual exercise provides insight into decision quality even when simpler approaches like investing everything in the S&P 500 might be more financially optimal.
AI enables rapid automation of workflows, but this acceleration of doing things well does not necessarily make those things important or worth doing. The volume of low-value work being completed efficiently is increasing significantly, while genuinely valuable applications remain limited.
A 20-year retrospective analysis that would require a year of full-time work with multiple people can be completed in a few hours using Claude code with Gmail API. The output quality exceeds manual capabilities, though the full potential of these tools remains unexplored.
Training AIs on personal writing creates highly capable writing assistants that can produce in 30 seconds what would take 30 hours manually. This capability comparison, similar to how AlphaGo's victory affected professional Go players' motivation, creates a net negative impact on creative drive. The comparison to AlphaGo illustrates how competing against machine output can drain motivation for putting in extensive manual effort.
Current AI systems excel at refining and expanding existing ideas but struggle with generating original thesis statements or novel concepts that would engage audiences. While AI can improve copy and suggest expansions, the generation of core creative concepts remains a human strength.
The method involves asking AI models (Claude, ChatGPT, etc.) questions like: "What do you think I should do in the next 5 years? What might be some rewarding paths of exploration?" or "What are three to five ideas that you think could be rewarding career exploration for me in the next X period of time?" The AI should have sufficient conversation history, with cross-conversation memory enabled for holistic understanding.
When provided with comprehensive personal context across scientific interests, investing, writing, relationships, and organizational activities, AI generates outstanding suggestions that friends rate as high quality. The suggestions include both business ideas and non-revenue philosophically aligned concepts that inform major life decisions.
Regular podcast listeners develop deep knowledge of hosts through consistent exposure, demonstrating that machines with perfect memory can achieve even greater understanding. This creates both opportunities and concerns about the depth of AI knowledge about individuals.
Better AI responses come from asking open-ended personal questions similar to those posed to close friends, rather than queries more suitable for robotic processing. Questions like "What do you think are three to five creative ways I might explore things professionally in the next 5 years?" yield more interesting and applicable responses than narrow, technical prompts.
Keep Tim Ferriss in your library
Save the videos and channels worth coming back to, and find them again in one place.
