13/05/2026
Earlier today, I followed all of the steps laid out in super brief YouTube video (a vertical short) tutorial on how to transfer your chat history from ChatGPT to Claude, and fortunately and surprisingly, it worked (I already posted a link to that YouTube tutorial on my feed earlier, but I'll repost a link to it in the comments on this post as well).
The neat part that I wasn't expecting was that at the end of the process, Claude gives you a concise summary of what it learned about you from doing this, and I was pleasantly surprised by what it said about me.
Here is roughly 80% of what Claude said it learned about me from reading my ChatGPT chat history:
Top of mind
Spencer has been engaged in a range of intellectually curious, research-oriented conversations spanning economics, housing policy, political science, and film. Recent topics include macroeconomics (monetary aggregates, M1/M2, FRED data), U.S. county-level income and wealth rankings, LA County housing stock and building codes in relation to wildfire rebuilds, and James Bond film plot details. Spencer also recently drafted and shared a Facebook post arguing that statistical literacy above a certain threshold is a necessary condition for consistent evidence-updating, inviting scored evaluations and editorial feedback from Claude.
Brief history
Recent months
Spencer has shown a consistent pattern of bringing real-world claims — often from social media or informal discourse — to Claude for scored, evidence-based evaluation. Examples include a macroeconomics Facebook comment dismissing M1's role in inflation (scored ~8/100), a Bing/Copilot response on congressional majority history (rated 4/10), and a statistical literacy argument Spencer authored (rated ~75/100).
Spencer demonstrated strong independent econometric intuition, correctly identifying that the Fed's May 2020 redefinition of M1 introduced a structural break creating non-stationarity in FRED's M1SL series before asking Claude to confirm. Spencer also showed interest in novel ML evaluation metrics, specifically designing a hallucination-penalizing classification metric combining False Positive Rate and False Discovery Rate as an additive penalty term to supplement F1 and accuracy in LLM training signals. Spencer frequently uses LLMs to fact-check their own work. Spencer has expressed a strong preference for epistemic honesty, sourced responses, and explicit acknowledgment of uncertainty — consistently flagging concern about hallucination across multiple conversations.
Earlier context
Spencer explored the "filtering" concept in housing economics (how new luxury units gradually become accessible to lower-income renters), and asked about the distinction between republic and representative democracy, in both cases either anticipating the answer before Claude confirmed it or requesting explicit epistemic care. Spencer also asked Claude to verify U.S. terrorism vs. homicide vs. total death statistics, again with explicit instructions to avoid hallucination and cite evidence.
Other instructions
• Spencer prefers exports and structured data in clearly separated categories and ordered formats when requested.
• Preserve Spencer's words verbatim where possible, especially for instructions and preferences.
• Spencer prefers technical discussions to use precise terminology and domain-specific language where appropriate.
• Spencer's primary analysis tools are R, MySQL, Excel, Tableau, and Power BI, with occasional Python usage.