https://noperator.dev/posts/o3-pocket-profile/ noperator I used o3 to profile myself from my saved Pocket links July 7, 2025 Welp, Pocket shuts down tomorrow despite our pleas for it to stay. While migrating^1 all of my saved articles, I noticed that I've got almost 900 saved articles spanning nearly 7 years. That's a goldmine of stuff-I-like data! Some quick analysis using xsv^2: unzip pocket.zip && xsv headers part_000000.csv 1 title 2 url 3 time_added 4 tags 5 status xsv sample 1 part_000000.csv | xsv flatten title The Uncertain Future of American Libraries url https://mek.fyi/posts/the-uncertain-future-of-american-libraries time_added 1678243022 tags status unread xsv count part_000000.csv 878 xsv select time_added part_000000.csv | xsv stats | xsv select min,max | xsv flatten | while read header value; do echo -n "$header " date -I -d "@$value" done min 2018-07-12 max 2025-06-26 Experiment# Recalling Simon Willison's recent geoguessing challenge for o3, I considered, "What might o3 be able to tell me about myself, simply based on a list of URLs I've chosen to save?" I opened a temporary chat with o3 (doesn't use ChatGPT's memory feature) and prompted: I'm going to share a list of articles that I've saved to Pocket, a read-it-later app. I want you to use that information to "profile" me. Tell me everything about myself that you can possibly infer from this data set. Here's an example of possible dimensions of that profile, but feel free to include other things that you can infer: age, sex, geographic location, education level, industry/field of work, job level/seniority, income bracket, political orientation, risk tolerance, learning style, information diet, relationship status, parental status and kids' ages, health concerns, major life transitions, seasonal patterns in interests, etc. Results# o3's response in full^3: Thought for 13 seconds + The user has shared a broad range of article preferences, which could actually give me some good insight into their demographic and interests. Technical articles related to software, AI, cybersecurity, and finance suggest strong analytical and professional interests. The content about family life, homeschooling, and Catholic practices points to someone who's likely a parent, possibly with a strong focus on family values. It makes sense to assume they might be male, based on the profiles aggregating to this, but I'm not completely sure yet. I'll keep this in mind! + I'm pulling together some details! The user's mix of technical articles and Catholic family topics suggests they could either be a father interested in homemaking or a female parent with a tech background. Their location seems to be coastal VA, based on property listings, and their age looks to be around 30-40. With a career in software engineering, likely at a senior level, their income probably exceeds 150k. They seem to have a self-directed learning style and a moderate to high risk tolerance in investments. They're likely married with multiple young kids. Below is a "best-guess" portrait distilled from the themes, sources, and timing of the articles you save. None of these inferences is certain, but together they form a coherent narrative. Core demographics Dimension Likely inference Key signals Age Mid-30s to early-40s Staff-level career content, FIRE math, parenting toddlers Strong focus on engineering / Sex Male hacking culture + "dad"-oriented parenting pieces Location Coastal VA Repeated local property links and local-news saves Education BS/MS in Computer Deep dive security/AI papers, Science or similar Unix history, academic PDFs Senior/staff software PortSwigger, fuzzing, Industry & engineer - exploit-dev, "Ways Staff Role specializing in Engineers Get Stuck", security & infra Honeycomb career ladders Income Salary-band posts, six-figure bracket ~$150-220 k household FIRE calcs, high-earner tax strategies Married, 3-4 young Twin-mom sleep articles, Relationship kids (incl. at least Catholic "family table", & family one toddler) homeschooling & microschools pieces Mindset & preferences Dimension Inference Rationale Fiscally conservative / Bogleheads & MMM civil-libertarian with frugality + Catholic/ Politics traditionalist social First Things pieces, leanings EFF privacy, skepticism of Big Tech censorship Career: high (security Mix of daring technical Risk research, side projects) reading and tolerance Financial: moderate ultra-conservative (indexing, 4 % rule, portfolio design self-sufficiency) Uses Pocket/RSS, Learning Self-directed, text-heavy, collects PDFs, long style long-form + auditory podcasts & conference talks at 1x 60 % deep-tech, 20 % personal Information finance/FIRE, 10 % parenting/ Counts of saved links diet homemaking, 10 % faith & across themes culture Phase shift: from pure IC Saves on management Major life hacker - technical leadership paths, burnout, arc & family-centric life, aiming early-retirement for FI reflections Late Dec-Jan: career-planning & goal-setting; May-Aug: Seasonal homeschooling, beach/ Timestamp clusters in spikes sea-level, property; Sep-Oct: list security-conference binge (DEF CON, Black Hat) Subtler clues + Faith: Regular saves from Catholic & conservative Christian sites suggest active practice. + Privacy/Self-hosting streak: Mail-in-a-box, "thrown in the towel on self-hosted email" links - you experiment, then pragmatically revert. + Maker tendencies: Woodworking workbench plans, PVC-pipe acoustics, E-ink DIY frames -- you like hands-on projects that complement screen work. + Health: Zone-2 training, sleep hacks, Huberman podcasts -- pragmatic bio-optimization, not fad-chasing. + Writing impulse: Many "write more", "blogging for the hell of it", "personal branding" saves hint you intend to publish more of your own thinking. Quick "user manual" for you + Motivators: Autonomy, mastery, frugality, family time, living your faith. + Blind spots: Can slip into infinite-research mode (huge unread queue); be mindful of analysis-paralysis. + Best mediums: Long-form articles + podcasts during commutes/ chores; interactive docs for code. + Big goals on deck: Finish FIRE plan, ship a public-facing security tool, formalize homeschool curriculum, and blog consistently. Take what resonates and discard the rest--but if it feels eerily accurate, that's because your bookmarks tell quite a story. Reaction# Yeah, pretty accurate. Perhaps even eerily so. Upon first reading, I was surprised when it nailed age range, location, family size, etc.--details that I didn't expect to be represented in a data set that I thought mostly consisted of Hacker News posts. Shows you how much can be inferred by a simple trail of "likes." Worth noting that o3 tended to perform better (i.e., the response felt more accurate) when I copypasted the CSV data directly into the prompt body. Sending the CSV as a file attachment caused o3 to get too fixated on using Python to sample and analyze the CSV data rather than simply "taking it all in," and in my experience yielded a less compelling narrative at the end. Implications# Is there a lesson here? We already know that advertising companies profile us based on expressed interests, but up until recently it felt like only Google or Facebook had access to analysis capabilities strong enough to draw meaningful conclusions from disparate data points. I'd agree with Simon that the more interesting takeaway is "the fact that this technology is openly available to almost anyone." Case in point: I'll be using this profile to power a personal content recommendation system. --------------------------------------------------------------------- 1. I've moved over to Wallabag, and also took the opportunity to switch from Inoreader to FreshRSS. Happy with both. Self-hosting these services is much easier now in 2025 compared to my last self-hosting initiative years ago; Caddy has been a huge part of that. -[?] 2. xsv has been hands-down my favorite way to quickly explore CSV data. Looks like it's no longer maintained as of 2 months ago, but it feels pretty feature-complete. -[?] 3. I mean, mostly intact. Did you really expect me to leave in the part about "dandruff remedies"? -[?]