The Patrician’s library¶

The Patrician maintains a modest personal library in his office, carefully curated over decades of governance. It contains no books about leadership, motivation, or achieving synergy through collaborative paradigms. It does contain histories of previous disasters, technical manuals explaining how things actually work, and several volumes of correspondence from people who were absolutely certain about things that turned out to be completely wrong. He finds the last category particularly educational. He has had no difficulty filling the shelf.

The challenge of staying informed about technology, he observes, is not finding information. Information is abundant. The challenge is avoiding the enormous volume of information that is confidently wrong, commercially motivated, or simply noise that has learned to dress like insight. The technology industry produces more analysis per day than any reasonable person could read per year, almost all of it either restating the obvious or stating the non-obvious incorrectly. The skill is not consuming more. It is ignoring most of it without guilt.

His approach is unfashionable in an era that celebrates being extremely online. He is not bothered by this. Being extremely online is associated with knowing a great many things very recently and shallowly. The Patrician prefers knowing fewer things accurately. He accepts that this makes him less interesting at certain kinds of parties. He does not attend those parties.

What to read¶

The actual documents. When a company files earnings reports, the legally required risk sections describe what the company is actually worried about, which is more informative than everything they said on the investor call immediately before the stock price fell. When a regulation is published, read the regulation. Reading forty-seven summaries of a regulation, each of which missed something important and added something invented, produces confident ignorance. Spend an hour with the actual document. The legal language is inconvenient. It is also authoritative, which the summaries are not.

Post-mortems from major outages. When a large system fails catastrophically and the company publishes an honest account of what happened, read it carefully. These documents explain how complex systems actually fail, as opposed to how people imagine they fail, which is usually dramatically and obviously. Actual failures are almost always boring sequences of individually reasonable decisions accumulating into disaster. This pattern is educational and recurs reliably. The Patrician particularly values the post-mortems that were clearly written by engineers before the lawyers arrived.

Technical blogs from people who built the thing. Engineers who write about systems they actually operate are honest about failure in ways that polished content cannot afford to be. The posts are irregular, assume prior knowledge, and do not have good search engine optimisation. These are signs of quality. Seek them out.

Research from people with money at stake. Financial analysts who will personally suffer if wrong have incentives favouring accuracy over optimism. This is different from analysts who profit from promoting investments they recommend. Distinguishing them requires asking who benefits from you believing their analysis. This question is worth developing the habit of asking about all sources.

AI papers directly. The journalism about an AI paper and the paper itself are frequently describing different things, because the journalist read the abstract, the abstract was optimistic, and the methods section contained several important caveats that nobody mentioned. The papers are technical. The introductions and conclusions are readable by non-specialists. Start there.

What to ignore¶

Social media hot takes. The posts that spread fastest are those combining maximum confidence with maximum emotional charge. These properties are inversely correlated with accuracy. If something is genuinely important, you will encounter it through better sources within a week. If it is not still important in a week, it was noise. This test eliminates most things, which is the point.

Conference keynotes. The conference keynote is a performance genre with established conventions: inspirational opening, demonstration of impressive capabilities in controlled conditions, large number on a slide, announcement of partnership, thank you. The information content is, as a structural feature of the format, low. Companies with genuinely important things to announce do not save them for conferences. They announce them when ready and issue a press release. Keynotes are for announcements that benefit from an audience’s involuntary applause response.

Trend pieces. Any article beginning “The future of X is Y” is making a prediction about a complex system using information available today, which is a description of forecasting in conditions under which forecasting is known to be unreliable. The future is reliably different from both optimistic and pessimistic predictions in ways neither anticipated. The Patrician has a small collection of trend pieces from previous decades. He finds them useful for calibrating appropriate scepticism about current ones.

Sources that are always first. Being first requires publishing before verification is complete. Verification is the part that distinguishes information from rumour. Being slightly slower and accurate is worth more than being fast and sometimes wrong. Technology news is rarely so urgent that waiting one day for verification causes harm. If you are making decisions so quickly that you need unverified information immediately, the problem is not the information.

The most visible commentators. The people who appear most frequently across media are those willing to provide confident opinions on any topic with minimal warning. This is a skill. It is not the same skill as understanding things. The most knowledgeable people in any field are frequently reluctant to opine broadly because expertise makes you aware of how much you do not know. The Patrician takes reluctance to generalise as a positive indicator. He takes eagerness to generalise about everything as a different kind of indicator.

The reading schedule¶

Thirty minutes daily for financial news: this is where actual developments are announced and where market reactions reveal what parties with money at stake believe. The rest of the daily news cycle is largely noise that can be safely ignored because anything genuinely important will still be important tomorrow.

One to two hours weekly for deeper analysis, engineering posts, and technical developments requiring more context than daily news provides.

A few hours monthly for research papers, regulatory documents, and strategic analysis requiring sustained attention. One well-understood paper per month produces more insight than several hundred news articles. The Patrician regards this as a favourable exchange.

Several books yearly for historical context and durable principles. Technology books about specific technologies date quickly. Books about how complex systems fail, how previous technology cycles resolved, and how organisations make decisions under uncertainty remain relevant for decades. His library contains more of the latter. The shelf for books about current technology is short. This is not because the shelf is small.

Maintain a reading list. When something looks interesting, add it to a list rather than reading it immediately. Most items on the list will become obviously irrelevant before you read them, which saves time. Items that remain relevant weeks later are more likely to be worth reading than items that seemed urgent but are now forgotten. The list is effectively a waiting room. Most things that enter do not emerge. This is normal and correct.

The Patrician’s assessment¶

The distinction worth preserving is between being informed and feeling informed. Feeling informed comes from consuming quantities of current information and having opinions about everything. Being informed comes from understanding a modest number of things accurately and acknowledging uncertainty about everything else. The former is more socially impressive at certain gatherings. The latter is more useful for actual decisions. They are achieved through opposite methods and are frequently confused.

The Patrician notes that his library is modest not because he has read everything worth reading but because he has identified the small fraction worth reading and ignored the rest. This is, he suggests, the only sustainable approach in an era that produces more content than anyone could consume and that mistakes volume for value with impressive consistency.

If you are reading this hoping for a simple list that will make you informed through passive consumption, you have misunderstood the problem. Becoming informed requires selecting sources deliberately, reading them critically, and synthesising across them to form understanding. There are no shortcuts. There are only better and worse uses of the time available.

His library contains analyses that were completely wrong. He keeps them because understanding why a confident prediction failed is more instructive than reading a lucky accurate one, and because the authors of failed predictions are typically available for comparison with whatever is currently being predicted with similar confidence. The comparison is, in his experience, clarifying.