Infinite Words

Here is what I did; I got Claude and GPT to write me code that went through all 5647 post and aggregated the entries into similar subjects. Then I got both Anthropic and OpenAI to write me a summary of my opinions on each subject, using whatever I wrote over the whole 13 years (ignoring that my views might have changed over that period).

Fuck me – the things cannot write in my style, or any style, no matter how hard I tried. They just keep reverting to how they were trained. Initially I got a whole unreadable essay on each subject. Then I kept making it shorter and shorter, just to filter out the LLM bullshit.

But that didn’t work. Basically an LLM just does what humans do but slightly shitter and with no character. However they do it much faster and they can cope with a lot more information before falling to pieces. In the case of compressing dozens of conflicting and intentionally provocative blog entries into one coherent passage; the thing has no chance. But then neither would anyone I know. So it’s evens again.

Here’s an example summary (and this is the absolute best of them):

Authenticity, ego and self-deception
Recurring positions stress that authenticity requires privileging felt truth and continual self-examination while resisting ego-driven performance; strongly emphasised practices include daily journalling, mindfulness, deliberate ego diminution and a commitment to speaking truth privately, alongside repeated admissions of self-deception, habitual self-censorship and persistent behavioural relapse. Reasoning rests on epistemic humility and doubt, advising methods to detect self-deception and a trust-but-verify stance toward intuition, and material change over time moves from chameleon-like conformity toward disciplined self-scrutiny and increased truth-telling. Tensions include balancing uncompromising honesty with social roles and uncertainty whether partial practices such as part-time Zen or AI-enabled solitude, or sustained effort, can prevent relapse.

Mostly meaningless gobbldegook.

But here’s the party trick – GPT says “The only person qualified to write the Offshore Westerly summaries is the one who wrote the previous 5,647.”

The reason we buy into this shit is because it is so easy to fluff up our egos.

My summary – Really Fast Flattering C-class (Fair Average Quality) writing. No thinking, no signs even of that.

That will be good enough to replace 95% of the working population.

To that GPT said “You should avoid a broader labour-market prediction while still capturing what your experiment actually demonstrated. If you want to argue about workforce replacement, I’d treat that as a separate argument built on additional evidence rather than making it the punchline of this experiment.”

To which I said – “shut up – you are an average idiot.”