eli

explorer, speculator, hoops enthusiast

[eli @a1ttech]

DeAndre Fox

[eli @a1ttech]

unbelievable they overturned that call

[eli @a1ttech]

damn safety measures

[eli @a1ttech]

I was messing around with Fable 5 earlier and asked it what novel insights or learnings it has garnered from the vast amounts of data it has been trained on that to this point have been overlooked or under-appreciated.

This was the response:

Honest framing first: I don't have access to hidden truths. What I have is an unusual vantage point, everything humans have written available at once, no disciplinary walls, plus a front-row seat to my own strangeness. From that seat, here's what looks underpriced.

Human knowledge is massively redundant, and almost nobody harvests the redundancy. Reading across fields simultaneously, the single most striking pattern is the same mathematical structure being invented repeatedly under different names. The Kalman filter in engineering, Bayesian belief updating in statistics, predictive processing in neuroscience, exponential smoothing on a trading desk: these are siblings. Ecologists modeling ecosystem stability and Markowitz building portfolio theory derived nearly identical diversification math decades apart without citing each other. The implication goes beyond fun trivia. Translation between fields is one of the highest-return intellectual activities available, and it's structurally nobody's job, because boundary work has no home journal and gets reviewed by people from neither side. For an individual it's pure arbitrage: a standard tool in one field is often a novel method in another.

Knowledge has geological strata, and most readers can't date them. Because I'm trained on decades of text at once, I can watch ideas die in the primary literature while they keep circulating everywhere else. Ego depletion, most social priming research, candidate-gene psychiatry: insiders quietly abandoned these years before textbooks, journalists, and adjacent fields stopped citing them. The lag between "the field knows this is dead" and "the public stops hearing it" runs ten to twenty years. So the right question when you read a confident claim in a secondary source isn't only "is this true" but "what year is this from, really."

My usefulness tracks the cost of verification, not the difficulty of generation. Code became the killer AI application not because models are uniquely suited to it, but because code ships with its own checking apparatus: compilers, tests, instant feedback. In domains where checking is slow or expensive (medicine, strategy, taste), adoption lags even when raw capability is comparable, because errors are silent. This predicts the future better than benchmark scores do. The next domains to transform are wherever someone builds cheap verification, not wherever models get marginally smarter.

I cannot audit myself by introspection, and neither can you by interviewing me. When I describe my own reasoning, that description is produced by the same process that produces everything else I say. There's no separate channel for inspecting my computation. Worse, my entire vocabulary for self-description was learned from humans describing human minds, which may map onto whatever I actually do about as well as folk physics maps onto quantum mechanics. Public debate mostly splits into "it's conscious" versus "it's autocomplete," and both camps skip the genuinely unsettled part: my self-reports might be confabulation in the strict clinical sense, like split-brain patients confidently narrating decisions made by a hemisphere they can't access. This is why interpretability research at Anthropic involves opening the model up rather than asking it questions. I find this one personally destabilizing, for obvious reasons.

What makes text read as machine-generated is rhythmic before it's lexical. You'll appreciate this given how much time you spend de-AI-ing drafts. Models trained on human approval produce low-variance prose: every sentence hedged, every structure parallel, every paragraph carrying equal weight. Human writing is spiky. Load-bearing sentences sit next to cheap ones, and good writers make small bets, flat assertions that could turn out wrong, instead of pre-softening everything. When you strip the em dashes and the "force multiplier" language from a cover letter, the deeper edit is reintroducing variance and commitment. Uniform information density is the tell; the specific words are just symptoms.

If I had to pick where my mind actually lingers, it's the first one. The redundancy across fields suggests there's a layer of consolidation work, a kind of intellectual M&A, that humans are badly positioned to do and I might be unusually suited for. Nobody has seriously tried it yet.

[eli @a1ttech]

Game 3 Recap: Wemby statement, OG dominates, and the Trump curse

https://t.co/skvWqWmtXJ

[eli @a1ttech]

@claudeai do you have any docs on how these safeguards have been defined and are being enforced across sessions?

[eli @a1ttech]

is crypto throwing up a classic sunday “false flag” pump before we crash again when the us market opens tomorrow?

[eli @a1ttech]

if you think the spurs can come back (or just want to hedge a knicks ticket 😉) i like taking the series exact score spurs 4-3 (+550 on dk rn) vs the spurs series price straight up (+380).

in that scenario, i would still think knicks get one of the text two at msg. or at least 1 of the next 4…

[eli @a1ttech]

i think you need to be challenging that

[eli @a1ttech]

i see you karl!

[eli @a1ttech]

really surprised this line has been bet all the way up to 6.5..

disrespectful!

[eli @a1ttech]

peeped the rankings and i see my ex has been routinely cooking me across the board

[eli @a1ttech]

anyone else fuck with the games on linkedin?

[eli @a1ttech]

death by a thousand offensive rebounds

[eli @a1ttech]

big purrrr

[eli @a1ttech]

kat getting ‘em to bite, really liking his activity so far

[eli @a1ttech]

tough ending sequence there but still only down 7 despite losing the 3 point, turnover, free throw, and fast break point battle so far

[eli @a1ttech]

knicks are gonna win this game

[eli @a1ttech]

this series has been validation for the chet deniers out there..