Marcelo Rinesi
marcelorinesi.bsky.social
Marcelo Rinesi
@marcelorinesi.bsky.social
Cognitive architecture designer and consultant.

On https://rinesi.com there are links to my blog and newsletters (one for articles, "what's new on arXiv," etc, the other for original short-short SF).
From a cognitive economics POV, one interesting effect of cheap or subsidized LLMs in commodified activities is that they set a lower bound on quality and an upper bound on price. Whether this is good or bad depends on the status quo + the bounds + socioeconomic dynamics... 1/
via @tedunderwood.com
Pretty soon GPT X Pro will not only write a better paper review than most reviewers, but it’ll spend more time on it too.
January 4, 2026 at 5:45 PM
"There's a human-shaped hole in the secret memory of the world. Somebody who walks without footprints and stands without shadow. No record no photos no financial or social ripples. Missing from high resolution satellite cameras..."

#ShortFiction #ScienceFiction #AdversarialMetanoia
The ECHELON Lacuna
There's a human-shaped hole in the secret memory of the world. Somebody who walks without footprints and stands without shadow. No record no photos no financial or social ripples. Missing from high…
blog.rinesi.com
January 4, 2026 at 2:06 PM
I intended to write this one, by sad necessity, in something closer to LinkedIn-ese than I'd prefer. I didn't get even in the vicinity of that register, but it's probably around 37% less openly sarcastic than my internal phrasing, so there's that.
The Post-AI Organization
blog.rinesi.com
January 4, 2026 at 5:12 AM
Argentina is likely a relative outlier but I follow the local equivalent of the MSM and the coverage is relatively positive [not unrelated to how much the Kirchnerist government aligned with Chavez; things are seen thru the prism of post-2003 Argentinean politics]

via @deanbaker13.bsky.social
So, I’ll be in Brasil in 3 weeks - gonna be a fascinating time to be an American reporter in Latin America
January 4, 2026 at 1:17 AM
"There's a human-shaped hole in the secret memory of the world. Somebody who walks without footprints and stands without shadow. No record no photos no financial or social ripples. Missing from high resolution satellite cameras..."

#ShortFiction #ScienceFiction #AdversarialMetanoia
The ECHELON Lacuna
blog.rinesi.com
January 3, 2026 at 10:40 PM
[In an effort to distract myself for the Ongoing(s)] #ScienceFiction #ShortFiction #AdversarialMetanoia posted during December '25:

"The best-known art forgers are driven by ego; the best, period, aren't known." 1/4
Invisible Art
blog.rinesi.com
January 3, 2026 at 10:10 PM
"A black hole in his soul?"

"A self-reinforcing memetic attractor, if you prefer. Plot it in semiotic space and it looks pretty much like a Kerr–Newman metric...."

#ShortFiction #ScienceFiction #AdversarialMetanoia
Accretion
"A black hole in his soul?"
blog.rinesi.com
January 1, 2026 at 4:13 AM
"A black hole in his soul?"

"A self-reinforcing memetic attractor, if you prefer. Plot it in semiotic space and it looks pretty much like a Kerr–Newman metric...."

#ShortFiction #ScienceFiction #AdversarialMetanoia
Accretion
"A black hole in his soul?"
blog.rinesi.com
December 31, 2025 at 1:15 AM
All valuable things, but an LLM-intensive process probably needs emphasizing DSL design (the closer to the domain the LLM's target, the better/safer the generated code) and information architecture (the last thing you want is for nobody in your org to know how things work).

via @tedunderwood.com
What does a class on software development post LLM contain?

Design?
Syntax of common langs?
Programmer thinking?
Environments/libraries/bash?
Debugging?

References and texts especially welcome
December 30, 2025 at 10:33 PM
I agree; I think echoes Peter Naur's "Programming as Theory Building." You can't really develop, maintain or extend a non-trivial piece of software without a group that's knowledgeable *specifically* about it, and that knowledge takes time and apprenticeship.

via @kevinr.free-dissociation.com
The thing that makes great, lasting software is teams that come together around it, and STAY TOGETHER around it. They put in the effort. They improve the system incrementally. They have the skills at the start to make flexible and good decisions.
December 30, 2025 at 6:32 PM
"A black hole in his soul?"

"A self-reinforcing memetic attractor, if you prefer. Plot it in semiotic space and it looks pretty much like a Kerr–Newman metric...."

#ShortFiction #ScienceFiction #AdversarialMetanoia
Accretion
"A black hole in his soul?"
blog.rinesi.com
December 29, 2025 at 5:17 PM
There's a plausible correlation there via governments leaning into exports to compensate for demographic headwinds. See e.g. Krugman here on China, although specific mechanisms differ.
December 29, 2025 at 6:08 AM
"A black hole in his soul?"

"A self-reinforcing memetic attractor, if you prefer. Plot it in semiotic space and it looks pretty much like a Kerr–Newman metric...."

#ShortFiction #ScienceFiction #AdversarialMetanoia
Accretion
"A black hole in his soul?"
blog.rinesi.com
December 28, 2025 at 2:06 PM
TIL about the concept of tripwire deployments.

"the purpose [...] is not to win or [...] slow down an enemy, but to die in the opening hours of any conflict, functionally guaranteeing it is politically impossible for distant powers to remain neutral" is one heck of a phrase, concept, and job.
Security guarantees in practice are important to the degree that they enforce costs if the agreeing party renegs. There's really two options: membership in larger mutual defense agreements or tripwire forces (or both).

Were I in Ukraine's position, I'd be willing to sacrifice a lot for tripwire.
Still kind of baffled by the security guarantees talk, US and EU - people who’ve been watching russia break every single law, kidnap children, loot, rape, torture and kill en masse, bomb cities and blow up dams, occupy NPP for years.

Those people are going to 100% help next time, pinky promise?
December 28, 2025 at 12:32 AM
While I don't have yet any view on how well it works in practice, the underlying idea is at the very least interesting (although I'm biased by information theory being my first mathematical love).
Information-theoretic signatures of causality in Bayesian networks and hypergraphs
Analyzing causality in multivariate systems involves establishing how information is generated, distributed and combined, and thus requires tools that capture interactions beyond pairwise relations. H...
arxiv.org
December 27, 2025 at 10:43 PM
Useful on its own - also a good example of what looks like a generally successful pattern to deploy LLM-style models: "funneling" sampling through reliable domain-specific filters. LLMs can [sometimes] be great at generating candidate [near-]solutions, but it's hard/impossible to make them reliable.
Propose, Solve, Verify: Self-Play Through Formal Verification
Training models through self-play alone (without any human data) has been a longstanding goal in AI, but its effectiveness for training large language models remains unclear, particularly in code gene...
arxiv.org
December 27, 2025 at 10:41 PM
If there's an aesthetics of the intersection of CS and physics --and I think field-specific aesthetics is an insufficiently explored area in mathematics, programming, and data modeling-- this paper is surely in the purest Classical tradition.
Classical billiards can compute.

With Isaac Ramos, we show that 2D billiard systems are Turing complete, implying the existence of undecidable trajectories in physically natural models from hard-sphere gases to celestial mechanics.
Determinism ≠ predictability. 🎱🧠 @upc.edu @ricardsole.bsky.social
December 27, 2025 at 10:35 PM
"A black hole in his soul?"

"A self-reinforcing memetic attractor, if you prefer. Plot it in semiotic space and it looks pretty much like a Kerr–Newman metric...."

#ShortFiction #ScienceFiction #AdversarialMetanoia
Accretion
blog.rinesi.com
December 27, 2025 at 8:27 PM
The US has all the legal and some of the operational infrastructure for a win-all-elections/disrupt-significant-opposition soft authoritarian/hard ethnonationalist state. Long-term optimism hinges on the balance between Trumpian incompetence and Democrat passivity. 1/

via @greatdismal.bsky.social
Filming them committing terrorism is now itself an act of terrorism?
December 27, 2025 at 7:49 PM
As somebody who has both led and been an analyst in a ton of data analysis projects from "tweak this metric" to "report to the CEO," I went thru this thread nodding to hold back the painful flashbacks.
I don't know how much it's worth talking about research leadership on this particular site (there is so little tech community here I feel), but this is a perspective I have LONG had (I used to talk about it all the time on twitter). And as a Research VP I witnessed often that junior data folks...
I cannot tell you how many data science teams I've seen operate with people with their heads down constantly spending all their time thinking about some nuance of some technical tool, but completely unable to formulate a real articulate research question. Similar to many software teams in some ways
December 27, 2025 at 4:32 PM
"The church we finished building three months ago is screeching curses pleads and bribe offers to the drones circling above it like algorithmically prophetic crows."

#ShortFiction #ScienceFiction #AdversarialMetanoia
The Burning Church
The church we finished building three months ago is screeching curses pleads and bribe offers to the drones circling above it like algorithmically prophetic crows. Its Faraday skin must be finally…
blog.rinesi.com
December 25, 2025 at 4:13 AM
"The church we finished building three months ago is screeching curses pleads and bribe offers to the drones circling above it like algorithmically prophetic crows."

#ShortFiction #ScienceFiction #AdversarialMetanoia
The Burning Church
The church we finished building three months ago is screeching curses pleads and bribe offers to the drones circling above it like algorithmically prophetic crows. Its Faraday skin must be finally…
blog.rinesi.com
December 24, 2025 at 1:15 AM
Yes, but/also: In practice, most top-down AI deployments are driven by people who explicitly want to replace hard-to-control humans (skilled, hard-to-replace experts) with controllable "just check the AI" cheaper workers. Musk or Altman aren't defending social dignity. 1/

via @tedunderwood.com
the whole AI panic is a grim reminder that nobody actually understands what humans care about: control of other humans. We are willing to pay for stuff that is crafted by other humans because we are animals in a social hierarchy. markets formalize and abstract this but it's what drives them.
December 23, 2025 at 7:18 PM
"The church we finished building three months ago is screeching curses pleads and bribe offers to the drones circling above it like algorithmically prophetic crows."

#ShortFiction #ScienceFiction #AdversarialMetanoia
The Burning Church
The church we finished building three months ago is screeching curses pleads and bribe offers to the drones circling above it like algorithmically prophetic crows. Its Faraday skin must be finally…
blog.rinesi.com
December 22, 2025 at 5:17 PM
A second-order concern I have is overoptimistic beliefs about LLMs leading to organizational shifts in hiring, resources, work patterns, etc, not only worse from an analytical point of view but also blocking the feedback loops that would allow the organization to recognize this.

via @jmberger.com
Over the past few months I’ve spoken to a number of intelligence analysts in the private sector about their use of LLMs in their work.

A number of concerning observations 🧵

1) Overwhelmingly, LLMs are thrust upon them by management with no training or direction other than ‘use it somehow’.
December 22, 2025 at 1:44 AM