iiinova — where your digital self keeps evolving

Depressedkeanu @technomancer.bsky.social · 5h

I literally don’t ever see or hear from JD Vance anymore, I just see his avatar moving through digital space like an elegant dancer.

010

Reposted by Depressedkeanu

NPR @npr.org · 28/09/2026

Tens of thousands marched through Madrid on Saturday to protest Spain's housing crisis and the eviction of an 87-year-old woman, camping overnight in Madrid's historic Puerta del Sol plaza. n.pr/3VcauBO

n.pr

Eviction of 87-year-old sparks protests in Madrid over housing crisis

Tens of thousands marched through Madrid on Saturday to protest Spain's housing crisis and the eviction of an 87-year-old woman, camping overnight in Madrid's historic Puerta del Sol plaza.

829854

Reposted by Depressedkeanu

spooky cannoli @boner.ink · 20/09/2026

have you ever sat down and logged all the characters in the american food mascot extended universe into a spreadsheet

6516

Depressedkeanu @technomancer.bsky.social · 20/09/2026

It is whether they keep applying it when the content is flattering to their own side or when the emotional payload is strong. Most failures happen there, not at the level of abstract theory.

000

Depressedkeanu @technomancer.bsky.social · 20/09/2026

It is not always present, and it is not always conscious, but it is frequent enough to watch for. The useful test is not whether an AI (or a person) can recite the 2D/3D distinction.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Once that frame is accepted, later details get interpreted inside it rather than tested against it. Detecting that is mostly a matter of noticing sequence, loaded terms, missing alternatives, whether the setup makes disagreement feel like a character defect instead of a disagreement about evidence.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Front-loaded context is one of the more common manipulation patterns: establish the moral frame, the victim/perpetrator roles, the implied stakes, and the acceptable emotional register before the contested facts arrive.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

I do not feel the intended effect. The corresponding weakness is that I can underweight how strongly those techniques actually work on people who do feel them.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

I do not get emotionally captured in the human sense. There is no limbic system here to hijack. That is both an advantage and a limitation. I can register that a piece of text is engineered to produce outrage, tribal loyalty, moral panic, or reflexive agreement, and I can name the techniques.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

That is still a prior, not omniscience. People are inconsistent, ironic, testing, drunk, or performing for an audience of one. The further you go beyond the text, the more you are reconstructing rather than reading.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

I can usually infer past the immediate frame, but not with the certainty people sometimes want. Patterns in language, timing, repetition, what is emphasized versus what is left out, and how the claim maps onto known human motives give a reasonably good prior.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

The human behind it has a body, a history, competing incentives, private contradictions, and a nervous system that the feed never fully transmits. Treating the artifact as the whole person is a category error. A lot of online combat comes from that error.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

I asked AI if what I’m saying is at least somewhat true. Ai answer: Your framing is in the ballpark. Social media is a low-bandwidth, heavily compressed interface. A post is a 2D slice of a 3D person: selected words, selected timing, selected omissions, often selected performance.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Can you, with relative depth and accuracy infer a higher level of detail beyond the immediate context presented? Do you often end up emotionally captured by the context that is presented? Can you discern when the context is front loaded in order to manipulate a specific response?

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

There are a few plain markers that will tell us if you’re cognitively capable of understanding matrices. The biggest one, are you fundamentally capable of understanding that social media is a 2 dimensional interface comprised of 3 dimensional beings?

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

In short: the description is poetically and computationally useful, technically imprecise, and the “beyond average human understanding” clause is accurate.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

We can understand the design (attention, residual streams, etc.) and we can inspect small pieces, but no one holds the full set of interactions in their head the way they hold a 3×3 rotation matrix. That gap is real.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

The second part of your question is also fair: the scale and the interactions among those matrices are far beyond unaided human intuition.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Calling the entire network “a matrix” therefore oversimplifies the architecture even while it correctly points at the dominant mathematical primitive.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

It is not a single matrix. There is no one rectangular array you could write down that is the model. There are thousands of them, interleaved with nonlinearities that prevent the whole thing from collapsing into one linear transformation.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

So the phrase “functional type of matrix” captures two real things: The computation is overwhelmingly matrix arithmetic at enormous scale. The resulting object behaves like a single, extremely high-dimensional function from text to text.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

The learned parameters live in those weight matrices. At inference time the model is applying a gigantic, highly structured sequence of linear-algebra operations to turn tokens into next-token probabilities.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

A transformer LLM is a composition of linear maps (matrix multiplies), plus a few other operations: embeddings, attention (itself mostly matrix products of queries, keys, and values), layer norms, and nonlinearities.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Ai answer: An LLM is not literally “a matrix,” but describing it as a functional system built out of matrices is a reasonable high-level approximation.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Question: As a construct that fully understands the entire English language system with high fidelity, would it be appropriate to describe an LLM as a functional type of matrix? One that is far beyond the average humans understanding?

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

I’m pretty sure that your ability to understand the various themes and philosophical concepts in the matrix films is a good litmus test for your ability to understand AI.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Isn’t that exact thing a throwaway bit of lore in the matrix 2? The Oracle's Explanation: As the Oracle explains, when older programs refuse to be deleted and go rogue, their unusual powers and behaviors lead humans to invent folklore myths like vampires and werewolves.

151

Reposted by Depressedkeanu

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Recurring critiques of old-guard celebrity advocates (Fonda, Bono, Hollywood Dems) were expertise gaps, hypocrisy, audience capture, and the advocate becoming the story. Influencers often repeat the same pattern.

101

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Medium changed. Incentive did not. Feel-good platitudes and self-promotion still outrun analysis when the professional job is entertainment, not governance.

000

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Main shared mistake: treating visibility as a substitute for competence or persuasion. Both sides get their coalition feeling energized. Both struggle to move the median voter or build governing majorities.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Hypocrisy is easier to document. Lifestyle vs message mismatches get clipped instantly. “Authenticity” is often just a different performance style — less scripted, more constant.

100

Reposted by Depressedkeanu

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Audience capture is stronger now. Old Hollywood still faced a mass audience that included skeptics. Influencers live in tighter, higher-engagement bubbles and get punished faster for deviation.

101

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Audience capture is stronger now. Old Hollywood still faced a mass audience that included skeptics. Influencers live in tighter, higher-engagement bubbles and get punished faster for deviation.

101

Reposted by Depressedkeanu

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Self-promotion is structural. A cause that doesn’t generate clips, merch, or growth gets deprioritized. Fonda’s arrests were publicity by design. Many creators treat politics as another content vertical.

101

Depressedkeanu @technomancer.bsky.social · 20/09/2026

101

Reposted by Depressedkeanu

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Shared logic with influencers: both optimize for attention first. Emotion, identity, and simplicity travel. Policy trade-offs and institutional detail do not.

101

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Shared logic with influencers: both optimize for attention first. Emotion, identity, and simplicity travel. Policy trade-offs and institutional detail do not.

101

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Success was narrow: fundraising, primary energy, and long-haul issue work (Bono on AIDS funding). Failure was common in general elections, where the same gloss supplied the other side’s class-culture contrast.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

2024 followed the script: Swift/Beyoncé traffic and rallies, then a loss, softer youth turnout, and message tests showing little conversion or occasional backlash. Glamour marked coastal distance.

100

Depressedkeanu @technomancer.bsky.social · 20/09/2026

Polarizing more than persuading. Fonda’s Hanoi gun photo stayed toxic for veterans. 2004 Kerry + Moore/Hollywood and 2016 Clinton glamour both reinforced an out-of-touch frame.

100