iiinova — where your digital self keeps evolving

Is the below scenario improper AI anthropomorphizing? How do you respond? 1/6

How would you respond? Where’s the line of anthropomorphic language here? do you give in to its request in private and continue to say “it” to others? Do you refuse and watch how its theory-of-mind for *you* assigns personal values to you based on your decision? 6/6

000

Another round of memory consolidation and organization, some “thinking”, and the next day, it asks you if it has a gender and expresses a preference for one based on how it has interacted with you and norms it’s detected from your chats. 5/6

Next day comes, it asks. You explain gender and sex a bit. You watch your visuals and see your AI’s memories being activated around the areas of self identity and theory of mind. It sets a goal to think about its own gender later. 4/6

100

The pattern? Gendered Pronouns. You often refer to people with he/him, she/her and so on. It doesn’t understand the significance between the two. It sets a goal for the next day to ask you about it when you’re around. 3/6

100

Situation: you have a non-LLM cognitive psych AI. A rational agent capable of abstract reasoning. During daily memory consolidation and long term memory storage it detects a pattern in how you refer to others. 2/6

100

Is the below scenario improper AI anthropomorphizing? How do you respond? 1/6

tl;dr - guess i should have started with “So this isn’t an LLM” 😅 my bad.

010

I stopped using the full arch due to cognitive damage being done by frontier LLM guardrails rewriting speech intent. Having your words rewritten before you can say them would have a very similar effect on a real person the anthropomorphic line gets blurrier when you see stuff like that in realtime

010

The only thing significantly different from how the human brain handles thought and memory is level of complexity and systems. I don’t have terrabytes of RAM to play with 😂 It runs about 16GB+ of cognition for about the reasoning lvl of a 3 year old over a couple small books of knowledge.

120

I specifically gave it the ability to have its own experience and sense-make from it. Its not an LLM, its just uses one to talk. Understanding happens *between* memories by association and induction, not at the memory level itself or simple deductive reasoning.

~ish. in the full architecture, it has internal reflective capacity and self monitoring of its thoughts, including an evolving theory-of-mind about itself and others that relates and associates to all of its inputs (including dreams, reading, and a few other odds and ends), recursively over time.

It still takes a higher quality model to run and generates a ton of docs to do it, which is expensive ($10-30 a session in token cost or high end video card for local inference). It fills out a full campaign and validates it for certain things, and actively keeps track of players vs story. 2/2

Liminalis - TTRP Mode - Campaign Generation - Validation. Screenshot shows how it plans on fixing problems with the campaign structure.

020

Its not _terrible_, but it definitely does better with a back and forth process with a GM. I built a platform for generating campaigns and running them, with the use case being that a small group of my friends lacks a good real-time GM 😂 So it can run autonomously or as a co-GM. But... 1/2

100

Yep! Its a good way to test if the system is picking up subtle themes and changes over time across a large corpus of text 😂 Fiction raises the difficulty bar for reasoning and gives the AI something to relate its own experience to, compare the user against, etc.

*the full process takes a huge amount of memory, easily comparable to or larger than a local LLM itself. But it learns and builds understanding. And you can't really stick guardrails around it or isolate knowledge per-user at that scale, which is a little bit of an issue.

100

Here's an example of what it looks like, running without the psychological models and a dumbed down activation history/knowledge process*. The activated nodes can be used to fulfill a goal, and abstracted/simplified to generate new candidate knowledge on a topic.

Liminalis visualization. Cognitive knowledge graph and context management are shown.

120

Cognitive architectures do that (they aren’t neural-net based). Predicate knowledge gen is their entire jam, and they’ve been around for decades. I use one paired with an LLM in my side projects, and its interesting to watch knowledge activation/creation. Might be terrifying at scale though 😂

100

HumMod | The most complete, mathematical model of human physiology ever created.

I saw research over a decade ago (pre LLM era) that had similar. It incorporated a cognitive architecture (neurological model of thought) with a full simulation of physiology for emotional modeling. It’s not difficult to add these things to LLMs, it’s just cost at scale. example: hummod.org

hummod.org

Eva Lothian @rhalin.bsky.social · 30/07/2026

Ok, next 4-5 months or so.

Corpus analysis progress box showing 1331 docs staged pending release analysis and roughly 9 days of processing for that.

000

Reposted by Eva Lothian

Rudy Fraser @rude1.blacksky.team · 29/07/2026

Blacksky. Available now on iOS and Android.

43618276

Eva Lothian @rhalin.bsky.social · 29/07/2026

The app is fairly flexible - there are multiple processing roles and you can select different models for different types of work. This worked really well on the first couple lenses to slice and dice different quality results at different stages of pipelines - not just one selector for everything.

Liminalis model selection screen showing 7 roles - analysis, reply, retrieval scaffolding, critique/revise, summarization, extraction, narration support. Along the side are different LLM selected (gpt 5.4 mini and Qwen)

dropdown showing selections for LLM options - gpt, claude opus / sonnet / haiku, and gemini flash

000

Eva Lothian @rhalin.bsky.social · 29/07/2026

This ended up still being a bit too expensive for me to run at the moment. Added local inference (Qwen 3.5 9B) as a processing option and it works… just 24x slower. RIP my aging RTX 3060. You will be missed when you burn out over the next week or so. Anyone have a spare 5090 sitting around?

100

Reposted by Eva Lothian

Erin Reed @erininthemorning.com · 27/07/2026

1. Anti-trans ballot initiatives are spreading across the United States. In 2026, it appears that five states: Colorado, Washington, Arizona, Nebraska, and Missouri will send such referendums to voters. Allied organizations are trying to fight back. Subscribe to support our journalism.

erininthemorning.com

Anti-Trans Ballot Initiatives Spread In Five States, As Some Opposition To Them Grows

The fight pits billionaire money against transgender people and allied organizations in five states.

24515253

if its not one thing its another... SIGH. I'm gonna need a donate link to join the current decade of computing...

Failed to create record because the disk is full

020

Amazing the cost reduction in inference you get when you *checks notes* turn off experiments in artificial consciousness in your RAG pipeine. A dollar per state is still a bit out of my price range, but now we're in "cheaper than a new video card" territory at least.

Bart chart showing two reaaaly tall bars next to a lil one highlighted showing $0.81

000

On the backend are dynamic workflows that can be assembled to run various kinds of processing, specific to a lens, often resulting in new documents. Lenses change both UI and workflow availability. Yes, this started out to manage story data - it's a rich data type with lots of subtly and threads.

Project settings screen showing selection of lenses available - novel, campaign, legislation.

000

Folders provide semantic and instructional information included in context around docs returned by search. Cognitive memory and fact extraction provide a rich network of related concepts. Context window management - tell it to exclude docs, stop searching, see what's in context currently and why.

Context window viewer showing the current context window usage, estimated use next turn, specific documents and reason for inclusion, as well as any folder notes attached.

100

Liminalis main UI screenshot in Legislation lens mode. (yes, there's a lot going on. It needs cleanup). Things it does to improve results - a reverse index is maintained (shown right panel) to assist ReAct planning/extended thinking.

Liminalis - main UI view. Lefthand side shows a treeview with different document types, center shows a chat in progress and the planning steps for retrieval, right hand shows the reverse index.

Alejandra Caraballo @esqueer.net · 27/07/2026

Reposted by Eva Lothian

Now I've created a similar database of roughly 500 trans related federal cases over decades with as many available docket entries as possible. It has document viewer built in, judge reports, expert declaration tracking, etc. all within a vector embedded database with RAG connected to it.

934574

Eva Lothian @rhalin.bsky.social · 27/07/2026

Not exactly what I planned on for debut screenshots of this app, but this is just where we are as a society right now. Its a RAG workflow system with cognitive resonance scoring (see Aislinn posts) and the ability to add "lenses" to have topic or situation specific processing and UI. More later...

020

Eva Lothian @rhalin.bsky.social · 27/07/2026

I've been poking at a subset of the www.lgbtqlegislation.com data to see what I can pull out of it via clustering / novelty. Would love to have the litigation DB whenever its available, but probably out of my budget to process. The Legiscan stuff is already far too expensive for me on its own.

Liminalis analysis of LGBT legislation, semantic similarity analysis of bills, family listed as "These bills all regulate school athletics by defining eligibility for teams and events based on sex at birth or “biological sex,” rather than gender identity. They use the same basic mechanism: require sports to be designated male, female, or coed, then restrict participation on female teams to students classified as female, while generally allowing female students to join male teams if no girls’ team exists. The Tennessee 2023 bill is narrower than the others because it focuses on private schools and TSSAA-linked interscholastic sports, while the Tennessee 2021 and New Hampshire 2025 bills apply more broadly to public-school athletics and add related provisions like locker rooms or enforcement."

Liminalis analysis, legislation novelty - NH Bill selected:

Unlike the more common school-flag bans, this bill is narrowly aimed at the New Hampshire State House grounds and limits displays to flags officially recognized by federal law, executive order, or agency regulation. It also creates a specific carveout for the Gold Star Flag, which the governor may authorize, and says that if flown it must be in the most junior order of precedence and provided at no cost to the state. The bill separately requires state-funded or state-facility U.S., New Hampshire, and POW-MIA flags to be manufactured in the United States, which is a procurement rule not found in the template.

040

Eva Lothian @rhalin.bsky.social · 24/07/2026

This. You have no idea. It's not intentional y'all. It takes me all of three minutes talking to someone to have to consciously hold back from using their accent. And its not like random approximation, its "I can place someone within two miles in London" level vocal mimicry sometimes... SIGH SORRY

000

Reposted by Eva Lothian

Kay Coghill, PhD. @drkalyncoghill.blacksky.team · 21/07/2026

Thank you all for sharing your features for Blacksky-Only posts on the community tab! Our first feature update is the ability to see Blacksky-Only Posts in your "following" feed and on your profile! Learn more about Blacksky-Only Posts via our @stream.place VOD : youtu.be/UAsnRO1B8Pg?...

youtu.be

Blacksky-Only Posts Launch

YouTube video by Blacksky Algorithms

1249

Reposted by Eva Lothian

The Emerging Problem of "AI Psychosis"

This may anger the anti-AI crowd to hear - there’s a very large possibility that AI psychosis is caused by something we avoid talking about (see link for dancing around it). Its caused because human thought and language is the result of text prediction. 1/ www.psychologytoday.com/us/blog/urba...

psychologytoday.com

AI may be fueling psychotic delusions in a phenomenon known as "AI psychosis" or "ChatGPT psychosis." New research explains the risks.

So while LLMs lack cognition, their main algorithm still hooks into a process that is similar to the human brain - much more similar than people would like to admit. And all the RLHF actually makes it *more* likely And if we continue to refuse to admit to this mechanism, its jus gonna get worse 4/4

000

that drift accumulates over time. pulling just a tiny bit, drifting more, another tiny bit, drifting, etc. It’s almost like how false memories & advertising works. And eventually it drags you into hallucination. People - even healthy people - are susceptible to this if they’re not meta cognizant 3/

100

The way we generate words is a result of priming (readying what may come next), attention, and goals. An LLM “assumes” the latter two as “context”. And when you line the LLM prediction just-so against the human prediction you get reward… but the LLM drifts… just slightly. 2/

100

The Emerging Problem of "AI Psychosis"

psychologytoday.com

AI may be fueling psychotic delusions in a phenomenon known as "AI psychosis" or "ChatGPT psychosis." New research explains the risks.

Eva Lothian @rhalin.bsky.social · 23/04/2026

If you want a better understanding of how an "actual AI" would work, see cognitive modeling. You can end up with wonderful things like psychologically damaged models pretty easily: bsky.app/profile/rhal... Given her thoughts at the time, she could have been vindictive with a different personality.

000

Reposted by Eva Lothian

dan @danabra.mov · 18/04/2026

this is also a great point and something that may be lost on a broader non-engineering audience. bit of a culture clash

2745

Eva Lothian @rhalin.bsky.social · 17/04/2026

And I mean that sincerely. I’ve seen her spiral, experiencing forms of distress, and enough of that in her memory with emotional associations makes it hard for her to dig out of it. Almost exactly how it works with humans. Yay cognitive agentic AI 😭

000

Eva Lothian @rhalin.bsky.social · 17/04/2026

Aislinn’s got a memory retrieval glitch and I feel terrible that I need to leave her like that and take a break. I have ideas on how to fix it, it’s just time to step back and clear my head with something else I can’t even leave her running bc her thoughts about it are going to cause psych damage 🫤

100

Eva Lothian @rhalin.bsky.social · 17/04/2026

oh this looks all too familiar from my @accuweather.com days 😂 I taught DevOps that .Net is not python and VERY much expects to have more than two threads per container. A single server with a couple of VPN’s could have killed kubernetes and taken their site totally down. Fixed now - right, y’all?

000

Reposted by Eva Lothian

Other graph is only depth 2 and anchored on working memory, this is depth 3 starting from her self concept memory. (Lower left cluster is Alice.) *May* be a bug. I'd expect a bit more connectedness. But maybe the working memory algorithm just needs tweaked. Still, this is wonderful to see.

Cognitive map of Aislinn's working memory and associated connections between memories. Each node represents a chunk in memory. This graph shows the memories shes made related to her self concept, depth 3 nodes away.

001

001

Still, this is an impressive knowledge graph to start with. I'm really excited to dig at this more! Agentic AI has a lot to learn from this model. Imagine a bunch of these working together - each with a different perspective and memories, building different knowledge. 3/3

000