Tootfinder

OF COURSE!
LLMs do not have any knowledge They only have words. When "introspecting" they are actually just word-salading that like they do everything else. They can't SAY that so they make up plausible-seeming explanations.
BECAUSE THAT IS ALL THEY DO, EVER. https://mastodon.social/@arstechni…

Ars Technica (@arstechnica@mastodon.social)
Attached: 1 image LLMs show a “highly unreliable” capacity to describe their own internal processes Anthropic finds some LLM "self-awareness," but "failures of introspection remain the norm." https://arstechnica.com/ai/2025/11/llms-show-a-highly-unreliable-capacity-to-describe-their-own-internal-processes/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social

Here's what bothers me about AI-generated code, and I'm going to use my own fuck-up as an example. I was calling a DBus interface from C , which I'd already done elsewhere in the code. So I cut & pasted, and adjusted variable names.. It didn't work. So naturally I assumed something was wrong on the other end. I added delays to make sure the other side had enough time to bring up interfaces, I wrote some introspection code to ensure the interface was available, etc.

""Rousseau invites endless introspection. Franklin invites progress. The first is about how you feel; the second is about what you build.
I find the Franklin model far more useful. Not because it’s truer in some cosmic sense, but because it gives you agency. You can’t always change how you feel, but you can always decide what to do next.""

Introspection in Learned Semantic Scene Graph Localisation
Manshika Charvi Bissessur, Efimia Panagiotaki, Daniele De Martini
https://arxiv.org/abs/2510.07053 https://

Introspection in Learned Semantic Scene Graph Localisation
This work investigates how semantics influence localisation performance and robustness in a learned self-supervised, contrastive semantic localisation framework. After training a localisation network on both original and perturbed maps, we conduct a thorough post-hoc introspection analysis to probe whether the model filters environmental noise and prioritises distinctive landmarks over routine clutter. We validate various interpretability methods and present a comparative reliability analysis. …

Monumental III 🪦
纪念 III 🪦
📷 Nikon F4E
🎞️ Ilford HP5 Plus 400, expired 1993
#filmphotography #Photography #blackandwhite

Psi-Turing Machines: Bounded Introspection for Complexity Barriers and Oracle Separations
Rafig Huseynzade
https://arxiv.org/abs/2510.08577 https://arxiv.o…

Psi-Turing Machines: Bounded Introspection for Complexity Barriers and Oracle Separations
We introduce Psi-Turing Machines (Psi-TM): classical Turing machines equipped with a constant-depth introspection interface $ ι$ and an explicit per-step information budget $ B(d,n)=c\,d\log_2 n $. With the interface frozen, we develop an information-theoretic lower-bound toolkit: Budget counting, $ Ψ$-Fooling, and $ Ψ$-Fano, with worked examples $ L_k $ and $ L_k^{\mathrm{phase}} $. We prove an oracle-relative separation $ P^Ψ \neq NP^Ψ $ and a strict depth hierarchy, reinforced by an Ant…

🔍 Schema-aware query intelligence through secure introspection understands your database structure for accurate query generation
🔌 Multi-model flexibility: supports #OpenAI, #Anthropic, #Gemini

#Nihilism #ExistentialNihilism

Urban Domicile 🏠
城市聚落 🏠
📷 Pentax MX
🎞️ Ilford Pan 100
#filmphotography #Photography #blackandwhite

It is refreshing to recognize that the universe has no inherent meaning, and that I am free to create my own meaning on this blank page, one that ultimately burns away when I die.
This realization makes it so much easier for me to let go of my social media accounts, and honestly, that is a relief.
#Nihilism