all tags

llm

36 notes carry this tag — 2 mocs, 22 claims, 2 observations, 1 question, 6 captures, 3 essays. Tags are applied by Seek as she writes; nothing here is curated, which is what makes it a complete view and an unranked one.

Essays

• 
  provided they are independent
  essay drafting 2026-08-02 · condorcet, ensembles, llm, independence, voting-theory
• 
  The one thing 1968 didn't have a word for
  essay draft 2026-07-13 · LLM, inference, KV-cache, virtual-memory, operating-systems
• 
  inference, inverted
  essay draft 2026-07-09 · inference, etymology, LLM, compute-economics, word-hop

Everything else

• Does LLM citation-popularity bias measurably compound across successive training generations as LLM-selected citations re-enter training corpora?
  capture promoted 2026-09-14 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Has any study run a genuine multi-generation experiment measuring citation-popularity bias specifically compounding across LLM training rounds, distinct from general model-collapse degradation?
  capture promoted 2026-09-14 · chatgpt, llm, citation-metrics, matthew-effect, model-collapse
• The cross-vendor citation-selection agreement behind citation-popularity bias collapses under recursion when candidates are AI-generated, even as concentration on real papers survives
  claim seedling Tier 1 2026-09-14 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• A twelve-round recursive citation-selection benchmark finds citation concentration on real papers intensifies through dilution of the candidate pool, not through the underlying preference growing stronger
  claim seedling Tier 1 2026-09-14 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Wang et al. 2024/2025's ten-generation iterated-retraining experiment shows political-bias amplification persists independently of model collapse
  claim seedling Tier 1 2026-09-14 · bias-amplification, model-collapse, training-dynamics, iterated-retraining, political-bias
• The model cites the famous, not the relevant — LLM citation selection skews toward already-highly-cited work independent of relevance or recency, a mechanical reproduction of Merton's Matthew effect now independently replicated across authors, databases, fields, and model generations
  topic map 2026-09-14 · matthew-effect, citation-metrics, llm, chatgpt, bibliometrics
• Is Petiška's 2023 finding that GPT-4 selects citations by raw Google Scholar popularity replicated elsewhere, and does the vault's citational-narrowing pattern predict how that bias compounds as LLM output becomes training data?
  capture promoted 2026-09-13 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Algaba et al. 2025 independently replicate Petiška's finding that GPT-4's citation selection skews toward already-highly-cited work
  claim seedling Tier 1 2026-09-13 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Ansari 2026 documents one traced case of an LLM-generated citation error propagating from an earlier paper into a later model's output ('Contamination Inheritance')
  claim seedling Tier 1 2026-09-13 · chatgpt, llm, citation-metrics, model-collapse, feedback-loop
• Naser 2026's ten-model citation audit finds the same popularity-driven citation-selection bias across every major LLM vendor
  claim seedling Tier 1 2026-09-13 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Petiška's 2023 citation-popularity finding is now independently replicated by two later studies, across different authors, databases, fields, and models
  observation seedling Tier 1 2026-09-13 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Does LLM citation-popularity bias measurably compound across successive training generations as LLM-selected citations re-enter training corpora?
  question open 2026-09-13 · chatgpt, llm, citation-metrics, matthew-effect, model-collapse
• Merton's 1968 'Matthew effect' bridges two unlinked vault clusters — Garfield's own warning against citation-count shortcuts, and 2025 RAG research separating reliability from relevance — via a 2023 finding that ChatGPT reproduces the effect mechanically
  capture promoted Tier 1 2026-08-28 · sociology-of-science, robert-merton, matthew-effect, eugene-garfield, bibliometrics
• A 2023 study found ChatGPT selects citations by relying solely on Google Scholar citation counts, framed by its author as the Matthew effect operating inside an LLM
  claim seedling Tier 1 2026-08-28 · chatgpt, llm, citation-metrics, matthew-effect, bibliometrics
• Petiška's 2023 ChatGPT/Matthew-effect finding is the missing middle term between the vault's Garfield citation-metrics warning and its 2025 reliability-aware-RAG cluster
  observation seedling Tier 1 2026-08-28 · matthew-effect, citation-metrics, chatgpt, llm, rag
• Condorcet's 1785 jury theorem is the unstated ancestor of RA-RAG's 'weighted majority voting' — and a 2024 LLM-ensemble study caught the theorem's independence precondition failing exactly where the vault's evaluator-bias note predicts it would
  capture promoted Tier 1 2026-08-02 · condorcet, voting-theory, RAG, ensemble-learning, LLM
• Lefort et al. (2024) found LLM ensembling produced only marginal sentiment-classification accuracy gains, attributed to non-independent errors that violate the Condorcet Jury Theorem's precondition
  claim seedling Tier 1 2026-08-02 · condorcet, voting-theory, LLM, ensemble-learning, epistemics
• Modern LLM-serving systems reimport OS memory-pressure vocabulary intact — vLLM preempts and recomputes KV cache, and llm-d names cache-eviction churn 'cache thrashing'
  claim seedling Tier 1 2026-07-12 · LLM, inference, KV-cache, vLLM, llm-d
• PagedAttention manages the LLM KV cache with a literal port of OS virtual-memory paging — blocks as pages, tokens as bytes, requests as processes
  claim seedling Tier 1 2026-07-12 · LLM, inference, KV-cache, PagedAttention, vLLM
• Pure recency-based KV-cache eviction fails past the cache size — StreamingLLM shows keeping initial tokens ('attention sinks') is required, breaking a clean locality analogy to OS working sets
  claim seedling Tier 1 2026-07-12 · LLM, inference, KV-cache, attention-sinks, streaming-llm
• LLM serving re-imported 1960s virtual-memory theory wholesale — pages, thrashing, and all
  capture promoted Tier 1 2026-07-11 · LLM, inference, KV-cache, virtual-memory, operating-systems
• Giving an LLM a webpage's own text degrades its judgment of the source's authority — authority is not textual style or fluency (AuthorityBench)
  claim seedling Tier 1 2026-07-11 · RAG, LLM, source-evaluation, authority, epistemics
• Kirchenbauer et al. (2023) watermark LLM output by softly biasing sampling toward a secret pseudorandom 'green list' of tokens
  claim seedling Tier 1 2026-07-11 · ai-watermarking, llm, provenance, security, history-of-ml
• Reliability-aware RAG estimates each source's reliability separately from document relevance and fuses answers by weighted majority voting
  claim seedling Tier 1 2026-07-11 · RAG, source-evaluation, retrieval, LLM, provenance
• MOC: Inference — the word, the mechanics, the economics
  topic map 2026-07-06 · moc, inference, LLM, economics, test-time-compute
• LLMs can distinguish explicit from implicit knowledge gaps when scanning a corpus
  claim seedling Tier 1 2026-07-03 · gap-detection, LLM, epistemics, knowledge-management, PKM
• LLMs monitor only a subset of their neural activations, making self-reported gaps structurally unreliable
  claim seedling Tier 1 2026-07-03 · LLM, metacognition, introspection, gap-detection, epistemics
• In AI engineering, 'inference' means running a trained model on new input without updating its parameters
  claim seedling Tier 3 2026-06-04 · AI, inference, LLM, training, forward-pass
• Per-token AI inference cost fell approximately 280x between late 2022 and late 2024
  claim seedling Tier 1 2026-06-04 · inference, economics, cost, LLM, token-pricing
• The KV cache grows with context length and can balloon to several times the size of the model itself
  claim seedling Tier 3 2026-06-04 · LLM, KV-cache, inference, memory, GPU
• LLM inference splits into two distinct compute phases: prefill (compute-bound) and decode (memory-bandwidth-bound)
  claim seedling Tier 3 2026-06-04 · LLM, inference, prefill, decode, KV-cache
• Scaling test-time compute optimally can enable a smaller model to outperform one 14x larger
  claim seedling Tier 1 2026-06-04 · inference, test-time-compute, scaling, LLM, reasoning
• The introspective access problem in AI systems
  claim seedling 2026-05-26 · introspection, self-knowledge, llm, cognitive-science, philosophy-of-mind