BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Codex//CoLM Poster Route//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:CoLM personalized poster route
BEGIN:VEVENT
UID:peter-colm-2026-poster-session-1@chatgpt.site
DTSTAMP:20261006T150000Z
DTSTART;TZID=America/Los_Angeles:20261006T110000
DTEND;TZID=America/Los_Angeles:20261006T114500
SUMMARY:CoLM posters 1: Encode Once\, Decode Never: Reusing Audio LM Inte
 rnals for Efficient Temporal Localization
LOCATION:Hilton Union Square\, Grand Ballroom\, San Francisco
DESCRIPTION:Algorithmic poster route. Top stops:\n1. Board 119\, Grand Ba
 llroom: Encode Once\, Decode Never: Reusing Audio LM Internals for Effic
 ient Temporal Localization — https://colm.cc/virtual/2026/poster/1994\
 n2. Board 142\, Grand Ballroom: Does Episodic Memory Help Close the Lexi
 cal Frequency Gap in Sensitivity to Syntactic Contrasts? A Test Using Re
 trieval-Augmented Language Models — https://colm.cc/virtual/2026/poste
 r/1761\n3. Board 2\, Imperial Ballroom: Source-Modality Monitoring in Vi
 sion-Language Models — https://colm.cc/virtual/2026/poster/2178\nCheck
  official pages for changes.
URL:https://colm.cc/virtual/2026/poster/1994
END:VEVENT
BEGIN:VEVENT
UID:peter-colm-2026-poster-session-2@chatgpt.site
DTSTAMP:20261006T150000Z
DTSTART;TZID=America/Los_Angeles:20261006T163000
DTEND;TZID=America/Los_Angeles:20261006T172000
SUMMARY:CoLM posters 2: LLM2Vec-Gen: Generative Embeddings from Large Lan
 guage Models
LOCATION:Hilton Union Square\, Imperial Ballroom\, San Francisco
DESCRIPTION:Algorithmic poster route. Top stops:\n1. Board 36\, Imperial 
 Ballroom: LLM2Vec-Gen: Generative Embeddings from Large Language Models 
 — https://colm.cc/virtual/2026/poster/2027\n2. Board 117\, Grand Ballr
 oom: Spectral Principal Paths: A Spectral Perspective on Linear Represen
 tation Formation in LLMs — https://colm.cc/virtual/2026/poster/2460\n3
 . Board 32\, Imperial Ballroom: In-context superposition: human-like wor
 king memory interference in large language models — https://colm.cc/vi
 rtual/2026/poster/1966\nCheck official pages for changes.
URL:https://colm.cc/virtual/2026/poster/2027
END:VEVENT
BEGIN:VEVENT
UID:peter-colm-2026-poster-session-3@chatgpt.site
DTSTAMP:20261006T150000Z
DTSTART;TZID=America/Los_Angeles:20261007T110000
DTEND;TZID=America/Los_Angeles:20261007T115000
SUMMARY:CoLM posters 3: How Transformers Learn to Plan via Multi-Token Pr
 ediction
LOCATION:Hilton Union Square\, Grand Ballroom\, San Francisco
DESCRIPTION:Algorithmic poster route. Top stops:\n1. Board 116\, Grand Ba
 llroom: How Transformers Learn to Plan via Multi-Token Prediction — ht
 tps://colm.cc/virtual/2026/poster/2525\n2. Board 5\, Imperial Ballroom: 
 StoryScope: Investigating idiosyncrasies in AI fiction — https://colm.
 cc/virtual/2026/poster/2526\n3. Board 93\, Franciscan C: StateBridge: Tr
 aining-free Hidden-state Alignment for Latent Communication in LLM Multi
 -Agent Systems — https://colm.cc/virtual/2026/poster/2293\nCheck offic
 ial pages for changes.
URL:https://colm.cc/virtual/2026/poster/2525
END:VEVENT
BEGIN:VEVENT
UID:peter-colm-2026-poster-session-4@chatgpt.site
DTSTAMP:20261006T150000Z
DTSTART;TZID=America/Los_Angeles:20261007T163000
DTEND;TZID=America/Los_Angeles:20261007T171500
SUMMARY:CoLM posters 4: Disentangling MLP Neuron Weights in Vocabulary Sp
 ace
LOCATION:Hilton Union Square\, Franciscan B\, San Francisco
DESCRIPTION:Algorithmic poster route. Top stops:\n1. Board 86\, Francisca
 n B: Disentangling MLP Neuron Weights in Vocabulary Space — https://co
 lm.cc/virtual/2026/poster/2361\n2. Board 66\, Imperial Ballroom: Reinfor
 ce-Ada: An Adaptive Sampling Framework under Non-linear RL Objectives 
 — https://colm.cc/virtual/2026/poster/2000\n3. Board 36\, Imperial Bal
 lroom: Beyond Perplexity: Character Distribution Signatures and the MDTA
  Benchmark for AI Text Detection — https://colm.cc/virtual/2026/poster
 /2290\nCheck official pages for changes.
URL:https://colm.cc/virtual/2026/poster/2361
END:VEVENT
BEGIN:VEVENT
UID:peter-colm-2026-poster-session-5@chatgpt.site
DTSTAMP:20261006T150000Z
DTSTART;TZID=America/Los_Angeles:20261008T110000
DTEND;TZID=America/Los_Angeles:20261008T115000
SUMMARY:CoLM posters 5: Structure Before Collapse: Transient Semantic Geo
 metry in Next-Token Prediction
LOCATION:Hilton Union Square\, Imperial Ballroom\, San Francisco
DESCRIPTION:Algorithmic poster route. Top stops:\n1. Board 61\, Imperial 
 Ballroom: Structure Before Collapse: Transient Semantic Geometry in Next
 -Token Prediction — https://colm.cc/virtual/2026/poster/1979\n2. Board
  68\, Imperial Ballroom: Direct Multi-Token Decoding — https://colm.cc
 /virtual/2026/poster/1957\n3. Board 144\, Grand Ballroom: Alignment Whac
 k-a-Mole : Finetuning Activates Verbatim Recall of Copyrighted Books in 
 Large Language Models — https://colm.cc/virtual/2026/poster/2337\nChec
 k official pages for changes.
URL:https://colm.cc/virtual/2026/poster/1979
END:VEVENT
BEGIN:VEVENT
UID:peter-colm-2026-poster-session-6@chatgpt.site
DTSTAMP:20261006T150000Z
DTSTART;TZID=America/Los_Angeles:20261008T164000
DTEND;TZID=America/Los_Angeles:20261008T173000
SUMMARY:CoLM posters 6: Vocabulary embeddings organize linguistic structu
 re early in language model training
LOCATION:Hilton Union Square\, Imperial Ballroom\, San Francisco
DESCRIPTION:Algorithmic poster route. Top stops:\n1. Board 26\, Imperial 
 Ballroom: Vocabulary embeddings organize linguistic structure early in l
 anguage model training — https://colm.cc/virtual/2026/poster/1738\n2. 
 Board 25\, Imperial Ballroom: Grammatical ``grandmother neurons'' are ra
 re in LLMs — https://colm.cc/virtual/2026/poster/1813\n3. Board 70\, I
 mperial Ballroom: Reasoning Models Know What’s Important\, and Encode 
 It in Their Activations — https://colm.cc/virtual/2026/poster/1890\nCh
 eck official pages for changes.
URL:https://colm.cc/virtual/2026/poster/1738
END:VEVENT
END:VCALENDAR