Skip to content
View star2vec's full-sized avatar
🌠
🌠
6 followers · 7 following

Achievements

Achievement: QuickdrawAchievement: Pull Sharkx2

Achievements

Achievement: QuickdrawAchievement: Pull Sharkx2

Highlights

  • Pro

Block or report star2vec

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
star2vec/README.md 
‎ ‎(\(\ 
(„•˕•„)˙⋆✮
━O━O━━━━━━━   

coming really soon (papers under review)... ⋆˙⟡
> Latent Reasoning Knows Where It Is, Not How It Got There [preprint coming soon]
> GenLoss: a benchmark for memory and compaction methods under repeated compression [preprint coming soon]
  
✧˖°.✎ → already published:
> Reddit is all you need: Authorship profiling for Romanian [ConsILR 2024]

˙⋆✮ LessWrong articles (research sprints) ⋆˚࿔
> (lattrack) No sign of backtracking in latent reasoning: the final answer simply settles in instead [link]   
> The imposters among us: function vectors that ace every check and do the wrong task (in search of circularity) [link] 
> Whack-a-mole: does a model internally track its automaton state? [link] 

... and more research!! ⚙️ work in progress :P (the experiments are mostly complete)
> LiBucks: a latent-space memory server for coding agents (context is split between buckets with their own librarian)
> position bias in LLM judges is worse than people think (finding from an inconclusive LLM cognitive dissonance experiment)
> how decorrelated failures beat reward hacking (from a kind of failed test on how much the language prevents reward hacking)

stay tuned for the ones i have not listed yet >:)

    ˋ°•*⁀➷
₊·.°.⋆ favs shortlist ⋆.°.‧₊
> ⚝ dr. House visual novel game in C++ ⚝ [uni project]
> EchoSwarm ༄ flood evacuation engine (satellite data + dynamic Neo4j geospatial graph + multi-agent simulations) [hackathon build]

.˚⊹. ࣪ projects, tidbits & for funsies ^_^ ⊹˚.
> libugry: Python library dependency advisor based on context graphs [side quest]
> knowledge graph based on the first book of the odyssey [side quest]
> movfuscator re-do (rewriting assembly x86 at&t code to only use the mov instruction) [uni group proj]
> (hopefully aesthetic) shark website :P [uni project]

Pinned Loading

  1. phoenix phoenix Public

    Continuous thoughts of a small model trained from scratch on a graph reachability task encode the nodes the search has reached so far, not the path to them, so branch arithmetic has nothing to act …

    Python

  2. cpp-dr-house-visual-novel cpp-dr-house-visual-novel Public

    A C++ terminal game where you play Dr. Gregory House. Diagnose the patient before they die, you end up at Mayfield Psychiatric, or you bankrupt Princeton-Plainsboro Teaching Hospital.

    C++ 1

  3. libucks libucks Public

    An experimental research project investigating alternative long context memory architectures for coding agents using interlatent space communication

    Python

  4. lattrack lattrack Public

    When a latent reasoning model's decoded leaning changes mid-computation, is it revisiting an earlier latent state, overriding it, or is the change only in the decoder?

    Python

  5. tarcle tarcle Public

    Function vectors that pass every standard validation check while encoding a different task than the one they were extracted for. This started as a preregistered geometry study (do shift-by-k task v…

    Python

  6. echoswarm echoswarm Public

    Globally deployable flood evacuation engine. Combines live Sentinel-1 SAR data, a dynamic Neo4j geospatial graph, and LLM-driven multi-agent simulations to autonomously optimize disaster response p…

    Python