#llms
17 briefs tagged #llms.
Researchers propose aDSL for agentic 3D creation
The paper pairs an agent-centric DSL with a multi-agent loop for programmatic 3D content creation.
StartupBench tests agents on real startup workflows
The new benchmark finds the strongest tested model completes only about 30% of tasks.
Researchers propose hybrid-policy self-editing for LLMs
The arXiv paper targets composability in unstructured knowledge editing for language models.
Researchers map activation spikes in hybrid attention LLMs
The study finds recurring pre-attention spikes and inter-spike plateaus across hybrid linear attention models.
Researchers improve logical compound-answer reasoning
A new framework decomposes AND, OR and NEITHER/NOR options before composing predictions.
Google DeepMind introduces sign-language-to-text model
SL2T powers new sign language features for Deaf and hard of hearing users.
Paper maps blueprint for economic world models
The arXiv paper outlines a six-level roadmap for building generative economic simulations.
Continual learning work targets forgetting in AI models
New papers propose CP-MoE and frame continual learning as system-level adaptation.
Researchers outline blueprint for economic world models
The arXiv paper maps a capability ladder for agent-based simulations of economies.
New papers probe self-distillation for LLM reinforcement learning
ArXiv reports propose ICE and OCSD while warning that privileged-information teachers can fail.
Reddit adds AI moderation tools for subreddits
Rules Hub uses LLMs to help moderators enforce community rules, with broader launch planned later this year.
Study finds LLMs fabricate user-profile claims
MirageBench reports pervasive over-inference across 12 personalized LLMs with memory.
Hank Green steps back after criticism over AI use
The YouTuber said his LLM use was “not healthy” but not for scriptwriting.
Researchers probe latent reasoning in language models
New arXiv papers test whether continuous hidden-state reasoning can improve LLM reasoning and agent collaboration.
CALVER challenges voting for LLM causal reasoning
A symbolic verifier outperformed voting and judge methods on multi-answer causal queries.
ReflectRL trains on failed expert reasoning traces
The paper proposes using flawed expert trajectories as reflection signals in on-policy training.
Hank Green steps back after criticism over LLM use
The creator said his AI use was “not healthy,” while saying he used it for research sources, not scripts.