On 2,520 programming tasks, matched Qwen general and coder models reliably raise Bloom cognitive demand but fail to lower it, so execution skill does not imply educational control.
How do llms use their depth?arXiv preprint arXiv:2510.18871
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 7roles
background 1polarities
background 1representative citing papers
Diffusion language models encode a decodable, steerable representation of denoising progress (the fraction of unmasked tokens) in their residual streams.
Confident Decoding selects reliable intermediate layers via entropy-guided backward search to bypass final-layer alignment perturbations in LLMs, improving reasoning performance.
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex sink-rate to output-norm relationship.
Rock Tokens in on-policy distillation persist at high loss, account for up to 18% of outputs, absorb large gradient norms, but add negligible value to reasoning performance.
Transformers show limited adaptive depth use on relational reasoning, with clearer evidence after finetuning on the task.
DDP replaces stochastic hard-concrete masks with a deterministic soft surrogate for l0-constrained structured pruning, delivering 1% performance loss on Qwen3 models at 20% sparsity and faster convergence than prior methods.
citing papers explorer
-
From Execution to Education: A Bloom-Aligned Framework for Measuring Educational Control in LLMs
On 2,520 programming tasks, matched Qwen general and coder models reliably raise Bloom cognitive demand but fail to lower it, so execution skill does not imply educational control.
-
Subliminal Clocks: Latent Time Modelling in Diffusion Language Models
Diffusion language models encode a decodable, steerable representation of denoising progress (the fraction of unmasked tokens) in their residual streams.
-
Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding
Confident Decoding selects reliable intermediate layers via entropy-guided backward search to bypass final-layer alignment perturbations in LLMs, improving reasoning performance.
-
Contribution Weights: A Geometrical Analysis of Self-Attention Transformers
Contribution Weights combine attention, value magnitude, and directional alignment to measure token influence more faithfully than attention alone, and show attention sinks actively suppress information via a convex sink-rate to output-norm relationship.
-
Cornerstones or Stumbling Blocks? Deciphering the Rock Tokens in On-Policy Distillation
Rock Tokens in on-policy distillation persist at high loss, account for up to 18% of outputs, absorb large gradient norms, but add negligible value to reasoning performance.
-
Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task
Transformers show limited adaptive depth use on relational reasoning, with clearer evidence after finetuning on the task.
-
Deterministic Differentiable Structured Pruning for Large Language Models
DDP replaces stochastic hard-concrete masks with a deterministic soft surrogate for l0-constrained structured pruning, delivering 1% performance loss on Qwen3 models at 20% sparsity and faster convergence than prior methods.