Daily Archive
Updated from current HFEPX corpus (Feb 27, 2026). 7 papers are grouped in this daily page. Common evaluation modes: Automatic Metrics, Simulation Env. Most common rater population: Domain Experts. Frequently cited benchmark: BrowseComp. Common metric signal: cost. Newest paper in this set is from Feb 11, 2026.
Updated from current HFEPX corpus (Feb 27, 2026). This page covers 7 papers centered on HFEPX Daily Archive: 2026-02-11. Common evaluation modes include Automatic Metrics, Simulation Env, with benchmark emphasis on BrowseComp, GSM8K. Use the anchored takeaways below to compare protocol choices and identify papers with stronger evidence depth.
Evaluation emphasis: Automatic Metrics and Simulation Env appear frequently in this slice.
Evidence: When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing , Embedding Inversion via Conditional Masked Diffusion Language Models , When Fusion Helps and When It Breaks: View-Aligned Robustness in Same-Source Financial Imaging
Benchmark concentration: BrowseComp, GSM8K helps control cross-paper variance.
Evidence: Embedding Inversion via Conditional Masked Diffusion Language Models , When Fusion Helps and When It Breaks: View-Aligned Robustness in Same-Source Financial Imaging , LoRA-Squeeze: Simple and Effective Post-Tuning and In-Tuning Compression of LoRA Modules
Metric concentration: cost, latency is repeatedly reported in this group.
Evidence: When Fusion Helps and When It Breaks: View-Aligned Robustness in Same-Source Financial Imaging , LoRA-Squeeze: Simple and Effective Post-Tuning and In-Tuning Compression of LoRA Modules , Search or Accelerate: Confidence-Switched Position Beam Search for Diffusion Language Models
Stratify by benchmark (BrowseComp vs GSM8K) before comparing methods.
Evidence: LoRA-Squeeze: Simple and Effective Post-Tuning and In-Tuning Compression of LoRA Modules , Search or Accelerate: Confidence-Switched Position Beam Search for Diffusion Language Models , Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters
Track metric sensitivity by reporting both cost and latency.
Evidence: Search or Accelerate: Confidence-Switched Position Beam Search for Diffusion Language Models , Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters , TestExplora: Benchmarking LLMs for Proactive Bug Discovery via Repository-Level Test Generation
Papers with explicit human feedback is visible in approximately 14.3% of papers in this set.
Evidence: Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters , TestExplora: Benchmarking LLMs for Proactive Bug Discovery via Repository-Level Test Generation , When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing
Papers with explicit human feedback
Coverage is a replication risk (14.3% vs 45% target).
Papers reporting quality controls
Coverage is a replication risk (0% vs 30% target).
Papers naming benchmarks/datasets
Coverage is usable but incomplete (28.6% vs 35% target).
Papers naming evaluation metrics
Coverage is a replication risk (14.3% vs 35% target).
Papers with known rater population
Coverage is a replication risk (14.3% vs 35% target).
Papers with known annotation unit
Coverage is a replication risk (0% vs 35% target).
1. When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing
Start with this anchor paper for scope and protocol framing. Covers Automatic Metrics.
2. Embedding Inversion via Conditional Masked Diffusion Language Models
Covers Automatic Metrics.
3. When Fusion Helps and When It Breaks: View-Aligned Robustness in Same-Source Financial Imaging
Covers Automatic Metrics.
4. LoRA-Squeeze: Simple and Effective Post-Tuning and In-Tuning Compression of LoRA Modules
Covers Automatic Metrics.
5. Search or Accelerate: Confidence-Switched Position Beam Search for Diffusion Language Models
Covers Automatic Metrics.
6. Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters
Covers Simulation Env. Includes human-feedback signal: Pairwise Preference.
7. TestExplora: Benchmarking LLMs for Proactive Bug Discovery via Repository-Level Test Generation
Covers Simulation Env.
automatic_metrics vs simulation_env
both=0, left_only=5, right_only=2
0 papers use both Automatic Metrics and Simulation Env.
Zachary Pedram Dadfar · Feb 11, 2026
Large language models produce rich introspective language when prompted for self-examination, but whether this language reflects internal computation or sophisticated confabulation has remained unclear.
Han Xiao · Feb 11, 2026
We frame embedding inversion as conditional masked diffusion, recovering all tokens in parallel through iterative denoising rather than sequential autoregressive generation.
Rui Ma · Feb 11, 2026
To control label ambiguity from near-zero moves, we use an ex-post minimum-movement threshold min_move (tau) based on realized absolute next-day return, defining an offline benchmark on the subset where the absolute next-day return is at le
Ivan Vulić, Adam Grycner, Quentin de Laroussilhe, Jonas Pfeiffer · Feb 11, 2026
Despite its huge number of variants, standard Low-Rank Adaptation (LoRA) is still a dominant technique for parameter-efficient fine-tuning (PEFT).
Mingyu Cao, Alvaro H. C. Correia, Christos Louizos, Shiwei Liu, Lu Yin · Feb 11, 2026
Across mathematical reasoning and code generation benchmarks (GSM8K, MBPP, HumanEval) on Dream-7B and LLaDA-8B, SOAR improves generation quality while maintaining competitive inference speed, offering a practical way to balance quality and
Ailin Huang, Ang Li, Aobo Kong, Bin Wang, Binxing Jiao · Feb 11, 2026
We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency.
Steven Liu, Jane Luo, Xin Zhang, Aofan Liu, Hao Liu · Feb 11, 2026
Current evaluations systematically overlook the third goal.
