LMs compare quantities via number and unit heuristics rather than exact conversion
A new arXiv paper investigates how language models handle quantity comparisons involving measurement units (e.g., 110 cm vs. 1.2 m) across multiple unit systems. The authors find that accuracy degrades near comparison boundaries and that LM behavior is explained by linear surrogate models using numerical-difference and unit-scale-difference cues. Causal interventions confirm these heuristics are mechanistically operative, suggesting LMs do not perform exact unit conversion but instead rely on a bag of numerical and unit-scale shortcuts.
Related guides (2)
Related events (8)
Systematic study reveals effectiveness-fluency trade-offs in LLM conditioning methods
A new arXiv paper systematically evaluates a range of LLM conditioning methods across both concept injection and removal scenarios, finding that efficient steering methods often degrade fluency significantly. A key finding is that activation steering is substantially less effective on instruction-tuned models than on base models, a previously overlooked interaction. Simple prompting and supervised fine-tuning work for concept injection but not removal, and cheap textual metrics are found to correlate well with expensive LLM-as-judge evaluations.
LLMs Show Inverted Compositional Strengths vs. Humans on Reference Resolution Task
This paper evaluates LLMs and humans on the Personal Relation Task (Paperno 2022), distinguishing between Extensional tasks (resolving what an expression refers to) and Intensional tasks (representing structured sense/formula). The study finds that humans outperform LLMs on Extensional tasks while LLMs outperform humans on Intensional tasks—an inverted pattern of strengths. The authors argue this asymmetry reflects the absence of referential grounding in LLM training as a key gap in human-like language understanding.
Benchmarking study finds LLMs fail at counterintuitive probability problems despite strong standard performance
A new arXiv paper evaluates 8 state-of-the-art LLMs on discrete probability problems using two datasets: standard exercises (average accuracy 0.96) and counterintuitive exercises designed to trigger heuristic reasoning (average accuracy 0.59). The authors document token bias causing 20%+ performance drops when canonical problem formulations are disguised, and up to 34% degradation when misleading suggestions are embedded in prompts. The findings argue that current LLMs are not genuine probabilistic reasoners despite their success on advanced math benchmarks.
Optimizing your LLM in production
A Hugging Face blog post covering practical techniques for optimizing large language models in production environments. The post likely addresses inference efficiency methods such as quantization, batching, caching, and hardware utilization strategies. It serves as a practitioner-oriented guide for deploying LLMs at scale.
Language Models Learn Constructional Semantics, Not To Mention Syntax: Investigating LM Understanding of Paired-Focus Constructions
This paper investigates whether language models can learn the semantics of rare English constructions (e.g., 'let alone', 'much less'), constructing a novel dataset to test form-meaning pairing understanding. Testing models across parameter counts, architectures, and pretraining dataset sizes, the authors find that modestly sized open-source models can grasp Paired-Focus construction semantics, while models trained on human-scale data fail. Training dynamics analysis reveals that semantic understanding of these constructions emerges later than syntactic knowledge and correlates with gains in world knowledge more broadly.
Mechanistic analysis of how LLMs encode essay quality in internal representations
Researchers systematically probe the hidden representations of eight LLMs across three essay datasets (ASAP++, CSEE, ENEM) to understand how automated essay scoring (AES) works internally. Using linear probing, dimensionality reduction, and neuron-level analysis, they find essay quality is encoded in a linearly accessible form that emerges progressively across layers and partially transfers across prompts. Individual 'essay scoring neurons' are identified whose activations correlate with scores and respond to targeted interventions, with longer essays relying more on deeper layers. The work contributes to mechanistic interpretability of LLM-based scoring systems.
SPACENUM: Revisiting Spatial Numerical Understanding in VLMs
SpaceNum is a new evaluation framework probing whether Vision-Language Models genuinely ground numerical outputs (coordinates, action magnitudes) in spatial perception, rather than relying on shallow cues. The benchmark defines two bidirectional tasks—Num2Space and Space2Num—across dynamic and static spatial settings. Results show current VLMs perform near random chance on spatial numerical grounding, with explicit reasoning providing only marginal improvement and fine-tuning offering partial gains.
Semantic vs. Surface Noise in LLM Agents: 68-Cell Measurement Study with Held-Out Validation
This paper documents an empirical phenomenon across 10 LLMs from 7 architecture families: meaning-bearing perturbations (paraphrase, synonym substitution) cause final-answer inconsistency ~19.69 percentage points more often than presentation-level perturbations (formatting, reordering) of comparable severity, across GSM8K, MATH, and HotpotQA benchmarks. The effect is validated on a held-out 11th model (qwen2.5-14B-Instruct) with 1,800 trajectories. Trace-level analysis supports a 'stealth-divergence' picture where semantic perturbations preserve the first action but induce divergence in intermediate reasoning steps, while two prior mechanism claims are explicitly retracted. The study is notable for its honest reporting of stress-test failures and pre-registered replication.