Can Large Language Models Recover Semantic Optimization Opportunities That Compilers Miss?
LLMs recover missing semantic optimizations in C/C++ code that compilers overlook, evaluated on SeGaBench with 120 cases.
Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
LLMs recover missing semantic optimizations in C/C++ code that compilers overlook, evaluated on SeGaBench with 120 cases.
Unconditional separations prove low-depth quantum circuits compute tasks intractable for constant-round diffusion and transformer LMs.
Layer-wise analysis shows sensitivity, causality, and repair capacity dissociate when LLMs fail on perturbed input; identifies spike-and-suppress vs. late-accumulation regimes.
LLM capability for testing Terminal User Interfaces: benchmark across ratatui/Rust, bubbletea/Go, textual/Python showing 12% coverage in real applications.
Self-augmentation method for MLLMs using model failure signals to generate targeted image augmentations without external supervision.
CARE-Bench evaluates 11 LLMs on medical triage safety, testing when patient-facing models recommend escalation to clinicians.
Empirical study across 13 models showing letter casing modulates attention allocation in LLMs and VLMs.
DiagLoop generates counterfactual training data from clinical guidelines to improve diagnostic LLM reasoning in low-data settings.
MuEvo uses LLMs to evolve multi-component heuristic ensembles for combinatorial optimization, addressing inter-component dependencies.
Don't be a meat proxy Niklas Gruhn coins an excellent new term - meat proxy - for people who blindly copy and paste the output of AI systems to their peers. By all means, prompt AI. But don't just relay the output. Read it, understand it, validate it, and then write a…
GradCuit enables test-time latent reasoning in LLMs via gradient flow through inserted continuous states, improving interpretability and credit assignment over decoded-token methods.
MedPRESS: 600-dialogue benchmark measuring patient-pressure-induced sycophancy in LLMs across medication, self-care, and triage scenarios.
Analysis of 18 open-source LLMs showing cultural bias in mythology knowledge; models encode cross-cultural distinctions in residual streams but fail to decode non-Western traditions.
CTRAG framework uses in-context retrieval with LLMs for automated regulatory compliance checking across financial, privacy, and cybersecurity domains.
Solution Hacking identifies shortcuts where LLMs achieve correct answers without valid reasoning on frontier science benchmarks.
Systematic comparison of training-free vs training-based intent classification methods for LLM prompt routing and domain specialization.
Empirical analysis of why frontier LLMs fail at tabular prediction without fine-tuning, foundational insight for tabular models.
Antares: compact LLMs (350M–3B) for agentic vulnerability localization via SFT and RL on cybersecurity reasoning over code.
Simon Willison argues LLMs make open-source dev tools more practical by lowering code-reading barriers through AI-assisted comprehension.
LLMs exhibit medical sycophancy—abandoning correct answers under user pushback—driven by conversational factors, not fixed model properties.
Attribute-level unlearning for MLLMs enables fine-grained removal of sensitive identity information while preserving model utility.
LLMs as examiners silently omit valid answers when authoring test sets; greedy one-shot generation fails to enumerate complete solution spaces.
FinHardBench evaluates LLM-generated FPGA hardware for financial trading (latency-critical 5-10ns); 33 tasks span module generation, pipeline tuning, adaptation.
Green offered a remarkable apology, saying that "the level of dopamine that I've been getting from interacting with LLMs ... is not healthy for me or good for the world."
FriendBench evaluates 26 multimodal LLMs on dyadic familiarity inference from video; top models match human performance but via different mechanisms.
Proposes optimization-path organization framework for parameter-efficient LLM fine-tuning reducing catastrophic forgetting across heterogeneous task sequences.
Benchmark comparing 10 LLMs and 6 prompting strategies for automated code generation in fluid system simulation (WNTR, Modelica).
As experts have warned for the last two years, some companies — like Microsoft and now Google — are finding and patching an exponential number of bugs in their products, thanks to the use of LLMs and AI tools.
Study shows LLMs reproduce standard language ideologies, privileging Inner Circle English norms and marginalizing non-dominant World Englishes variants.
LLMs exhibit salience bias, prioritizing explicit numerical distractors over implicit commonsense knowledge in reasoning tasks.