Pith. sign in

REVIEW 60 cited by

Ministral 3 derives 3B, 8B, and 14B dense models through iterative pruning and distillation for constrained hardware.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-05-14 19:08 UTC pith:KM6EM2C5

load-bearing objection Ministral 3 is a model release announcement for 3B/8B/14B variants with image support via cascade distillation, but it supplies zero benchmarks or method details so the claims cannot be checked.

arxiv 2601.08584 v1 pith:KM6EM2C5 submitted 2026-01-13 cs.CL

Ministral 3

Alexander H. Liu , Kartik Khandelwal , Sandeep Subramanian , Victor Jouault , Abhinav Rastogi , Adrien Sad\'e , Alan Jeffares , Albert Jiang
show 111 more authors
Alexandre Cahill Alexandre Gavaudan Alexandre Sablayrolles Am\'elie H\'eliou Amos You Andy Ehrenberg Andy Lo Anton Eliseev Antonia Calvi Avinash Sooriyarachchi Baptiste Bout Baptiste Rozi\`ere Baudouin De Monicault Cl\'emence Lanfranchi Corentin Barreau Cyprien Courtot Daniele Grattarola Darius Dabert Diego de las Casas Elliot Chane-Sane Faruk Ahmed Gabrielle Berrada Ga\"etan Ecrepont Gauthier Guinet Georgii Novikov Guillaume Kunsch Guillaume Lample Guillaume Martin Gunshi Gupta Jan Ludziejewski Jason Rute Joachim Studnia Jonas Amar Jos\'ephine Delas Josselin Somerville Roberts Karmesh Yadav Khyathi Chandu Kush Jain Laurence Aitchison Laurent Fainsin L\'eonard Blier Lingxiao Zhao Louis Martin Lucile Saulnier Luyu Gao Maarten Buyl Margaret Jennings Marie Pellat Mark Prins Mathieu Poir\'ee Mathilde Guillaumin Matthieu Dinot Matthieu Futeral Maxime Darrin Maximilian Augustin Mia Chiquier Michel Schimpf Nathan Grinsztajn Neha Gupta Nikhil Raghuraman Olivier Bousquet Olivier Duchenne Patricia Wang Patrick von Platen Paul Jacob Paul Wambergue Paula Kurylowicz Pavankumar Reddy Muddireddy Philom\`ene Chagniot Pierre Stock Pravesh Agrawal Quentin Torroba Romain Sauvestre Roman Soletskyi Rupert Menneer Sagar Vaze Samuel Barry Sanchit Gandhi Siddhant Waghjale Siddharth Gandhi Soham Ghosh Srijan Mishra Sumukh Aithal Szymon Antoniak Teven Le Scao Th\'eo Cachet Theo Simon Sorg Thibaut Lavril Thiziri Nait Saada Thomas Chabal Thomas Foubert Thomas Robert Thomas Wang Tim Lawson Tom Bewley Tom Edwards Umar Jamil Umberto Tomasini Valeriia Nemychnikova Van Phung Vincent Maladi\`ere Virgile Richard Wassim Bouaziz Wen-Ding Li William Marshall Xinghui Li Xinyu Yang Yassine El Ouahidi Yihan Wang Yunhao Tang Zaccharie Ramzi
This is my paper
classification cs.CL
keywords dense language modelscascade distillationmodel pruningparameter efficientinstruction tuningreasoning modelsmultimodal capabilitiesmodel compression
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces the Ministral 3 family of dense language models sized at 3B, 8B, and 14B parameters, built specifically for applications with limited compute and memory. It details a derivation recipe called Cascade Distillation that repeatedly prunes the model and continues training with distillation to shrink size while aiming to keep performance. Each size offers three variants: a base pretrained model, an instruction-finetuned version, and a reasoning model, all equipped with image understanding. The work centers on releasing these under an open license so they can run where larger models cannot.

Core claim

Ministral 3 is a series of parameter-efficient dense language models at 3B, 8B, and 14B parameters obtained by Cascade Distillation, an iterative process of pruning followed by continued training with distillation, yielding base, instruction-tuned, and reasoning variants that each support image understanding.

What carries the argument

Cascade Distillation: the iterative pruning and continued training with distillation technique that shrinks model size while transferring capabilities from larger teachers.

Load-bearing premise

Iterative pruning plus distillation training preserves strong instruction following, reasoning, and image understanding at the reduced parameter counts.

What would settle it

Benchmark results showing the 3B Ministral 3 model scores more than 20 points below a comparable 7B model on standard instruction-following and multimodal reasoning tests.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • The three model sizes enable deployment on hardware that cannot host larger dense models.
  • Instruction-tuned variants directly support user command following without further adaptation.
  • Reasoning variants target complex multi-step problem solving at reduced cost.
  • Built-in image understanding extends the models to multimodal tasks without separate vision components.
  • Apache 2.0 release permits commercial and research reuse without licensing restrictions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same cascade process could be tested on even smaller targets such as 1B parameters to map the size-performance curve.
  • Combining cascade distillation with post-training quantization might produce further efficiency gains for edge devices.
  • The approach suggests a repeatable path for converting existing large models into families of progressively smaller siblings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

No mathematical derivations, free parameters, or background axioms are invoked. The only potential new element is the named 'Cascade Distillation' process, presented without external references or validation.

invented entities (1)
  • Cascade Distillation no independent evidence
    purpose: Iterative pruning combined with continued training via distillation to create smaller capable models
    Described as the authors' recipe in the abstract with no prior citations or independent evidence provided.

pith-pipeline@v0.9.0 · 5965 in / 1198 out tokens · 51648 ms · 2026-05-14T19:08:31.206940+00:00 · methodology

0 comments
read the original abstract

We introduce the Ministral 3 series, a family of parameter-efficient dense language models designed for compute and memory constrained applications, available in three model sizes: 3B, 8B, and 14B parameters. For each model size, we release three variants: a pretrained base model for general-purpose use, an instruction finetuned, and a reasoning model for complex problem-solving. In addition, we present our recipe to derive the Ministral 3 models through Cascade Distillation, an iterative pruning and continued training with distillation technique. Each model comes with image understanding capabilities, all under the Apache 2.0 license.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 60 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. EHRNote-ChatQA: A Benchmark for Evidence-Grounded Multi-Turn Clinical Question Answering over Longitudinal Discharge Summaries

    cs.CL 2026-06 unverdicted novelty 8.0

    EHRNote-ChatQA is the first benchmark for evidence-grounded multi-turn clinical QA over longitudinal discharge summaries, containing 16,072 medical-expert-verified pairs across eight categories and revealing LLM weakn...

  2. Toward Agentic SysAdmin: Rethinking System Administration with AI Agents

    cs.NI 2026-06 unverdicted novelty 7.0

    NetLLMeval is an emulation-based framework for benchmarking LLM solvers on network admin tasks, with a 24000-run study showing solver architecture lifts a 14B model from 0.43 to 0.88 accuracy and allows local models t...

  3. AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model

    cs.CV 2026-06 unverdicted novelty 7.0

    AMALIA-VL introduces the first open-source instruction-tuned LVLM natively optimized for European Portuguese via vision-language alignment, instruction tuning, preference optimization, and a pt-PT-centric data mix.

  4. PorTEXTO: A European Portuguese Benchmark for Visual Text Extraction

    cs.CV 2026-06 unverdicted novelty 7.0

    PorTEXTO benchmark shows sharp real-world performance drops in pt-PT OCR and finds specialized multilingual data outperforms model size or resolution increases.

  5. LEDGER: A Long-Context Benchmark of Corporate Annual Reports for Grounded Financial Retrieval and Extraction

    cs.CL 2026-06 unverdicted novelty 7.0

    LEDGER provides a corpus of 4,999 annual reports with 31 labeled KPIs and three benchmarks for page-level retrieval, needle-in-haystack lookup, and full KPI extraction from long documents.

  6. SurgiQ: A Large-Scale Multi-Domain Benchmark for Evaluating Surgical Understanding in Large Language Models

    cs.CL 2026-06 unverdicted novelty 7.0

    SurgiQ is a new 13k-question surgical benchmark showing general-purpose LLMs reach 68.1% accuracy while most biomedical models lag and smaller models stay near random baseline.

  7. Anchored, Not Graded: Vision-Language Models Fail at Slant-from-Texture Perception

    cs.CV 2026-06 unverdicted novelty 7.0

    VLMs exhibit anchoring to discrete slant angles rather than graded responses across zero-shot, in-context, and fine-tuned settings, unlike human psychophysical patterns.

  8. Anchored, Not Graded: Vision-Language Models Fail at Slant-from-Texture Perception

    cs.CV 2026-06 unverdicted novelty 7.0

    VLMs across families and scales show anchoring to discrete slant angles in zero-shot and prompted settings rather than human-like graded texture-based slant perception.

  9. Synthetic Personalities: How Well Can LLMs Mimic Individual Respondents Using Socio-Economic Microdata?

    cs.CY 2026-06 unverdicted novelty 7.0

    LLMs achieve up to 78.8% accuracy and r=0.590 correlation mimicking individual SOEP respondents using cumulative microdata, with gains from more information but diminishing returns past the 75% entropy point.

  10. Linear Ensembles Wash Away Watermarks: On the Fragility of Distributional Perturbations in LLMs

    cs.CL 2026-05 conditional novelty 7.0

    Averaging output distributions across 3-5 LLMs recovers the unwatermarked distribution, suppressing detection z-scores below threshold while improving quality.

  11. SliceWorld: A Predictive and Controllable World-State Model for CT Report Generation

    cs.CV 2026-05 unverdicted novelty 7.0

    SliceWorld introduces a world-state model for CT report generation that uses predictive and factor-aware objectives on axial slice sequences.

  12. Generative Conversational Recommender System

    cs.IR 2026-05 unverdicted novelty 7.0

    A single autoregressive model for conversational recommendation that uses semantic item IDs, predicts response intent and target first, then generates the response, reporting up to 29% Recall@1 gains.

  13. Llamas on the Web: Memory-Efficient, Performance-Portable, and Multi-Precision LLM Inference with WebGPU

    cs.DC 2026-05 conditional novelty 7.0

    LlamaWeb is a WebGPU backend for llama.cpp that uses static memory planning, tunable kernels, and templated multi-precision support to cut memory use by 29-33% and raise decode throughput by 45-69% versus prior browse...

  14. ECUAS$_n$: A family of metrics for principled evaluation of uncertainty-augmented systems

    cs.AI 2026-05 unverdicted novelty 7.0

    Introduces the ECUAS_n family of proper scoring rules for evaluating uncertainty-augmented systems, where n tunes the trade-off between prediction accuracy costs and uncertainty quality.

  15. To Call or Not to Call: Diagnosing Intrinsic Over-Calling Bias in LLM Agents

    cs.LG 2026-05 conditional novelty 7.0

    LLM agents have an intrinsic over-calling bias diagnosed via SAE activation margins and corrected by adaptive margin-calibrated steering, improving overall decision accuracy.

  16. Causal Bias Detection in Generative Artificial Intelligence

    cs.AI 2026-05 unverdicted novelty 7.0

    Develops a causal framework unifying generative AI fairness with standard ML, with new decompositions, identification conditions, and estimators demonstrated on LLM race and gender bias.

  17. K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs

    cs.CL 2026-05 conditional novelty 7.0

    K12-KGraph is a textbook-derived knowledge graph that powers a new benchmark revealing LLMs' poor curriculum cognition and a small training corpus that outperforms general instruction data on educational tasks.

  18. DocScope: Benchmarking Verifiable Reasoning for Trustworthy Long-Document Understanding

    cs.CL 2026-05 unverdicted novelty 7.0

    DocScope benchmarks long-document QA by requiring models to predict evidence pages, regions, facts, and answers, finding that even correct answers have complete evidence chains only 29% of the time at best.

  19. DocScope: Benchmarking Verifiable Reasoning for Trustworthy Long-Document Understanding

    cs.CL 2026-05 unverdicted novelty 7.0

    DocScope is a new benchmark for long-document understanding that audits models via four independent stages of reasoning trajectory, showing that correct answers frequently lack complete verifiable evidence chains with...

  20. Stateful Agent Backdoor

    cs.CR 2026-05 unverdicted novelty 7.0

    A stateful backdoor for LLM agents, modeled as a Mealy machine with a decomposition framework, enables incremental malicious actions across sessions and achieves 80-95% attack success rate on four models.

  21. When LLMs Stop Following Steps: A Diagnostic Study of Procedural Execution in Language Models

    cs.CL 2026-05 unverdicted novelty 7.0

    LLM first-answer accuracy on procedural arithmetic drops from 61% on 5-step tasks to 20% on 95-step tasks, with frequent failures including skipped steps, premature answers, and hallucinated operations.

  22. RESP: Reference-guided Sequential Prompting for Visual Glitch Detection in Video Games

    cs.CV 2026-04 unverdicted novelty 7.0

    RESP uses reference-guided sequential prompting with VLMs to improve frame-level and video-level visual glitch detection in games by establishing per-video baselines.

  23. BAS: A Decision-Theoretic Approach to Evaluating Large Language Model Confidence

    cs.CL 2026-04 unverdicted novelty 7.0

    BAS aggregates utility from an answer-or-abstain model across risk thresholds and is uniquely maximized by truthful confidence estimates.

  24. LLM-ODE: Data-driven Discovery of Dynamical Systems with Large Language Models

    cs.LG 2026-03 unverdicted novelty 7.0

    LLM-ODE integrates large language models into genetic programming to guide symbolic search for governing equations of dynamical systems, outperforming classical GP on 91 test cases in efficiency and solution quality.

  25. Reasoning over Video: Evaluating How MLLMs Extract, Integrate, and Reconstruct Spatiotemporal Evidence

    cs.CV 2026-03 unverdicted novelty 7.0

    VAEX-BENCH shows state-of-the-art MLLMs perform substantially worse on abstractive spatiotemporal reasoning tasks than on matched extractive tasks in video understanding.

  26. Beyond One-Size-Fits-All: Adaptive Subgraph Denoising for Zero-Shot Graph Learning with Large Language Models

    cs.LG 2026-03 unverdicted novelty 7.0

    GraphSSR introduces an adaptive SSR pipeline with SSR-SFT data synthesis and SSR-RL (Authenticity-Reinforced and Denoising-Reinforced stages) to overcome one-size-fits-all subgraph noise in zero-shot LLM graph reasoning.

  27. MosaicKV: Serving Long-Context LLM with Dynamic Two-D KV Cache Compression

    cs.LG 2026-07 unverdicted novelty 6.0

    MosaicKV achieves up to 16x attention speedup, 4.8x lower decode latency, 7.3x higher throughput, and 3x memory reduction with 1.76% accuracy loss via dynamic two-D KV cache compression and management on H800 GPUs.

  28. AURORA: Asymmetry and Update-Induced Rotation for Robust Hallucination Detection in Large Language Models

    cs.CL 2026-06 unverdicted novelty 6.0

    AURORA detects hallucinations via skewness of cosine similarities between weights and gradients plus a rotation ratio from SVD on update-induced changes to singular vectors.

  29. What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

    cs.LG 2026-06 unverdicted novelty 6.0

    Proposes SCSuff metric for evaluating LLM explanation sufficiency via model-generated alternative inputs, showing explanations are typically insufficient and predictable from hidden states.

  30. AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model

    cs.CV 2026-06 unverdicted novelty 6.0

    AMALIA-VL is the first open-source LVLM natively optimized for European Portuguese via three-stage training on a pt-PT-centric data mix combining curated, translated, and novel datasets.

  31. AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model

    cs.CV 2026-06 unverdicted novelty 6.0

    Introduces AMALIA-VL, the first open-source instruction-tuned LVLM for European Portuguese, using a high-resolution vision encoder, pt-PT language model, learned connector, and three-stage training on a custom data mix.

  32. PorTEXTO: A European Portuguese Benchmark for Visual Text Extraction

    cs.CV 2026-06 unverdicted novelty 6.0

    PorTEXTO is the first benchmark for contemporary European Portuguese visual text extraction, documenting performance gaps between synthetic and real data and the advantage of specialized multilingual training.

  33. Grounding Spoken LLMs in Multi-Speaker Audio via Diarization Conditioning

    eess.AS 2026-06 unverdicted novelty 6.0

    Dixtral uses diarization conditioning on a Whisper-based encoder within Voxtral to outperform baselines on multi-speaker transcription and match or exceed on QA tasks.

  34. Soft-Prompt Tuning for Fair and Efficient LLM Benchmark Evaluation

    cs.CL 2026-06 unverdicted novelty 6.0

    Soft-prompt tuning with 10 vectors improves format compliance on LLM benchmarks and provides a low-cost proxy for comparing base models.

  35. Beyond Coverage and Kill Scores: Empirically Measuring Test Suite Behavioural Gaps

    cs.SE 2026-06 unverdicted novelty 6.0

    An empirical study extracts 20,729 expected behaviors from ten Java libraries and finds 17.5% remain untested, independent of line coverage and mutation scores.

  36. Reasoning Arena: Trace Tournaments When Verifiable Rewards Fall Short

    cs.LG 2026-06 unverdicted novelty 6.0

    Reasoning Arena converts non-diverse reward groups in RLVR into relative rewards via adaptive trace tournaments and Bradley-Terry fitting on anchor comparisons, claiming 7.6% average gains and 27-41% faster training o...

  37. PACT: Learning Diverse Diagnostic Strategies via Privileged Synthesis and Branch Consensus

    cs.CL 2026-06 unverdicted novelty 6.0

    PACT combines privileged multi-paradigm dialogue synthesis from EMRs with consensus aggregation of paradigm-specific LoRA branches to reach SOTA on a new Chinese interactive medical diagnosis benchmark.

  38. The Masked Advantage: Uncovering Local-Language Access to Cultural Knowledge in LLMs

    cs.CL 2026-06 unverdicted novelty 6.0

    Using a 1PL IRT model on real cultural questions across 13 locales, the study identifies a local-language knowledge-access advantage masked by lower proficiency in raw accuracy.

  39. Compress-Distill: Reasoning Trace Compression for Efficient Knowledge Distillation

    cs.LG 2026-06 unverdicted novelty 6.0

    Post-hoc model-based compression of reasoning traces cuts training tokens to 12-30% and speeds training 2-7.6x while retaining up to 96% of raw-trace accuracy, though raw traces remain superior at every scale.

  40. Deep Research as Rubric for Reinforcement Learning

    cs.CL 2026-05 unverdicted novelty 6.0

    DR-rubric is a two-stage framework using iterative agentic search to generate atomic verifiable constraints for GRPO-based RL, achieving competitive performance on 6 benchmarks with 1K-3K examples via bootstrap or fro...

  41. EHRBench: An Automated and Reliable EHR-based Benchmark for Clinical Decision Making with LLMs

    cs.AI 2026-05 unverdicted novelty 6.0

    EHRBench uses an EHR-LLM-KB pipeline to automatically create 960,067 reliable QA items spanning diagnosis, treatment, and prognosis for large-scale LLM evaluation in clinical decision making.

  42. Pruning and Distilling Mixture-of-Experts into Dense Language Models

    cs.CL 2026-05 unverdicted novelty 6.0

    A systematic MoE-to-dense conversion via expert scoring, grouping, and distillation yields +6.3 pp average accuracy over dense-to-dense pruning at matched parameter count on tested models.

  43. An Efficient and Privacy-Preserving Architecture for Cross-Institutional Collaborative RAG

    cs.CR 2026-05 unverdicted novelty 6.0

    FedRAG uses a Scrambled Distributed Attention protocol with feature scrambling and token permutation to enable high-throughput, privacy-preserving federated RAG without special hardware or retraining.

  44. ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention

    cs.LG 2026-05 unverdicted novelty 6.0

    ThriftAttention recovers 89.1% of the FP16 quality gap versus pure FP4 attention by running only 5% of query-key blocks in FP16 on long-context benchmarks.

  45. TempGlitch: Evaluating Vision-Language Models for Temporal Glitch Detection in Gameplay Videos

    cs.CV 2026-05 unverdicted novelty 6.0

    TempGlitch is a controlled benchmark showing that 12 evaluated VLMs perform near chance level on detecting five types of temporal glitches in gameplay videos, with denser sampling and larger models providing no reliab...

  46. ECUAS$_n$: A family of metrics for principled evaluation of uncertainty-augmented systems

    cs.AI 2026-05 unverdicted novelty 6.0

    ECUAS_n is a parameterized family of proper scoring rules for jointly assessing prediction accuracy and uncertainty quality in automated decision systems.

  47. ECUAS$_n$: A family of metrics for principled evaluation of uncertainty-augmented systems

    cs.AI 2026-05 unverdicted novelty 6.0

    Proposes ECUAS_n metrics as proper scoring rules for evaluating uncertainty-augmented systems, with n controlling cost trade-offs between predictions and uncertainties.

  48. TRACE: Trajectory Correction from Cross-layer Evidence for Hallucination Reduction

    cs.AI 2026-05 unverdicted novelty 6.0

    TRACE uses cross-layer candidate trajectories inside frozen LLMs to dynamically select and apply one of three correction operators, delivering mean gains of +12.26 MC1 and +8.65 MC2 points across 15 models and 3 bench...

  49. Towards Human-Level Book-Writing Capability

    cs.AI 2026-05 unverdicted novelty 6.0

    A prompt-to-book training framework that derives hierarchical summaries from public-domain novels and inverts them to supervise long-context models toward human literary prose instead of assistant-style output.

  50. Towards Human-Level Book-Writing Capability

    cs.AI 2026-05 unverdicted novelty 6.0

    A supervised fine-tuning approach using inverted multi-resolution planning scaffolds from public-domain novels trains models to generate book-length stories with more human-like literary qualities than standard instru...

  51. Jobs' AI Exposure Should Be Measured from Evidence, Not Model Priors

    cs.IR 2026-05 conditional novelty 6.0

    The authors propose a retrieval-augmented framework that grounds AI exposure labels for 18,796 O*NET occupation-task pairs in retrieved news and academic abstracts, outperforming zero-shot prompting in 72% of disagree...

  52. PanoWorld: Towards Spatial Supersensing in 360$^\circ$ Panorama World

    cs.CV 2026-05 unverdicted novelty 6.0

    PanoWorld adds spherical geometry to MLLMs via cross-attention and pano-specific instruction data, yielding better performance on panoramic spatial reasoning benchmarks than standard perspective-based pipelines.

  53. PanoWorld: Towards Spatial Supersensing in 360$^\circ$ Panorama World

    cs.CV 2026-05 unverdicted novelty 6.0

    PanoWorld adds spherical spatial cross-attention and pano-native training data to MLLMs for improved spatial reasoning on ERP panoramas, outperforming baselines on new and existing benchmarks.

  54. On-Policy Self-Evolution via Failure Trajectories for Agentic Safety Alignment

    cs.AI 2026-05 unverdicted novelty 6.0

    FATE lets LLM agents self-evolve safer behaviors by generating and filtering repairs from their own failure trajectories using verifiers and Pareto optimization.

  55. PRISM: A Geometric Risk Bound that Decomposes Drift into Scale, Shape, and Head

    cs.CL 2026-05 unverdicted novelty 6.0

    PRISM supplies a geometric upper bound on LLM variant risk that splits drift into scale, shape, and head axes and doubles as a differentiable regularizer against forgetting.

  56. Causal Bias Detection in Generative Artificial Intelligence

    cs.AI 2026-05 unverdicted novelty 6.0

    A causal framework unifies fairness analysis across generative AI and standard ML by deriving decompositions that separate biases along causal pathways and differences between real-world and model mechanisms.

  57. Mirror, Mirror on the Wall: Can VLM Agents Tell Who They Are at All?

    cs.AI 2026-05 unverdicted novelty 6.0

    Stronger VLM agents use mirror reflections for self-identification in controlled 3D tests, while weaker ones inspect but fail to extract or correctly attribute self-relevant information.

  58. LLMs are not (consistently) Bayesian: Quantifying internal (in)consistencies of LLMs' probabilistic beliefs

    cs.LG 2026-05 unverdicted novelty 6.0

    LLMs show inconsistent belief updates from evidence, with learned heuristics sometimes beating exact Bayesian computation due to misspecified world models.

  59. LLMs are not (consistently) Bayesian: Quantifying internal (in)consistencies of LLMs' probabilistic beliefs

    cs.LG 2026-05 unverdicted novelty 6.0

    LLMs do not consistently perform Bayesian updates on probabilistic beliefs; heuristic approaches often outperform exact Bayesian computation on downstream tasks, indicating misspecified internal models of the world.

  60. Skill Neologisms: Towards Skill-based Continual Learning

    cs.LG 2026-05 unverdicted novelty 6.0

    Skill neologisms are optimized soft tokens that improve LLM performance on targeted skills without weight updates and allow zero-shot composition for continual learning.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages · cited by 78 Pith papers · 18 internal anchors

  1. [1]

    Pixtral 12B

    Pravesh Agrawal, Szymon Antoniak, Emma Bou Hanna, Baptiste Bout, Devendra Chaplot, Jessica Chudnovsky, Diogo Costa, Baudouin De Monicault, Saurabh Garg, Theophile Gervet, et al. Pixtral 12b.arXiv preprint arXiv:2410.07073,

    work page internal anchor Pith review arXiv

  2. [2]

    GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints

    Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebrón, and Sumit Sanghai. Gqa: Training generalized multi-query transformer models from multi-head checkpoints. arXiv preprint arXiv:2305.13245,

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    Program Synthesis with Large Language Models

    Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, et al. Program synthesis with large language models.arXiv preprint arXiv:2108.07732,

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [4]

    Qwen3-VL Technical Report

    URLhttps://arxiv.org/abs/2511.21631. Dan Busbridge, Amitis Shidani, Floris Weers, Jason Ramapuram, Etai Littwin, and Russ Webb. Distillation scaling laws.arXiv preprint arXiv:2502.08606,

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [5]

    Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge. arXiv preprint arXiv:1803.05457,

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [6]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    URLhttps://arxiv.org/abs/2501.12948. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv e-prints, pages arXiv–2407,

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Distilled pretraining: A modern lens of data, in-context learning and test-time scaling.arXiv preprint arXiv:2509.01649, 2025

    URLhttps://arxiv.org/abs/2509.01649. Shangmin Guo, Biao Zhang, Tianlin Liu, Tianqi Liu, Misha Khalman, Felipe Llinares, Alexandre Rame, Thomas Mesnard, Yao Zhao, Bilal Piot, Johan Ferret, and Mathieu Blondel. Direct language model alignment from online ai feedback,

    work page arXiv

  8. [8]

    Direct Language Model Alignment from Online AI Feedback.arXiv Preprint arXiv:2402.04792, 2024

    URLhttps://arxiv.org/abs/2402.04792. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding.arXiv preprint arXiv:2009.03300,

    work page arXiv 2009

  9. [9]

    Measuring Mathematical Problem Solving With the MATH Dataset

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset.arXiv preprint arXiv:2103.03874,

    work page internal anchor Pith review Pith/arXiv arXiv

  10. [10]

    LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code

    Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida Wang, Armando Solar-Lezama, Koushik Sen, and Ion Stoica. Livecodebench: Holistic and contamination free evaluation of large language models for code.arXiv preprint arXiv:2403.07974,

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension

    Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension.arXiv preprint arXiv:1705.03551,

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    Gemma 3 Technical Report

    URL https://arxiv.org/abs/2503.19786. Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, et al. Natural questions: a benchmark for question answering research.Transactions of the Association for Computational Linguistics, 7:453–466,

    work page internal anchor Pith review Pith/arXiv arXiv

  13. [13]

    Race: Large-scale reading comprehension dataset from examinations

    Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang, and Eduard Hovy. Race: Large-scale reading comprehension dataset from examinations. InProceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 785–794,

    work page 2017

  14. [14]

    From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline

    12 Tianle Li, Wei-Lin Chiang, Evan Frick, Lisa Dunlap, Tianhao Wu, Banghua Zhu, Joseph E Gonzalez, and Ion Stoica. From crowdsourced data to high-quality benchmarks: Arena-hard and benchbuilder pipeline.arXiv preprint arXiv:2406.11939,

    work page Pith review arXiv

  15. [15]

    Wildbench: Benchmarking llms with challenging tasks from real users in the wild, 2024 a

    Bill Yuchen Lin, Yuntian Deng, Khyathi Chandu, Faeze Brahman, Abhilasha Ravichander, Valentina Pyatkin, Nouha Dziri, Ronan Le Bras, and Yejin Choi. Wildbench: Benchmarking llms with challenging tasks from real users in the wild.arXiv preprint arXiv:2406.04770,

    work page arXiv

  16. [16]

    Phybench: Holistic evaluation of physical perception and reasoning in large language models, 2025

    Zihan Liu, Zijian Wang, Yue Zhang, Jianing Wang, Jian Tang, Xiang He, and Xiangyu Zhang. Phybench: Holistic evaluation of physical perception and reasoning in large language models. arXiv preprint arXiv:2504.16074,

    work page arXiv

  17. [17]

    Compact language models via pruning and knowledge distillation

    Saurav Muralidharan, Sharath Turuvekere Sreenivas, Raviraj Joshi, Marcin Chochowski, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro, Jan Kautz, and Pavlo Molchanov. Compact language models via pruning and knowledge distillation.arXiv preprint arXiv:2407.14679,

    work page arXiv

  18. [18]

    Nakanishi

    Ken M Nakanishi. Scalable-softmax is superior for attention.arXiv preprint arXiv:2501.19399,

    work page arXiv

  19. [19]

    YaRN: Efficient Context Window Extension of Large Language Models

    Bowen Peng, Jeffrey Quesnelle, Honglu Fan, and Enrico Shippole. Yarn: Efficient context window extension of large language models.arXiv preprint arXiv:2309.00071,

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    Are we done with mmlu?arXiv preprint arXiv:2406.04127, 2024

    Aryo Perez, Tomasz Stanislawek, Andrzej Pohl, Kamil Dwojak, Dawid Jurkiewicz, Piotr Kobus, and Tomasz Trzci´nski. Are we done with mmlu?arXiv preprint arXiv:2406.04127,

    work page arXiv

  21. [21]

    Direct Preference Optimization: Your Language Model is Secretly a Reward Model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model.arXiv preprint arXiv:2305.18290,

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    arXiv preprint arXiv:2506.10910 , year=

    Abhinav Rastogi, Albert Q. Jiang, Andy Lo, Gabrielle Berrada, Guillaume Lample, et al. Magistral. arXiv preprint arXiv:2506.10910,

    work page arXiv

  23. [23]

    DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

    URLhttps://arxiv.org/abs/2402.03300. Noam Shazeer. Glu variants improve transformer.arXiv preprint arXiv:2002.05202,

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  24. [24]

    arXiv preprint arXiv:2408.11796 , url=

    13 Sharath Turuvekere Sreenivas, Saurav Muralidharan, Raviraj Joshi, Marcin Chochowski, Ameya Sunil Mahabaleshwarkar, Gerald Shen, Jiaqi Zeng, Zijia Chen, Yoshi Suhara, Shizhe Diao, Chenhan Yu, Wei-Chun Chen, Hayley Ross, Oluwatobi Olabiyi, Ashwath Aithal, Oleksii Kuchaiev, Daniel Korzekwa, Pavlo Molchanov, Mostofa Patwary, Mohammad Shoeybi, Jan Kautz, an...

    work page arXiv

  25. [25]

    RoFormer: Enhanced Transformer with Rotary Position Embedding

    Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.arXiv preprint arXiv:2104.09864,

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    A Simple and Effective Pruning Approach for Large Language Models

    Mingjie Sun, Zhuang Liu, Anna Bair, and J Zico Kolter. A simple and effective pruning approach for large language models.arXiv preprint arXiv:2306.11695,

    work page internal anchor Pith review arXiv

  27. [27]

    URLhttps://arxiv.org/abs/2505.09388. Xiang Yue, Yuansheng Ni, Kai Zhang, Tianyu Zheng, Ruoqi Liu, Ge Zhang, Samuel Stevens, Dongfu Jiang, Weiming Ren, Yuxuan Sun, Cong Wei, Botao Yu, Ruibin Yuan, Renliang Sun, Ming Yin, Boyuan Zheng, Zhenzhu Yang, Yibo Liu, Wenhao Huang, Huan Sun, Yu Su, and Wenhu Chen. Mmmu: A massive multi-discipline multimodal understa...

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [28]

    HellaSwag: Can a Machine Really Finish Your Sentence?

    Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. Hellaswag: Can a machine really finish your sentence?arXiv preprint arXiv:1905.07830,

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  29. [29]

    AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models

    Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen, and Nan Duan. Agieval: A human-centric benchmark for evaluating foundation models. arXiv preprint arXiv:2304.06364,

    work page internal anchor Pith review arXiv