scan-v5.json (27604B)
1 { 2 "scan_version": 5, 3 "paper_type": "empirical", 4 "paper": { 5 "title": "DeepReview: Improving LLM-based Paper Review with Human-like Deep Thinking Process", 6 "authors": [ 7 "Minjun Zhu", 8 "Yixuan Weng", 9 "Linyi Yang", 10 "Yue Zhang" 11 ], 12 "year": 2025, 13 "venue": "Annual Meeting of the Association for Computational Linguistics", 14 "arxiv_id": "2503.08569", 15 "doi": "10.48550/arXiv.2503.08569" 16 }, 17 "checklist": { 18 "claims_and_evidence": { 19 "abstract_claims_supported": { 20 "applies": true, 21 "answer": true, 22 "justification": "All major abstract claims are backed by tables: 88.21%/80.20% win rates against GPT-o1/DeepSeek-R1 appear in Table 4, and outperformance over CycleReviewer-70B on MSE appears in Table 2. Resources are released at ai-researcher.net.", 23 "source": "haiku" 24 }, 25 "causal_claims_justified": { 26 "applies": true, 27 "answer": false, 28 "justification": "The paper attributes adversarial robustness causally to the multi-stage framework ('we attribute this robustness to DeepReviewer's multi-stage reasoning framework') but runs no ablation removing stages to test robustness specifically; the fast/standard/best mode ablation tests accuracy, not robustness.", 29 "source": "haiku" 30 }, 31 "generalization_bounded": { 32 "applies": true, 33 "answer": false, 34 "justification": "The paper claims to 'set a new benchmark for LLM-based paper review' but evaluation is limited exclusively to ICLR 2024/2025 ML papers; no discussion of whether findings transfer to other venues, disciplines, or review formats.", 35 "source": "haiku" 36 }, 37 "alternative_explanations_discussed": { 38 "applies": true, 39 "answer": false, 40 "justification": "Key alternative explanations are absent: (1) DeepReviewer is trained on the same ICLR distribution as the test set, giving a distributional advantage over general-purpose LLMs; (2) using Gemini-2.0-Flash-Thinking as judge while also including it as a baseline creates a potential self-preference artifact that is not analyzed.", 41 "source": "haiku" 42 }, 43 "proxy_outcome_distinction": { 44 "applies": true, 45 "answer": false, 46 "justification": "Rating MSE (predicting reviewer scores) is the primary quantitative metric, but the paper does not explicitly discuss the gap between score prediction accuracy and actual review utility (catching errors, actionable suggestions); LLM-as-judge qualitative evaluation partially addresses this but without acknowledging the proxy limitation.", 47 "source": "haiku" 48 } 49 }, 50 "limitations_and_scope": { 51 "limitations_section_present": { 52 "applies": true, 53 "answer": true, 54 "justification": "A dedicated 'Limitations' section appears after the Conclusions, covering synthetic data quality, computational cost of Best mode, and incomplete adversarial robustness.", 55 "source": "haiku" 56 }, 57 "threats_to_validity_specific": { 58 "applies": true, 59 "answer": true, 60 "justification": "The limitations name specific threats: synthetic training data may not capture human review nuances, Best mode is computationally intensive, and adversarial robustness is incomplete — these go beyond generic boilerplate.", 61 "source": "haiku" 62 }, 63 "scope_boundaries_stated": { 64 "applies": true, 65 "answer": false, 66 "justification": "The paper does not state that results apply only to ML/AI conference reviews in ICLR format; no explicit scope boundary distinguishes what the findings do not show (e.g., other venues, disciplines, or review styles).", 67 "source": "haiku" 68 } 69 }, 70 "conflicts_of_interest": { 71 "funding_disclosed": { 72 "applies": true, 73 "answer": true, 74 "justification": "Corresponding author footnote states 'Supported by Research Center for Industries of the Future, Westlake University,' disclosing institutional support.", 75 "source": "haiku" 76 }, 77 "affiliations_disclosed": { 78 "applies": true, 79 "answer": true, 80 "justification": "All four authors have affiliations listed: Zhejiang University, Westlake University School of Engineering, and University College London.", 81 "source": "haiku" 82 }, 83 "funder_independent_of_outcome": { 84 "applies": true, 85 "answer": true, 86 "justification": "The funder is Westlake University's research center, an academic institution with no direct commercial interest in the evaluated system's performance.", 87 "source": "haiku" 88 }, 89 "financial_interests_declared": { 90 "applies": true, 91 "answer": false, 92 "justification": "No competing interests statement, patent declarations, or equity/consulting disclosures appear anywhere in the paper.", 93 "source": "haiku" 94 } 95 }, 96 "scope_and_framing": { 97 "key_terms_defined": { 98 "applies": true, 99 "answer": false, 100 "justification": "'Deep thinking,' 'expert reviewer,' and 'human-like' are used in the title and throughout but never precisely defined; the paper operationalizes stages but does not formally define what distinguishes the framework from prior structured prompting approaches.", 101 "source": "haiku" 102 }, 103 "intended_contribution_clear": { 104 "applies": true, 105 "answer": true, 106 "justification": "The introduction and conclusion explicitly enumerate three contributions: DeepReview-13K dataset, DeepReviewer-14B model, and DeepReview-Bench benchmark, alongside the multi-stage framework design.", 107 "source": "haiku" 108 }, 109 "engagement_with_prior_work": { 110 "applies": true, 111 "answer": true, 112 "justification": "Section 2 (Related Work) actively situates the paper against CycleReviewer, AI Scientist, AgentReview, and LLM reasoning literature, explaining how DeepReview extends or differs from each.", 113 "source": "haiku" 114 } 115 } 116 }, 117 "type_checklist": { 118 "empirical": { 119 "artifacts": { 120 "code_released": { 121 "applies": true, 122 "answer": true, 123 "justification": "Code repository (zhu-minjun/Researcher), model weights (DeepReviewer-7B and 14B), dataset (DeepReview-13K), and demo are released at ai-researcher.net.", 124 "source": "haiku" 125 }, 126 "data_released": { 127 "applies": true, 128 "answer": true, 129 "justification": "DeepReview-13K (13,378 training samples) and DeepReview-Bench (1,286 test samples) are stated to be publicly available.", 130 "source": "haiku" 131 }, 132 "environment_specified": { 133 "applies": true, 134 "answer": false, 135 "justification": "Training hardware (8x H100 80G, DeepSpeed+ZeRO3) is mentioned but no requirements.txt, Dockerfile, or package version list is provided.", 136 "source": "haiku" 137 }, 138 "reproduction_instructions": { 139 "applies": true, 140 "answer": false, 141 "justification": "No step-by-step reproduction instructions are included in the paper; readers are pointed to the repository without guidance on replicating experiments.", 142 "source": "haiku" 143 } 144 }, 145 "statistical_methodology": { 146 "confidence_intervals_or_error_bars": { 147 "applies": true, 148 "answer": false, 149 "justification": "No confidence intervals or error bars are reported in Tables 2, 3, or 4; all results are point estimates.", 150 "source": "haiku" 151 }, 152 "significance_tests": { 153 "applies": true, 154 "answer": false, 155 "justification": "No statistical significance tests are applied to any comparative claims despite large tables of numerical comparisons.", 156 "source": "haiku" 157 }, 158 "effect_sizes_reported": { 159 "applies": true, 160 "answer": true, 161 "justification": "Percentage improvements are reported with baseline context (e.g., 'Rating MSE: 44.80% ↑', '65.83% reduction' vs. prompt-based baselines), giving interpretable effect sizes.", 162 "source": "haiku" 163 }, 164 "sample_size_justified": { 165 "applies": true, 166 "answer": false, 167 "justification": "Test sets of 652 and 634 papers for ICLR 2024/2025 are used without power analysis or justification of sample adequacy.", 168 "source": "haiku" 169 }, 170 "variance_reported": { 171 "applies": true, 172 "answer": false, 173 "justification": "No standard deviations, variance, or cross-run statistics are reported; all tables present single-run point estimates.", 174 "source": "haiku" 175 } 176 }, 177 "evaluation_design": { 178 "baselines_included": { 179 "applies": true, 180 "answer": true, 181 "justification": "Two classes of baselines: prompt-based (AI Scientist, AgentReview with GPT-o1, Claude-3.5-Sonnet, Gemini, DeepSeek-V3/R1) and fine-tuned (CycleReviewer 8B and 70B).", 182 "source": "haiku" 183 }, 184 "baselines_contemporary": { 185 "applies": true, 186 "answer": true, 187 "justification": "Baselines include GPT-o1-2024-12-17, Claude-3.5-sonnet-20241022, Gemini-2.0-Flash-Thinking-01-21, DeepSeek-R1 — all state-of-the-art at submission time.", 188 "source": "haiku" 189 }, 190 "ablation_study": { 191 "applies": true, 192 "answer": true, 193 "justification": "Section 5.5 ablates reasoning depth (Fast/Standard/Best modes) and reviewer count (R=1 to R=6), showing per-metric impact of each component.", 194 "source": "haiku" 195 }, 196 "multiple_metrics": { 197 "applies": true, 198 "answer": true, 199 "justification": "Quantitative metrics cover MSE, MAE, Decision Accuracy, F1, Spearman correlation, and Pairwise Accuracy; qualitative metrics add LLM-as-judge win rates across five dimensions.", 200 "source": "haiku" 201 }, 202 "human_evaluation": { 203 "applies": true, 204 "answer": false, 205 "justification": "Qualitative evaluation uses Gemini-2.0-Flash-Thinking as judge, not human annotators; no humans evaluated the review text outputs.", 206 "source": "haiku" 207 }, 208 "held_out_test_set": { 209 "applies": true, 210 "answer": true, 211 "justification": "10% of the dataset (1,286 samples) was randomly held out as DeepReview-Bench, separate from the 13,378 training samples.", 212 "source": "haiku" 213 }, 214 "per_category_breakdown": { 215 "applies": true, 216 "answer": true, 217 "justification": "Table 3 breaks down results by Soundness, Presentation, and Contribution dimensions; Table 4 breaks win rates by constructive value, analytical depth, plausibility, and technical accuracy.", 218 "source": "haiku" 219 }, 220 "failure_cases_discussed": { 221 "applies": true, 222 "answer": false, 223 "justification": "The adversarial attack section notes a small score increase under attack (5.38→5.69) but presents no systematic failure case analysis or qualitative examples of where the model produces poor reviews.", 224 "source": "haiku" 225 }, 226 "negative_results_reported": { 227 "applies": true, 228 "answer": true, 229 "justification": "Performance variability in Reviewer Scaling (R≠4) is explicitly noted; DeepReviewer's relative weakness versus Gemini in technical accuracy (showing 20.79% baseline win rate) is reported in Table 4.", 230 "source": "haiku" 231 } 232 }, 233 "setup_transparency": { 234 "model_versions_specified": { 235 "applies": true, 236 "answer": true, 237 "justification": "Exact model versions are specified: GPT-o1-2024-12-17, Claude-3.5-sonnet-20241022, Gemini-2.0-Flash-Thinking-01-21; training backbone is Phi-4 14B (Abdin et al., 2024).", 238 "source": "haiku" 239 }, 240 "prompts_provided": { 241 "applies": true, 242 "answer": true, 243 "justification": "Figures 4, 5, 6, and 7 in the appendix provide full system prompts for the judge, review enhancement, paper analysis, and reliability verification stages.", 244 "source": "haiku" 245 }, 246 "hyperparameters_reported": { 247 "applies": true, 248 "answer": true, 249 "justification": "Training: 23,500 steps, batch size 16, learning rate 5e-6, 40K context window, 256K with LongRoPE; inference: temperature 0.4, max input 100K tokens, max output 16,384 tokens.", 250 "source": "haiku" 251 }, 252 "scaffolding_described": { 253 "applies": true, 254 "answer": true, 255 "justification": "The three-stage scaffold (novelty verification with literature retrieval, multi-dimension review, reliability verification with evidence chains) is described in detail in Section 4.2, including model assignments for each stage.", 256 "source": "haiku" 257 }, 258 "data_preprocessing_documented": { 259 "applies": true, 260 "answer": true, 261 "justification": "Section 3.1 documents PDF conversion via MinerU, LATEX source prioritization from arXiv, empty PDF filtering, and the automated quality control using Qwen-2.5-72B-Instruct.", 262 "source": "haiku" 263 } 264 }, 265 "data_integrity": { 266 "raw_data_available": { 267 "applies": true, 268 "answer": true, 269 "justification": "DeepReview-13K dataset is stated to be publicly released at ai-researcher.net with source papers from OpenReview/arXiv.", 270 "source": "haiku" 271 }, 272 "data_collection_described": { 273 "applies": true, 274 "answer": true, 275 "justification": "Section 3.1 describes collection of 18,976 papers from OpenReview across ICLR 2024-2025, the three components assembled per paper (textual assessments, rebuttal discussions, standardized scores), and filtering to 13,378 valid samples.", 276 "source": "haiku" 277 }, 278 "recruitment_methods_described": { 279 "applies": false, 280 "answer": false, 281 "justification": "No human participant recruitment; data was collected from OpenReview and arXiv public repositories.", 282 "source": "haiku" 283 }, 284 "data_pipeline_documented": { 285 "applies": true, 286 "answer": true, 287 "justification": "The full pipeline from raw paper collection through three synthesis stages (novelty verification, multi-dimension review, reliability verification) to quality control filtering is documented in Section 4.2.", 288 "source": "haiku" 289 } 290 }, 291 "contamination": { 292 "training_cutoff_stated": { 293 "applies": true, 294 "answer": false, 295 "justification": "The Phi-4 base model's training data cutoff is not stated; the paper does not disclose when Phi-4's pre-training data ends relative to ICLR 2024/2025 paper submission dates.", 296 "source": "haiku" 297 }, 298 "train_test_overlap_discussed": { 299 "applies": true, 300 "answer": false, 301 "justification": "The paper does not discuss whether ICLR 2024/2025 papers in the test set may have been included in Phi-4's pre-training corpus; only the case study paper is explicitly noted as not in training data.", 302 "source": "haiku" 303 }, 304 "benchmark_contamination_addressed": { 305 "applies": true, 306 "answer": false, 307 "justification": "DeepReview-Bench uses ICLR 2024/2025 papers that were publicly available before Phi-4's training, but potential contamination of the base model's knowledge of these specific papers is not discussed.", 308 "source": "haiku" 309 } 310 }, 311 "human_studies": { 312 "pre_registered": { 313 "applies": false, 314 "answer": false, 315 "justification": "No human subjects study.", 316 "source": "haiku" 317 }, 318 "irb_or_ethics_approval": { 319 "applies": false, 320 "answer": false, 321 "justification": "No human subjects study; ethics section discusses deployment implications, not IRB.", 322 "source": "haiku" 323 }, 324 "demographics_reported": { 325 "applies": false, 326 "answer": false, 327 "justification": "No human participants.", 328 "source": "haiku" 329 }, 330 "inclusion_exclusion_criteria": { 331 "applies": false, 332 "answer": false, 333 "justification": "No human participants.", 334 "source": "haiku" 335 }, 336 "randomization_described": { 337 "applies": false, 338 "answer": false, 339 "justification": "No human participants.", 340 "source": "haiku" 341 }, 342 "blinding_described": { 343 "applies": false, 344 "answer": false, 345 "justification": "No human participants.", 346 "source": "haiku" 347 }, 348 "attrition_reported": { 349 "applies": false, 350 "answer": false, 351 "justification": "No human participants.", 352 "source": "haiku" 353 } 354 }, 355 "cost_and_practicality": { 356 "inference_cost_reported": { 357 "applies": true, 358 "answer": false, 359 "justification": "Output token counts by mode (3K/8K/14.5K) are mentioned, but no inference latency in seconds, API cost, or GPU utilization during inference is reported.", 360 "source": "haiku" 361 }, 362 "compute_budget_stated": { 363 "applies": true, 364 "answer": false, 365 "justification": "Hardware is specified (8x H100 80G, DeepSpeed+ZeRO3) but total GPU-hours or dollar cost for training 23,500 steps is not reported.", 366 "source": "haiku" 367 } 368 } 369 } 370 }, 371 "claims": [ 372 { 373 "claim": "DeepReviewer-14B reduces Rating MSE by 44.80% compared to CycleReviewer-70B despite having fewer parameters", 374 "evidence": "Table 2 shows DeepReviewer-14B MSE of 1.3137/1.3410 vs CycleReviewer-70B MSE of 2.4870/2.4294 on ICLR 2024/2025", 375 "supported": "strong" 376 }, 377 { 378 "claim": "DeepReviewer achieves win rates of 88.21% and 80.20% against GPT-o1 and DeepSeek-R1 in LLM-as-judge evaluation", 379 "evidence": "Table 4 shows these win rates for 'overall judgment' on ICLR 2024; ICLR 2025 shows 91.67% and 87.39% respectively", 380 "supported": "strong" 381 }, 382 { 383 "claim": "DeepReviewer demonstrates superior robustness to adversarial attacks due to its multi-stage reasoning framework", 384 "evidence": "Figure 2 shows DeepReviewer's score increases only 0.31 points under attack vs 4.26 for Gemini; causal attribution to multi-stage design is asserted without ablation of robustness specifically", 385 "supported": "moderate" 386 }, 387 { 388 "claim": "Test-time scaling via reasoning depth and reviewer count consistently improves performance", 389 "evidence": "Figure 3 shows positive regression trends for both Fast→Best mode and R=1→R=6 reviewer scaling across most metrics, with noted variability at R≠4", 390 "supported": "moderate" 391 }, 392 { 393 "claim": "DeepReviewer reduces Rating MSE by an average of 65.83% and improves Decision Accuracy by 15.2 points compared to prompt-based baselines", 394 "evidence": "Reported in Section 5.2 body text with reference to Table 2; calculation is across multiple backbone models of AI Scientist/AgentReview", 395 "supported": "strong" 396 }, 397 { 398 "claim": "DeepReview-13K with structured reasoning annotations enables training a model that outperforms much larger fine-tuned competitors", 399 "evidence": "14B DeepReviewer outperforms 70B CycleReviewer across all metrics in Table 2; however, the advantage could partly reflect distributional fit since both use ICLR data", 400 "supported": "moderate" 401 }, 402 { 403 "claim": "Gemini-2.0-Flash-Thinking as judge validates DeepReviewer's superiority even when Gemini itself is a baseline being compared", 404 "evidence": "Table 4 shows 59.41% win rate for DeepReviewer even against Gemini, which Gemini judges; the self-evaluation conflict is noted but not corrected for", 405 "supported": "weak" 406 } 407 ], 408 "methodology_tags": [ 409 "benchmark-eval", 410 "case-study" 411 ], 412 "key_findings": "DeepReviewer-14B, trained on the synthetic DeepReview-13K dataset via a three-stage structured reasoning pipeline (novelty verification, multi-dimension review, reliability verification), substantially outperforms both larger fine-tuned models (CycleReviewer-70B) and frontier LLMs (GPT-o1, DeepSeek-R1) on rating prediction MSE, ranking, and paper selection tasks derived from ICLR 2024/2025 reviews. In LLM-as-judge qualitative evaluation, DeepReviewer achieves >88% win rates against GPT-o1 across five review quality dimensions. Test-time scaling experiments confirm that deeper reasoning paths and more simulated reviewers generally improve scoring accuracy, with the fastest mode (3K tokens) already outperforming prior 6K-token baselines.", 413 "red_flags": [ 414 { 415 "flag": "Judge-baseline conflict", 416 "detail": "Gemini-2.0-Flash-Thinking serves simultaneously as the LLM judge evaluating qualitative review quality AND as one of the baselines being evaluated, creating a potential self-preference artifact that is not statistically controlled for." 417 }, 418 { 419 "flag": "Training-test distributional overlap", 420 "detail": "Both training and test data come from ICLR 2024/2025 reviews; general-purpose LLM baselines have no such distributional advantage, making the comparison potentially unfair without explicit domain-adaptation controls." 421 }, 422 { 423 "flag": "No statistical testing", 424 "detail": "All comparative claims across large tables of metrics lack confidence intervals, significance tests, or variance estimates, making it impossible to assess whether improvements are reliable." 425 }, 426 { 427 "flag": "Base model contamination unaddressed", 428 "detail": "Phi-4's training data cutoff is not disclosed; ICLR 2024/2025 papers in the test set may have been seen during Phi-4 pre-training, potentially inflating performance on review content that references specific papers." 429 }, 430 { 431 "flag": "Self-reviewed paper", 432 "detail": "Appendix E states 'This article has been reviewed by DeepReviewer-14B and revised accordingly based on its review comments' — the paper being evaluated was itself revised using the system, raising questions about circularity." 433 }, 434 { 435 "flag": "No human evaluation of review quality", 436 "detail": "All qualitative evaluation relies on LLM-as-judge (Gemini); no human annotators assessed whether DeepReviewer's reviews are actually more useful, accurate, or actionable than baselines." 437 } 438 ], 439 "cited_papers": [ 440 { 441 "title": "CycleResearcher: Improving Automated Research via Automated Review", 442 "relevance": "Direct predecessor; provides CycleReviewer baseline and the CycleResearcher framework that DeepReview extends" 443 }, 444 { 445 "title": "The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery", 446 "relevance": "Primary prompt-based baseline for agent-driven paper review; represents the competing approach" 447 }, 448 { 449 "title": "AgentReview: Exploring Peer Review Dynamics with LLM Agents", 450 "relevance": "Second prompt-based baseline; multi-agent simulation of peer review process" 451 }, 452 { 453 "title": "OpenScholar: Synthesizing Scientific Literature with Retrieval-Augmented LMs", 454 "relevance": "Used as the literature retrieval backbone in DeepReview's novelty verification stage" 455 }, 456 { 457 "title": "Peer Review as a Multi-Turn and Long-Context Dialogue with Role-Based Interactions", 458 "relevance": "ReviewMT dataset and approach; prior work on structured LLM-based review generation" 459 }, 460 { 461 "title": "Are We There Yet? Revealing the Risks of Utilizing Large Language Models in Scholarly Peer Review", 462 "relevance": "Adversarial attack methodology and risk assessment for LLM review systems" 463 }, 464 { 465 "title": "Large Language Models for Automated Scholarly Paper Review: A Survey", 466 "relevance": "Comprehensive survey of the LLM paper review space that contextualizes this work" 467 }, 468 { 469 "title": "A Dataset of Peer Reviews (PeerRead): Collection, Insights and NLP Applications", 470 "relevance": "Foundational benchmark for peer review NLP tasks" 471 } 472 ], 473 "engagement_factors": { 474 "practical_relevance": { 475 "score": 3, 476 "justification": "Live demo at ai-researcher.net/deepreviewer, released model weights, and three inference modes make this immediately usable by researchers and conference organizers." 477 }, 478 "surprise_contrarian": { 479 "score": 1, 480 "justification": "A 14B model outperforming 70B is mildly surprising, but the core finding (structured multi-stage reasoning improves review quality) aligns with conventional wisdom." 481 }, 482 "fear_safety": { 483 "score": 1, 484 "justification": "The ethics section raises concerns about bias amplification and reviewer deskilling, but these are framed as responsible-use guidance rather than alarming findings." 485 }, 486 "drama_conflict": { 487 "score": 1, 488 "justification": "The adversarial robustness finding (other models boosted 4+ points under attack) has a provocative angle, but the paper does not emphasize it as a central controversy." 489 }, 490 "demo_ability": { 491 "score": 3, 492 "justification": "Working demo, released model weights, and dataset allow anyone to test the system immediately; the homepage explicitly links to the demo." 493 }, 494 "brand_recognition": { 495 "score": 1, 496 "justification": "Westlake University and UCL are respected institutions but not the dominant brand names (OpenAI, Google, Meta) that drive HN attention." 497 } 498 }, 499 "hn_data": { 500 "threads": [ 501 { 502 "hn_id": "45967079", 503 "title": "Show HN: Browser-based interactive 3D Three-Body problem simulator", 504 "points": 249, 505 "comments": 113, 506 "url": "https://news.ycombinator.com/item?id=45967079" 507 }, 508 { 509 "hn_id": "36349110", 510 "title": "The Distributed Tensor Algebra Compiler (2022)", 511 "points": 40, 512 "comments": 6, 513 "url": "https://news.ycombinator.com/item?id=36349110" 514 }, 515 { 516 "hn_id": "45202421", 517 "title": "UGMM-NN: Univariate Gaussian Mixture Model Neural Network", 518 "points": 31, 519 "comments": 12, 520 "url": "https://news.ycombinator.com/item?id=45202421" 521 }, 522 { 523 "hn_id": "44250248", 524 "title": "Thermal Detection of People with Mobility Restrictions for Barrier Reduction", 525 "points": 4, 526 "comments": 0, 527 "url": "https://news.ycombinator.com/item?id=44250248" 528 }, 529 { 530 "hn_id": "41139898", 531 "title": "End-to-End Amp Modeling: From Data to Controllable Guitar Amplifier Models", 532 "points": 2, 533 "comments": 2, 534 "url": "https://news.ycombinator.com/item?id=41139898" 535 }, 536 { 537 "hn_id": "43772236", 538 "title": "Hands-On: Segmenting Individual Signs from Continuous Sequences", 539 "points": 1, 540 "comments": 0, 541 "url": "https://news.ycombinator.com/item?id=43772236" 542 } 543 ], 544 "top_points": 249, 545 "total_points": 327, 546 "total_comments": 133 547 } 548 }