scan-v5.json (28171B)
1 { 2 "scan_version": 5, 3 "paper_type": "empirical", 4 "paper": { 5 "title": "Exploring Data-Efficient Adaptation of Large Language Models for Code Generation", 6 "authors": [ 7 "Xue Jiang", 8 "Yihong Dong", 9 "Zhiyuan Fan", 10 "Zhi Jin", 11 "Wenpin Jiao", 12 "Ge Li" 13 ], 14 "year": 2024, 15 "venue": "ACM Transactions on Software Engineering and Methodology", 16 "arxiv_id": "2403.00046", 17 "doi": "10.1145/3772721" 18 }, 19 "checklist": { 20 "claims_and_evidence": { 21 "abstract_claims_supported": { 22 "applies": true, 23 "answer": true, 24 "justification": "The abstract's '46.2% average relative improvement in Pass@1' matches the average of the five per-dataset improvements reported in Table 1 (29.5%, 33.0%, 27.1%, 37.6%, 103.8%); all major claims are supported by experimental results.", 25 "source": "haiku" 26 }, 27 "causal_claims_justified": { 28 "applies": true, 29 "answer": true, 30 "justification": "The paper makes causal claims about error-driven learning improving efficiency; these are backed by controlled ablation studies (RQ3 training data variants, RQ6 component ablations) that isolate the effect.", 31 "source": "haiku" 32 }, 33 "generalization_bounded": { 34 "applies": true, 35 "answer": false, 36 "justification": "The paper claims 'broad applicability' but experiments are exclusively on Python benchmarks with 2B–7B models; no explicit scope boundary for programming language or model scale is stated despite the sweeping framing.", 37 "source": "haiku" 38 }, 39 "alternative_explanations_discussed": { 40 "applies": true, 41 "answer": false, 42 "justification": "The gains are attributed solely to error-driven learning without considering alternative explanations such as quality-filtering effects (only passing revisions are kept), curriculum learning dynamics, or data augmentation from the iterative process.", 43 "source": "haiku" 44 }, 45 "proxy_outcome_distinction": { 46 "applies": true, 47 "answer": true, 48 "justification": "Pass@k is measured via automated test case execution, directly evaluating functional correctness; no conflation between proxy metrics (BLEU, token overlap) and actual correctness is made.", 49 "source": "haiku" 50 } 51 }, 52 "limitations_and_scope": { 53 "limitations_section_present": { 54 "applies": true, 55 "answer": true, 56 "justification": "Section 8 'Limitations' is a dedicated section listing two specific limitations: requirement for test cases during preprocessing and restriction to low-resource scenarios.", 57 "source": "haiku" 58 }, 59 "threats_to_validity_specific": { 60 "applies": true, 61 "answer": true, 62 "justification": "Section 7 discusses three threats with some specificity: dataset quality and generalizability, hyperparameter sensitivity with acknowledgment of 'small-range grid search,' and metric reliability justifying the unbiased Pass@k estimator.", 63 "source": "haiku" 64 }, 65 "scope_boundaries_stated": { 66 "applies": true, 67 "answer": false, 68 "justification": "While limitations note the test case requirement and low-resource focus, the paper does not explicitly state what the results do NOT show — no mention of language scope, model scale limits, or inapplicability to other task types.", 69 "source": "haiku" 70 } 71 }, 72 "conflicts_of_interest": { 73 "funding_disclosed": { 74 "applies": true, 75 "answer": true, 76 "justification": "Acknowledgments disclose funding from National Key R&D Program No. 2023YFB4503801, National Natural Science Foundation of China Nos. 62192733/62192730/62192731, and Hubei Province Major Program No. 2023BAA024.", 77 "source": "haiku" 78 }, 79 "affiliations_disclosed": { 80 "applies": true, 81 "answer": true, 82 "justification": "All six authors are disclosed as affiliated with the Key Laboratory of High Confidence Software Technologies, School of Computer Science, Peking University.", 83 "source": "haiku" 84 }, 85 "funder_independent_of_outcome": { 86 "applies": true, 87 "answer": true, 88 "justification": "Funders are Chinese government research programs (NSFC, National Key R&D) with no direct financial stake in the DEED method's adoption or commercialization.", 89 "source": "haiku" 90 }, 91 "financial_interests_declared": { 92 "applies": true, 93 "answer": false, 94 "justification": "No competing interests statement or financial interests declaration is included anywhere in the paper.", 95 "source": "haiku" 96 } 97 }, 98 "scope_and_framing": { 99 "key_terms_defined": { 100 "applies": true, 101 "answer": true, 102 "justification": "'Data-efficient adaptation' is contextualized as adapting with limited training data, 'error-driven learning' is explained via the four-step process, and DEED's acronym is explicitly defined.", 103 "source": "haiku" 104 }, 105 "intended_contribution_clear": { 106 "applies": true, 107 "answer": true, 108 "justification": "Three contributions are explicitly enumerated: demonstrating error-driven learning effectiveness, proposing DEED, and showing outperformance over mainstream approaches on five benchmarks.", 109 "source": "haiku" 110 }, 111 "engagement_with_prior_work": { 112 "applies": true, 113 "answer": true, 114 "justification": "Section 6 engages substantively with fine-tuning variants, prompting approaches, and related code refinement methods (Self-Refine, Self-Debug, Self-Edit, CYCLE, ILF), distinguishing DEED's contribution from each.", 115 "source": "haiku" 116 } 117 } 118 }, 119 "type_checklist": { 120 "empirical": { 121 "artifacts": { 122 "code_released": { 123 "applies": true, 124 "answer": false, 125 "justification": "No code repository URL or availability statement is provided anywhere in the paper.", 126 "source": "haiku" 127 }, 128 "data_released": { 129 "applies": true, 130 "answer": true, 131 "justification": "All five evaluation datasets (HumanEval, MBPP, HumanEval-ET, MBPP-ET, DS-1000/DataScience) are standard public benchmarks available independently of this work.", 132 "source": "haiku" 133 }, 134 "environment_specified": { 135 "applies": true, 136 "answer": false, 137 "justification": "Only 'a single A6000 GPU' is mentioned; no requirements.txt, Dockerfile, Python version, or library versions are specified.", 138 "source": "haiku" 139 }, 140 "reproduction_instructions": { 141 "applies": true, 142 "answer": false, 143 "justification": "Algorithm 1 provides pseudocode and Section 4.2 lists hyperparameters, but no step-by-step reproduction instructions sufficient to rerun experiments without guessing implementation details are provided.", 144 "source": "haiku" 145 } 146 }, 147 "statistical_methodology": { 148 "confidence_intervals_or_error_bars": { 149 "applies": true, 150 "answer": false, 151 "justification": "Results are averaged over five runs but no confidence intervals, standard deviations, or error bars are reported for any metric.", 152 "source": "haiku" 153 }, 154 "significance_tests": { 155 "applies": true, 156 "answer": false, 157 "justification": "No statistical significance tests are performed despite making multiple comparative claims across methods, datasets, and LLMs.", 158 "source": "haiku" 159 }, 160 "effect_sizes_reported": { 161 "applies": true, 162 "answer": true, 163 "justification": "Relative improvement percentages (e.g., ↑29.5%, ↑103.8%) are reported against the best-performing baseline, providing interpretable effect magnitudes.", 164 "source": "haiku" 165 }, 166 "sample_size_justified": { 167 "applies": true, 168 "answer": false, 169 "justification": "The training split (min(200, 40%*D)) is stated but not justified; no power analysis or reasoning about whether sample sizes are sufficient to detect the reported effects is provided.", 170 "source": "haiku" 171 }, 172 "variance_reported": { 173 "applies": true, 174 "answer": false, 175 "justification": "Results are 'averaged over five test runs' but no standard deviation, variance, or range across those runs is reported anywhere.", 176 "source": "haiku" 177 } 178 }, 179 "evaluation_design": { 180 "baselines_included": { 181 "applies": true, 182 "answer": true, 183 "justification": "Six baselines are included: Direct Generation, Fine-tuning (Full), Fine-tuning (LoRA), Few-shot Prompting, Self-Refine, and Self-Debug.", 184 "source": "haiku" 185 }, 186 "baselines_contemporary": { 187 "applies": true, 188 "answer": true, 189 "justification": "Baselines include Self-Refine (NeurIPS 2023), Self-Debug (2023), and LoRA (ICLR 2022); all are relevant and contemporary for the submission period.", 190 "source": "haiku" 191 }, 192 "ablation_study": { 193 "applies": true, 194 "answer": true, 195 "justification": "RQ3 ablates training data variants, RQ4 studies iteration counts, RQ5 varies the revision model, and RQ6 ablates Self-Revise input components (correct solution, error messages, failed test cases).", 196 "source": "haiku" 197 }, 198 "multiple_metrics": { 199 "applies": true, 200 "answer": true, 201 "justification": "Pass@1, Pass@5, and Pass@10 are reported throughout; Pass@any is added for revision quality experiments.", 202 "source": "haiku" 203 }, 204 "human_evaluation": { 205 "applies": false, 206 "answer": false, 207 "justification": "Code correctness is evaluated via automated test execution; human evaluation of generated code quality is not applicable to this setting.", 208 "source": "haiku" 209 }, 210 "held_out_test_set": { 211 "applies": true, 212 "answer": true, 213 "justification": "Each dataset is split into training (min(200, 40%*D) problems) and a held-out test set (remaining problems); evaluation is performed on the test portion only.", 214 "source": "haiku" 215 }, 216 "per_category_breakdown": { 217 "applies": true, 218 "answer": false, 219 "justification": "Results are reported per dataset but no per-category, per-difficulty, or per-problem-type breakdowns are provided within datasets.", 220 "source": "haiku" 221 }, 222 "failure_cases_discussed": { 223 "applies": true, 224 "answer": false, 225 "justification": "Figure 4 provides qualitative success cases for Self-Revise; no systematic discussion of where DEED fails or under what conditions it underperforms is included.", 226 "source": "haiku" 227 }, 228 "negative_results_reported": { 229 "applies": true, 230 "answer": true, 231 "justification": "The paper reports that ChatGPT/GPT-3.5-turbo as MRevise does not outperform Self-Revise (FT), that Fine-tuning (LoRA) underperforms Full fine-tuning, and that Llama-7B Fine-tuning underperforms Direct Generation — all reported without concealment.", 232 "source": "haiku" 233 } 234 }, 235 "setup_transparency": { 236 "model_versions_specified": { 237 "applies": true, 238 "answer": false, 239 "justification": "Models are named (CodeGen-2B, Llama-7B, CodeLlama-7B) but no checkpoint hashes, Hugging Face identifiers, or version dates are given; ChatGPT is cited without any version.", 240 "source": "haiku" 241 }, 242 "prompts_provided": { 243 "applies": true, 244 "answer": true, 245 "justification": "Appendix C provides the exact instruction text for automatic code revision, and Figure 3 shows the full template structure with all five input components labeled.", 246 "source": "haiku" 247 }, 248 "hyperparameters_reported": { 249 "applies": true, 250 "answer": true, 251 "justification": "Section 4.2 reports learning rates (5e-6 Full, 2e-4 LoRA), batch size (1), gradient accumulation (32), training epochs (10), temperature (0.8), LoRA rank (128), α (8), β1/β2 (0.9), and sampling counts (5 for errors, 30 for revisions).", 252 "source": "haiku" 253 }, 254 "scaffolding_described": { 255 "applies": false, 256 "answer": false, 257 "justification": "DEED is a fine-tuning pipeline, not an agentic scaffold; the iterative training process is fully described in Algorithm 1 but does not constitute agentic scaffolding.", 258 "source": "haiku" 259 }, 260 "data_preprocessing_documented": { 261 "applies": true, 262 "answer": true, 263 "justification": "Sections 3.1 and 3.2 document error code collection via rejection sampling and revision via acceptance sampling with test execution filtering in sufficient detail.", 264 "source": "haiku" 265 } 266 }, 267 "data_integrity": { 268 "raw_data_available": { 269 "applies": true, 270 "answer": false, 271 "justification": "The generated error codes and revised training data produced during DEED's preprocessing are not released; only the public benchmark sources are available.", 272 "source": "haiku" 273 }, 274 "data_collection_described": { 275 "applies": true, 276 "answer": true, 277 "justification": "Sections 3.1 and 3.2 describe error code collection (rejection sampling by log-probability) and revision (acceptance sampling with minimum Levenshtein distance selection) in detail.", 278 "source": "haiku" 279 }, 280 "recruitment_methods_described": { 281 "applies": false, 282 "answer": false, 283 "justification": "Standard public benchmarks are used; no participant recruitment is involved.", 284 "source": "haiku" 285 }, 286 "data_pipeline_documented": { 287 "applies": true, 288 "answer": true, 289 "justification": "Algorithm 1 documents the complete iterative pipeline from dataset input through error collection, revision, model optimization, and iteration termination.", 290 "source": "haiku" 291 } 292 }, 293 "contamination": { 294 "training_cutoff_stated": { 295 "applies": true, 296 "answer": false, 297 "justification": "Training data cutoffs for CodeGen, Llama-7B, or CodeLlama-7B are not stated, despite all being trained potentially after HumanEval and MBPP were publicly released.", 298 "source": "haiku" 299 }, 300 "train_test_overlap_discussed": { 301 "applies": true, 302 "answer": false, 303 "justification": "Contamination is not discussed for the main benchmarks; EvoCodeBench is used in Appendix B as a supplementary contamination-resistant evaluation but the main evaluation does not address overlap.", 304 "source": "haiku" 305 }, 306 "benchmark_contamination_addressed": { 307 "applies": true, 308 "answer": false, 309 "justification": "HumanEval (2021) and MBPP (2021) were publicly available before training cutoffs of most evaluated models; this is not discussed despite being a known confound for code LLM evaluation.", 310 "source": "haiku" 311 } 312 }, 313 "human_studies": { 314 "pre_registered": { 315 "applies": false, 316 "answer": false, 317 "justification": "No human participants.", 318 "source": "haiku" 319 }, 320 "irb_or_ethics_approval": { 321 "applies": false, 322 "answer": false, 323 "justification": "No human participants.", 324 "source": "haiku" 325 }, 326 "demographics_reported": { 327 "applies": false, 328 "answer": false, 329 "justification": "No human participants.", 330 "source": "haiku" 331 }, 332 "inclusion_exclusion_criteria": { 333 "applies": false, 334 "answer": false, 335 "justification": "No human participants.", 336 "source": "haiku" 337 }, 338 "randomization_described": { 339 "applies": false, 340 "answer": false, 341 "justification": "No human participants.", 342 "source": "haiku" 343 }, 344 "blinding_described": { 345 "applies": false, 346 "answer": false, 347 "justification": "No human participants.", 348 "source": "haiku" 349 }, 350 "attrition_reported": { 351 "applies": false, 352 "answer": false, 353 "justification": "No human participants.", 354 "source": "haiku" 355 } 356 }, 357 "cost_and_practicality": { 358 "inference_cost_reported": { 359 "applies": true, 360 "answer": false, 361 "justification": "The paper qualitatively states DEED incurs no additional inference overhead compared to direct generation, but provides no quantitative latency or cost measurements.", 362 "source": "haiku" 363 }, 364 "compute_budget_stated": { 365 "applies": true, 366 "answer": false, 367 "justification": "Only 'a single A6000 GPU' is mentioned; total training time, GPU-hours, or compute budget for the full experimental suite is not reported.", 368 "source": "haiku" 369 } 370 } 371 } 372 }, 373 "claims": [ 374 { 375 "claim": "DEED achieves an average relative improvement of 46.2% in Pass@1 over the best-performing mainstream baseline across five code generation benchmarks under limited data.", 376 "evidence": "Table 1 reports relative improvements over Fine-tuning (Full) of 29.5% (HumanEval), 33.0% (HumanEval-ET), 27.1% (MBPP), 37.6% (MBPP-ET), and 103.8% (DataScience), averaging to 46.2%.", 377 "supported": "strong" 378 }, 379 { 380 "claim": "Training on revised error codes (error-driven learning) is more data-efficient than training on original dataset samples.", 381 "evidence": "Table 3 shows DEED (32.8% Pass@1) outperforms Raw D_train fine-tuning (25.8%) using fewer training examples; representational distance analysis shows revised codes are closer to error codes than dataset samples (6.39 vs 12.35 Euclidean distance).", 382 "supported": "strong" 383 }, 384 { 385 "claim": "Self-Revise using the same base model (fine-tuning setting) yields better final model performance than using larger or more capable external models for revision.", 386 "evidence": "Table 5 shows Self-Revise (FT) with CodeGen-2B achieves M_θ* Pass@1 of 32.8% vs 27.0% for ChatGPT-based revision, despite ChatGPT having far higher MRevise Pass@any (92.1% vs 24.6%).", 387 "supported": "moderate" 388 }, 389 { 390 "claim": "DEED's iterative adaptation stabilizes after two iterations, capturing most achievable gains.", 391 "evidence": "Table 4 shows Pass@1 of 31.6% (iter 1), 32.8% (iter 2), 33.0% (iter 3), 33.2% (iter 4), with diminishing returns and Pass@10 oscillation after iteration 2.", 392 "supported": "moderate" 393 }, 394 { 395 "claim": "DEED is broadly applicable across LLMs of varying sizes and architectures.", 396 "evidence": "Table 2 shows 25–33% relative improvements over Fine-tuning across CodeGen-2B, CodeGen-6B, Llama-7B, and CodeLlama-7B, though all are 2B–7B models tested on Python-only tasks.", 397 "supported": "weak" 398 } 399 ], 400 "methodology_tags": [ 401 "benchmark-eval" 402 ], 403 "key_findings": "DEED, which fine-tunes LLMs on automatically self-revised versions of their own error outputs rather than raw dataset samples, achieves 27–104% relative improvement in Pass@1 over mainstream adaptation approaches on five Python code generation benchmarks under limited data conditions. The core empirical finding is that error-driven training data is more data-efficient than standard dataset samples, supported by representational distance analysis and ablation experiments. Self-Revise performs best using the same model being adapted in a fine-tuning setting, and performance gains stabilize after two iterations.", 404 "red_flags": [ 405 { 406 "flag": "No variance reported", 407 "detail": "Results are averaged over five runs but standard deviations are never reported, making it impossible to assess whether observed differences between methods exceed run-to-run variability." 408 }, 409 { 410 "flag": "No statistical significance tests", 411 "detail": "Multiple comparative claims across six baselines, five datasets, and four LLMs are made with no statistical tests applied; numerical differences may not be meaningful." 412 }, 413 { 414 "flag": "Benchmark contamination unaddressed", 415 "detail": "HumanEval and MBPP (both 2021) were publicly available before training cutoffs of CodeGen, Llama, and CodeLlama; this known confound is not discussed for the main evaluation." 416 }, 417 { 418 "flag": "Code not released", 419 "detail": "No implementation code is provided despite the method having non-trivial implementation complexity (rejection/acceptance sampling, iterative training loop, revision filtering)." 420 }, 421 { 422 "flag": "Generalizability overclaimed", 423 "detail": "Claims of 'broad applicability' are based solely on Python benchmark tasks and models ≤7B; no non-Python languages, larger models, or non-programming tasks are tested." 424 }, 425 { 426 "flag": "Model versions unspecified", 427 "detail": "Model names are given without checkpoint hashes, Hugging Face identifiers, or snapshot dates, preventing exact reproduction and confounding cross-study comparisons." 428 } 429 ], 430 "cited_papers": [ 431 { 432 "title": "Evaluating Large Language Models Trained on Code (Codex/HumanEval)", 433 "relevance": "Foundational code LLM and benchmark (HumanEval); primary evaluation dataset and baseline comparison point." 434 }, 435 { 436 "title": "Self-Refine: Iterative Refinement with Self-Feedback", 437 "relevance": "Direct competing baseline for iterative code improvement via prompting; DEED is evaluated against it." 438 }, 439 { 440 "title": "Teaching Large Language Models to Self-Debug", 441 "relevance": "Direct competing baseline using execution feedback for code correction; compared against in main evaluation." 442 }, 443 { 444 "title": "LoRA: Low-Rank Adaptation of Large Language Models", 445 "relevance": "Parameter-efficient fine-tuning method used as a baseline and within DEED for resource-constrained settings." 446 }, 447 { 448 "title": "CYCLE: Learning to Self-Refine the Code Generation", 449 "relevance": "Concurrent work on test-driven self-refinement for code; explicitly contrasted with DEED's adaptation focus." 450 }, 451 { 452 "title": "Generalization or Memorization: Data Contamination and Trustworthy Evaluation for Large Language Models", 453 "relevance": "Cited to contextualize data leakage in evaluation benchmarks; motivates supplementary EvoCodeBench experiment." 454 }, 455 { 456 "title": "EvoCodeBench: An Evolving Code Generation Benchmark with Domain-Specific Evaluations", 457 "relevance": "Contamination-resistant benchmark used in Appendix B to validate DEED in a data-leakage-aware setting." 458 }, 459 { 460 "title": "Program Synthesis with Large Language Models (MBPP)", 461 "relevance": "Primary benchmark dataset used for most experiments and preliminary representational distance analysis." 462 }, 463 { 464 "title": "CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis", 465 "relevance": "Primary base model (CodeGen-2B) used in most experiments; default model for full fine-tuning comparisons." 466 }, 467 { 468 "title": "Self-Edit: Fault-Aware Code Editor for Code Generation", 469 "relevance": "Related work training a separate editor model for code revision; contrasted with DEED's self-revision approach." 470 } 471 ], 472 "engagement_factors": { 473 "practical_relevance": { 474 "score": 2, 475 "justification": "Directly addresses the real-world scarcity of domain-specific training data with a deployable fine-tuning pipeline requiring no external resources beyond test cases." 476 }, 477 "surprise_contrarian": { 478 "score": 2, 479 "justification": "Counterintuitive finding that using a weak base model for self-revision outperforms ChatGPT-based revision, and that error-focused training beats learning from correct examples." 480 }, 481 "fear_safety": { 482 "score": 0, 483 "justification": "No safety, risk, or misuse concerns are raised; the paper is entirely focused on improving code generation performance." 484 }, 485 "drama_conflict": { 486 "score": 0, 487 "justification": "No controversy with established methods or high-profile conflict; straightforward incremental improvement paper." 488 }, 489 "demo_ability": { 490 "score": 1, 491 "justification": "Method is conceptually implementable on public benchmarks and models, but no code is released, requiring substantial re-implementation effort before practitioners can try it." 492 }, 493 "brand_recognition": { 494 "score": 0, 495 "justification": "Peking University research group; not a top-tier AI lab brand with mainstream tech community recognition." 496 } 497 }, 498 "hn_data": { 499 "threads": [ 500 { 501 "hn_id": "39651926", 502 "title": "An all-optical general-purpose CPU and optical computer architecture", 503 "points": 197, 504 "comments": 103, 505 "url": "https://news.ycombinator.com/item?id=39651926", 506 "created_at": "2024-03-09T14:49:53Z" 507 }, 508 { 509 "hn_id": "33426789", 510 "title": "Yoneda Hacking: The Algebra of Attacker Actions", 511 "points": 9, 512 "comments": 0, 513 "url": "https://news.ycombinator.com/item?id=33426789", 514 "created_at": "2022-11-01T20:10:20Z" 515 }, 516 { 517 "hn_id": "42496507", 518 "title": "Online Advertising Is a Regrettable Necessity", 519 "points": 6, 520 "comments": 2, 521 "url": "https://news.ycombinator.com/item?id=42496507", 522 "created_at": "2024-12-23T18:27:49Z" 523 }, 524 { 525 "hn_id": "41961564", 526 "title": "Easy real-time collision detection", 527 "points": 4, 528 "comments": 0, 529 "url": "https://news.ycombinator.com/item?id=41961564", 530 "created_at": "2024-10-27T11:06:23Z" 531 }, 532 { 533 "hn_id": "39610408", 534 "title": "Polyamorous Scheduling is NP-hard", 535 "points": 3, 536 "comments": 0, 537 "url": "https://news.ycombinator.com/item?id=39610408", 538 "created_at": "2024-03-05T23:27:01Z" 539 }, 540 { 541 "hn_id": "39329353", 542 "title": "Training microrobots to swim by a large language model", 543 "points": 2, 544 "comments": 1, 545 "url": "https://news.ycombinator.com/item?id=39329353", 546 "created_at": "2024-02-10T19:21:39Z" 547 }, 548 { 549 "hn_id": "41537027", 550 "title": "Towards Battery-Free Wireless Sensing via Radio-Frequency Energy Harvesting", 551 "points": 2, 552 "comments": 0, 553 "url": "https://news.ycombinator.com/item?id=41537027", 554 "created_at": "2024-09-14T02:26:33Z" 555 }, 556 { 557 "hn_id": "39352140", 558 "title": "Detecting Multimedia Generated by Large AI Models: A Survey", 559 "points": 2, 560 "comments": 0, 561 "url": "https://news.ycombinator.com/item?id=39352140", 562 "created_at": "2024-02-12T23:36:45Z" 563 }, 564 { 565 "hn_id": "45763351", 566 "title": "VaultDB: A Real-World Pilot of SMPC Within a Clinical Research Network", 567 "points": 1, 568 "comments": 0, 569 "url": "https://news.ycombinator.com/item?id=45763351", 570 "created_at": "2025-10-30T18:24:42Z" 571 }, 572 { 573 "hn_id": "41981519", 574 "title": "Easy real-time collision detection", 575 "points": 1, 576 "comments": 0, 577 "url": "https://news.ycombinator.com/item?id=41981519", 578 "created_at": "2024-10-29T09:41:11Z" 579 } 580 ], 581 "top_points": 197, 582 "total_points": 227, 583 "total_comments": 106 584 } 585 }