Skip to content
← Best of the week
AI Engineer

AI papers engineers are reading this week

Research papers, benchmarks and evaluation results that engineers who build with AI shared this week.

Updated Nov 12, 2026. Read from 25 sources every four hours.

See the full ranked list Free account. No card.
  1. 1PaperWorth trying this weekFirst seen Nov 12, 2026

    MACaQ: Fast, Model Agnostic Calibrated Quantization for Low-Precision Weights and Activations

    MACaQ is an MXFP4 W4A4 post-training quantization method that quantizes each linear layer independently using online activation transforms and Hessian-aware refinement, reducing cost while preserving accuracy across language and diffusion models. Code:…

  2. 2PaperWorth knowingFirst seen Nov 12, 2026

    Faster Retrieval for Better, More Efficient RAG

  3. 3PaperWorth trying this weekFirst seen Nov 12, 2026

    Don't Just Trust Your Memory: Evidence-Linked Memory Reduces Memory-Induced Degradation in LLM Agents

    WikiTrace links each distilled lesson to its source trajectory and rationale, enabling LLM agents to verify and correct memories before acting, improving performance and reducing memory-induced degradation across agentic benchmarks. Code:…

  4. 4PaperWorth knowingFirst seen Nov 12, 2026

    Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines

  5. 5PaperWorth knowingFirst seen Nov 12, 2026

    Few-Step Streaming Video Generation via Condition Shifting

    APEX extends a streaming video generation method by replacing a separate fake-score network with condition shifting to estimate the generated distribution’s velocity field, improving instruction following and visual aesthetics while reducing memory and…

  6. 6PaperWorth knowingFirst seen Nov 12, 2026

    Physically Controllable Relighting of Photographs

  7. 7PaperWorth knowingFirst seen Nov 12, 2026

    DeepSeek V4.1 Flash 1017 exl3 2.9bpw APEX 2025 benchmark results

    My main AI machine is a 6x3090 128GB DDR4 Threadripper 3960x system. I probably spend more time trying to optimize it than actually using it but it has been fun. I tried a bunch of saturated benchmarks like Terminal Bench 2.1, AIME 2026 and the performance…

  8. 8PaperWorth knowingFirst seen Nov 12, 2026

    One Table, Many Queries: Benchmarking Generalist Tabular Models Across Generation, Imputation, and Prediction

    A large aligned benchmark reveals that tabular generation, imputation, and prediction share useful conditional information but show distinct fidelity rankings and objective interactions, motivating unified models that improve categorical reconstruction and…

  9. 9PaperWorth knowingFirst seen Nov 12, 2026

    SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering

    Low-bit quantized LLM agents can recover much of their task success through closed-loop guidance from a full-precision teacher, followed by unassisted student training on corrected and reference turns. Code: github.com/air-embodied-brain/SteerDistill.

  10. 10PaperWorth knowingFirst seen Nov 12, 2026

    Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation

62 more ranked this week.

AI Engineer in Spezia ranks every find as it lands, today and this week, with filters by kind and by source, and you can ask Spezia whether each one is worth your time.

Open AI Engineer

How it is ranked

  • Each item is judged for an engineer who builds with AI: how much it could change how you work this week, whether it is new, and whether there is real code or results behind it.
  • How hot it is where it was found counts too: points, stars gained, upvotes and likes, each on its own scale.
  • A find picked up in several places at once ranks higher than one seen once.
  • Today fades an item by half its lift every two days. This week ranks on quality alone.
  • Nobody pays to be listed.

More on AI Engineer: what it does and what it reads.

More from AI Engineer

Other desks