Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

Monthly edition · August 2026 · Issue #2

Auditable long-horizon systems

Verified terminal tasks, evidence-based evaluation, and action traces.

Top 3 papers

Tracked since posting

  1. Stealing Reasoning Traces from Proprietary LLM APIs

    Exploits weaker sibling models as cross-provider decryption oracles, recovering 182 credentials from 315,320 reasoning blobs.

    Posted18d
    Stars—
    Likes116+73
    →
  2. Recursive Synthesis for Long-Horizon Terminal Tasks

    Recursively grows sandbox-verified terminal tasks from seeds, yielding 37,484 tasks at $0.05 each.

    Posted20d
    Stars—
    Likes250+27
    →
  3. Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

    Targets autonomous theory-level mechanistic discovery, grounding experiments in 13k interpretability papers and 32 tools.

    Posted16d
    Stars49+21
    Likes86+10
    →
  4. Previous leaderOrca: The World is in Your Mind

    #1 paper of July, where it is now

    LedJul
    Stars1k+486
    Likes507+172
    →

Top 3 repos

Tracked since posting

  1. ApodexAI/FrontierAgent

    Unlike most agent runtimes, it ships built-in evaluation and shows parallel Agent Team beating ReAct 38.5 vs 34.4 on APEX.

    Posted3d
    Stars1.3k+367
    Forks109+29
    →
  2. MirroS-Lab/HarnessEval-W

    Unlike scalar-only video benchmarks, it uses model-agnostic skill routing with auditable evidence trees.

    Posted10d
    Stars271+60
    Forks10+5
    →
  3. TencentARC/SCoPE

    Adds explicit camera-ray token coordinates to a pretrained video DiT with under 0.1% new parameters.

    Posted12d
    Stars189+69
    Forks13+4
    →
  4. Previous leaderdeepseek-ai/DeepSpec

    #1 repo of July, where it is now

    LedJul
    Stars7.1k+295
    Forks660+31
    →

Top 3 datasets

Tracked since posting

  1. hamzabagirsakci/turkish-court-decisions

    It offers 11 million CC0 full-text Turkish court decisions spanning 1962–2026 across five court systems.

    Posted4d
    Downloads2k+1.6k
    Likes100+39
    →
  2. Anthropic/claude-protein-binder-design

    It links 1,440 de novo binders to raw two-vendor kinetics, structures, and stepwise design provenance.

    Posted10d
    Downloads42.1k+40.4k
    Likes172+124
    →
  3. PatronusAI/figmatrace

    Unlike final-design datasets, it captures 3,469 expert Figma action-frame trajectories over 126 long-horizon tasks.

    Posted6d
    Downloads1.7k+850
    Likes12+2
    →
  4. Previous leadernvidia/Cosmos3-DROID

    #1 dataset of July, where it is now

    LedJul
    Downloads129.9k+45.8k
    Likes45+35
    →

And this week

One useful ML email a week.

The Weekly ML Signal.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.