Cover image

When a Good Probe Points in Different Directions

TL;DR for operators A model can distinguish two behavioral concepts accurately without providing a reproducible internal measurement of either. In From Stories to Features, Scherbakov and colleagues1 address this problem by extracting concept directions from large collections of matched synthetic stories. Their 1,036-feature dictionary supports natural-text retrieval, controlled behavioral steering, and exploratory analysis of agent failures and jailbreak attacks. ...

November 10, 2026 · 9 min · Zelina
Cover image

One Model, Different Decays: Rethinking Spike Inference Across Sensors

TL;DR for operators Scientific instruments can measure the same underlying activity while producing signals with different response times, noise patterns, and temporal distortions. Maintaining a separate predictive model for every sensor is one solution, but it becomes increasingly difficult to manage as measurement technologies diversify. SpikeSSL1 offers an alternative: one neural architecture that adjusts its temporal processing according to the characteristics of each recording. Across five held-out evaluation conditions, it achieves mean spike-rate correlation of 0.654 versus 0.483 for the strongest competing baseline. Biophysical simulation improves its five-seed average to 0.708, although individual conditions do not benefit equally. ...

November 2, 2026 · 8 min · Zelina
Cover image

Small Backbone, Large Memory: What CHARM Actually Improves

TL;DR for operators A small neural backbone is not necessarily a small learned system: auxiliary task memories also consume parameters, and multiple inference attempts consume compute. In Structured Sparse Memory for Recurrent Reasoning1, CoSE cuts task-memory parameters by about 94% while improving controlled ARC accuracy. But CHARM’s strongest results also rely on rule-preserving synthetic examples, carefully configured recurrence, longer training, and inference-time aggregation. For teams evaluating specialized reasoning models, the actionable comparison is task accuracy versus total learned footprint, training-data investment, and inference cost. Private-memory updates are promising for adding related tasks without changing shared weights; the experiments do not demonstrate production savings or reliable generalization to arbitrary new work. ...

October 24, 2026 · 6 min · Zelina
Cover image

When Another Training Round Stops Buying Learning

TL;DR for operators Closed-loop self-training should not be treated as productive merely because the system is still generating examples and updating weights. Wang et al. propose measuring whether each new round contains patterns that were not already predictable from the previous round, then using that signal to decide both which samples deserve gradient weight and whether another round is worth running at all.1 ...

October 14, 2026 · 8 min · Zelina
Cover image

A Skill Is Not an Environment: What Skill2Env Changes About Agent Post-Training

TL;DR for operators Organizations already possess large amounts of procedural knowledge: internal runbooks, tool instructions, operating procedures, and reusable agent skills. The difficult part is turning that knowledge into training tasks with the correct tools, files, constraints, hidden information, and evaluation logic. Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents1 treats that as an environment-construction problem rather than an instruction-generation problem. Its pipeline produced 2,963 executable tasks by organizing recurring agent demands into five capability dimensions and translating them into reusable difficulty patterns. ...

October 9, 2026 · 7 min · Zelina
Cover image

Teach the Primitive, Not the Picture

TL;DR for operators A synthetic-data budget creates a design choice: generate examples that resemble deployment tasks, or construct simpler exercises that isolate the operations a model will repeatedly need. SpatialBlock1 provides evidence for the second strategy. In a matched Qwen2.5-VL-3B experiment, synthetic training on conventional relative-direction and relative-distance questions improves those same in-domain tasks. But the SpatialBlock curriculum performs better on several external evaluations: 49.1 versus 32.8 on MindCube, 28.1 versus 25.5 on MMSI-Bench, and 46.4 versus 43.6 on MMMU. ...

September 29, 2026 · 7 min · Zelina
Cover image

Make the Structure Survive: Forma Turns Synthetic Clinical Cases Into Auditable Outputs

TL;DR for operators If synthetic cases are going to support training or evaluation, surface plausibility is a weak control objective. The harder requirement is preserving the relationships that make each case meaningful. Forma tests that idea by specifying a person-specific psychological structure before generation and asking whether those directional relationships can be recovered afterward. In the full condition, an external probe reaches MCC +0.41 and AUC 0.70 for directed-edge recovery. When the structural formulation is removed, performance falls close to chance: MCC +0.03/AUC 0.52 with demographics and self-report still present, and +0.01/0.50 under zero-shot generation. ...

September 25, 2026 · 7 min · Zelina
Cover image

Train the Graph Before You Query It: SelfGraphRAG Turns Structure Into Supervision

TL;DR for operators An internal document collection can contain the relationships needed to answer difficult questions while still lacking the labeled examples needed to teach a retriever which relationships matter. That usually leaves teams choosing between manual annotation and retrieval based mostly on embedding similarity. SelfGraphRAG1 tests a third option: build a knowledge graph, turn its structure into generated question-answer examples, and train the retriever on those examples. On MultiHop-RAG, the resulting system reports F1 of 24.62, compared with 2.60 for RAG, 0.98 for LightRAG, and 0.01 for GraphRAG. ...

September 19, 2026 · 7 min · Zelina
Cover image

The Right Answer Is Not Enough: Verify the Reasoning Before You Train on It

TL;DR for operators A synthetic reasoning trace can end with the correct answer and still contain intermediate steps you would not want a model to imitate. ORACLE1 addresses that data-quality problem by checking reasoning one step at a time: it uses a symbolic reasoning engine when a step can be formalized, and LLM-based correctness and feasibility judgments when it cannot. ...

September 17, 2026 · 7 min · Zelina
Cover image

One Model, Two Routes: Why Audio-Omni Unifies Audio by Splitting Its Controls

TL;DR for operators A product that must understand an instruction, preserve a voice or reference sound, and stay synchronized with video is handling several different kinds of control. Audio-Omni’s strongest architectural evidence says those controls should not be forced through the same interface. The system combines a frozen multimodal language model with a trainable audio generator. High-level meaning and transcript information are supplied as flexible context; synchronization and acoustic-reference information are attached directly to the evolving audio representation. In the paper’s conditioning ablation, that allocation performs best across text-to-audio, video-to-audio, text-to-speech, and audio editing. ...

September 14, 2026 · 9 min · Zelina