Model capability is the central variable in LLM pre-training, yet is never observed directly: data shapes it prospectively, while evaluation reveals it only retrospectively, compressing samples, prompts, decoding, and scoring rules into one noisy score. Practical optimization runs this backward: a failure is observed first, and the engineer must infer the corpus fix. The two sides speak incompatible vocabularies -- benchmark names and per-sample correctness versus data sources, domains, and quality labels -- so this inference is usually intuition, not method. We close this gap with the \emph{capability slice}: a group of evaluation samples sharing background condition, task type, solving operation, and output constraint -- precise enough to localize a single weakness yet stable enough to survive aggregation, unlike a benchmark name, too coarse, or a single sample, too noisy. Built around this unit, an evaluation taxonomy, a non-instruction data taxonomy, and mapping rules form a closed loop turning a benchmark-level failure into a targeted, testable data intervention. We test this loop on two case studies pulling in opposite directions. First, the loop rules the data out: continued pre-training drives BBH down by $-46.82\%$, but diagnosis traces this to a single masked \texttt{\textless EOS\textgreater} loss rather than weakened reasoning; restoring it recovers BBH to $66.44$, above the original checkpoint, without changing the data. Second, the loop rules the data in: a persistent math-reasoning weakness is decomposed by solving operation into specific failing combinations, and a weakness-targeted sampling procedure built from it lifts AIME2025/AIME2026 Pass@128 from $6.67$/$0.00$ to $26.67$ each. The same unmodified loop reaches opposite, correct verdicts in both cases, showing the evaluation-to-data inference can be routine, auditable, and experimentally validated rather than intuitive.
Off-policy distillation is now central to large language model pre-training, yet how training data, objective parameterization, and model capabilities interact remains poorly characterized. We studies top-$k$-truncated, temperature-scaled off-policy distillation by decomposing this problem into two questions: an \emph{objective-to-capability} analysis of how the training objective shapes token-level supervision and downstream performance, and a \emph{data-to-objective} analysis of how data heterogeneity should inform objective routing. We first show that the language-modeling objective ($L_{\mathrm{LM}}$) and the knowledge-distillation objective ($L_{\mathrm{KD}}$) induce systematically different capability profiles, and trace this divergence to a gradient-level tension between \emph{direct observed-token reinforcement} and \emph{teacher-supported alternative supervision}. To quantify this tension, we introduce diagnostic metrics -- support coverage, observed-token probability mass, and teacher-distribution concentration -- and show via controlled sweeps that the support size $k$ governs a coverage-sharpness trade-off, while distillation temperature controls within-support probability allocation. We then examine adaptive objective routing: a domain-level policy that applies $L_{\mathrm{LM}}$ to math and code and $L_{\mathrm{KD}}$ to general-domain data yields consistent gains over both single-objective baselines, whereas token-level routing based on observed-token probability mass or teacher entropy fails to consistently match the single-objective baseline. These results suggest that effective objective routing depends less on routing granularity than on the quality of the routing signal, reframing continued pre-training via off-policy distillation as a structured, data-conditional supervision-design problem rather than a global hyperparameter choice.
Tool-augmented multimodal agents show strong benchmark gains, often taken as evidence that agents have learned to use tools. We argue that this interpretation can be premature: a tool-call trace alone does not show whether the tool supplied answer-critical information. We study two representative ``thinking with images'' agents, Thyme and DeepEyesV2, across real-world understanding, OCR, chart understanding, and mathematical reasoning. Each agent is compared with its Tool-Free counterpart and with a Pure-Text Reasoner trained from the same source pool without tool-calling trajectories. Tool access yields little consistent aggregate improvement, does not reliably reduce generated-token cost, and leaves only a small tool-only solved set: 93% of DeepEyesV2's tool-solved problems and 96% of Thyme's are also solved by at least one non-tool setting. Mechanism ablations further show that the full tool-use loop does not consistently outperform either the tool-call format or the returned execution result alone. In the settings we study, the analyzed agents appear to learn tool-calling patterns more reliably than tool-contributed capabilities, suggesting that evaluation should distinguish tool availability from whether tools actually expand what agents can solve.
Vincent Henkel, Felix Gehlhoff, David Kube +13cs.AI
Foundation models, particularly large language models, are increasingly integrated into agent architectures for industrial tasks such as decision support, process monitoring, and engineering automation. Yet evidence on their purposes, capabilities, and limitations remains fragmented across domains. This work examines how mature foundation-model-based agent systems are in industrial contexts, how their functional profile differs from conventional agent systems, and which limitations persist. A systematic literature survey following the PRISMA 2020 guideline is presented, screening 2,341 publications and synthesising a corpus of 88 publications through a structured coding scheme. The results show that reported systems are predominantly at prototype and early validation stages (75.0% at TRL 4-6), with deployment-oriented evidence remaining rare (9.1%). Operational goals are most frequently positioned in user assistance, monitoring, and process optimisation, while conventional production-control purposes such as planning and scheduling are less prominent. Compared with an established baseline for industrial agent systems, the capability profile reveals substantial gains in human interaction (+37%) and dealing with uncertainty (+35%), but a pronounced deficit in negotiation (-39%). The most widely reported limitations concern lack of generalization, hallucination and output instability, data scarcity, and inference latency. A working definition of foundation-model-based industrial agents is also proposed, bridging conventional agent theory, automation-engineering standards, and the foundation-model paradigm.