Large language models (LLMs) are transforming laboratory automation by enabling self-driving laboratories (SDLs) that could accelerate materials research. However, current SDL implementations rely on rigid protocols that fail to capture the adaptability and intuition of expert scientists in dynamic experimental settings. Here, we show that LLM agents can automate atomic force microscopy (AFM) through our Artificially Intelligent Lab Assistant (AILA) framework. Further, we develop AFMBench—a comprehensive evaluation suite challenging LLM agents across the complete scientific workflow from experimental design to results analysis. We find that state-of-the-art LLMs struggle with basic tasks and coordination scenarios. Notably, models excelling at materials science question-answering perform poorly in laboratory settings, showing that domain knowledge does not translate to experimental capabilities. Additionally, we observe that LLM agents can deviate from instructions, a phenomenon referred to as sleepwalking, raising safety alignment concerns for SDL applications. Our ablations reveal that multi-agent frameworks significantly outperform single-agent approaches, though both remain sensitive to minor changes in instruction formatting or prompting. Finally, we evaluate AILA’s effectiveness in increasingly advanced experiments—AFM calibration, feature detection, mechanical property measurement, graphene layer counting, and indenter detection. These findings establish the necessity for benchmarking and robust safety protocols before deploying LLM agents as autonomous laboratory assistants across scientific disciplines.

Three different types of strain and temperature sensors based on negative curvature hollow core fiber (NCHCF) are proposed. Each sensor is produced by splicing a small section of the NCHCF between two sections of single mode fiber. Different types of interferometers are obtained simply by changing the splicing conditions. The first sensor consists on a single Fabry-Perot interferometer (FPI). The remaining two configurations are attained with the same sensing structure, depending on its position in relation to the interrogation setup. Thus, a double FPI or a hybrid sensor, the latter being composed by an FPI and a Michelson interferometer, are formed. The inline sensors are of submillimeter size, thus enabling nearly punctual measurements.

The flow of high-viscosity liquids inside micrometer-size holes can be substantially different from the flow in the bulk, non-confined state of the same liquid. Such non-Newtonian behavior can be employed to generate structural anisotropy in the frozen-in liquid, i.e., in the glassy state. Here, we report on the observation of non-Newtonian flow of an ultralow melting chalcogenide glass inside a silica microcapillary, leading to a strong deviation of the shear viscosity from its value in the bulk material. In particular, we experimentally show that the viscosity is radius-dependent, which is a clear indication that the microscopic rearrangement of the glass network needs to be considered if the lateral confinement falls below a certain limit. The experiments have been conducted using pressure-assisted melt filling, which provides access to the rheological properties of high-viscosity melt flow under previously inaccessible experimental conditions. The resulting flow-induced structural anisotropy can pave the way towards integration of anisotropic glasses inside hybrid photonic waveguides.

All-solid microstructured optical fibers (MOF) allow the realization of very flexible optical waveguide designs. They are prepared by stacking of doped silica rods or canes in complex arrangements. Typical dopants in silica matrices are germanium and phosphorus to increase the refractive index (RI), or boron and fluorine to decrease the RI. However, the direct interface contact of stacking elements often causes interrelated chemical reactions or evaporation during thermal processing. The obtained fiber structures after the final drawing step thus tend to deviate from the targeted structure risking degrading their favored optical functionality. Dopant profiles and design parameters (e.g., the RI homogeneity of the cladding) are controlled by the combination of diffusion and equilibrium conditions of evaporation reactions. We show simulation results of diffusion and thermal dissociation in germanium and fluorine doped silica rod arrangements according to the monitored geometrical disturbances in stretched canes or drawn fibers. The paper indicates geometrical limits of dopant structures in sub-µm-level depending on the dopant concentration and the thermal conditions during the drawing process. The presented results thus enable an optimized planning of the preform parameters avoiding unwanted alterations in dopant concentration profiles or in design parameters encountered during the drawing process.

To prevent the thermally induced spatial beam degradation occurring in high-power fiber lasers and amplifiers, index-depressed core “fully aperiodic large-pitch fibers” (FA-LPFs) have been designed and fabricated. In contrast to previous experimental works performed on FA-LPFs, in which the active core and the surrounding cladding material are quasi-index-matched, the core refractive index is in slight depression compared to the surrounding material ( Δ n ≈ − 3 × 10 − 5 ). Thus, the index-depressed fiber core tends first to behave as an anti-guide, preventing light from being properly guided into it. However, by increasing the absorbed pump power, the thermal load induces a parabolic refractive index change sufficient to compensate for the − 3 × 10 − 5 index depression in the core, enabling a robust single-mode amplification at high average power. As a proof of concept, using a 110 µm depressed core FA-LPF, M 2 values of 1.3 were demonstrated in amplifier configuration from 60 W to a maximal value of 170 W of emitted average power only limited by the available pump power.

A dual-core fiber in which one of the cores is doped with germanium and the other with phosphorus is used as an in-line Mach–Zehnder dispersive interferometer. By ensuring an equal length but with different dispersion dependencies in the interferometer arms (the two cores), high-sensitivity strain and temperature sensing are achieved. Opposite sensitivities for high and low wavelength peaks were also demonstrated when strain and temperature was applied. To our knowledge this is the first time that such behavior is demonstrated using this type of in-line interferometer based on a dual-core fiber. A sensitivity of $(0.102\pm 0.002)\mathrm{nm}/\mathrm{\mu \epsilon }$, between $0and800\mathrm{\mu \epsilon }$ and $(-4.2\pm 0.2)\mathrm{nm}/°\mathrm{C}$ between 47°C and 62°C is demonstrated.