Conversational agents are good at recall. But recall only reacts and an agent that answers only what it is asked can never raise what the user never thought to mention. We study the other direction which we call prospective memory. A background agent keeps a small set of pins in the assistant’s context before any question arrives and each pin must be retired by the right authority. A constraint from chat, such as a stolen bike, ends only on the user’s word. One from a city feed, such as a suspended tram line, ends only on the feed’s end message, whatever its forecast said. We release a benchmark of 60 days, ten users and 1,802 graded decisions in a mobility assistant, with automatic answer keys for pinning, forgetting, recommendation and habit discovery. Capture, at least of the templated reveals this benchmark renders, is the easy half. A dense 8B model captures events almost as well as a 397B/17B sparse one and what separates the two is forgetting. The small model hoards and invents. The large model’s one systematic failure is a release cascade, asked to rewrite its whole pin table, it treats any release signal as a reason to shorten it. Asked instead to name its removals as explicit operations, it names only the right ones, every swept pin survives, and required session recall rises from 0.77 to 0.99, from 0.80 to 1.00 on the events both runs captured.

Long-term conversational AI requires two distinct capabilities: remembering what was discussed and adapting to how a user’s preferences evolve over time. We argue these operate on different timescales and require separate but interoperable architectures. AgenticLOCI is a five-agent framework built on this premise: session retrieval fuses a short-term semantic ranker with a long-term keyword scorer via rank fusion, while personalization is maintained through a versioned deliberation-state memory that accumulates decision-episode snapshots across interactions. We evaluate on three benchmarks using Qwen 2.5-7B-Instruct. On LoCoMo, the AgenticLOCI Backbone achieves F1 = 0.342 versus BM25 = 0.285 while consuming 7.3× fewer input tokens per item. On LongMemEval-S, item-level analysis confirms that the two retrieval layers solve architecturally distinct sub-problems; at 7B scale the keyword layer leads, a reversal expected to resolve at larger model sizes where LLM ranking is more reliable. On PersonaMem, all configurations perform within a narrow band, establishing that memory benefit is conditional on context-budget pressure. The longitudinal personalization layer is evaluated on a synthetic mobility benchmark with known user trajectories, keeping 97.6% of stable preference fields unchanged and making warranted updates with 58.3% precision against an ablation baseline. We demonstrate the full framework as a personalized urban mobility assistant for Magdeburg.

It remains unclear whether agentic augmentation (iterative tool use, code execution, or retrieved memory) improves over single-pass VLM inference for medical image interpretation. We present a systematic quantitative evaluation of the Recursive Language Model (RLM) framework on a multimodal medical task, comparing three pipeline architectures (P1 direct vision, P2 iterative RLM tool use, P3 model-written code agent) across four retrieval-augmented memory conditions on 100 CheXpert Plus chest X-rays, scoring with CheXbert (14-label classification) and GREEN (radiology-specific LLM judge). Three findings emerge: (1) direct vision outperforms both agentic variants; the RLM loop consumes 7-9× more tokens without adding task-relevant information when its only tool calls the same VLM; (2) retrieval consistently helps, but the optimal pool size depends on metric and pipeline: top-5 is the sweet spot for direct-vision report quality (GREEN +37%), while 24 retrieved cases further benefit pathology classification (CheXbert +50%) and can improve GREEN for agentic pipelines; and (3) qualitative analysis shows code agents with a pretrained classifier succeed when it returns confident pathology scores but fail on normal cases where it returns noise-floor outputs, illustrating that tool quality, not framework complexity, determines outcome.

This study investigates efficient speech-to-intent recognition for human–robot interaction in elderly-care environments in German, targeting deployment on resource-constrained platforms such as the Jetson AGX Orin. To benchmark performance, we created a domain-specific German dataset with two sub-datasets (PaSID and PaSynTex) that simulate specific nursing home communication scenarios. Two alternative speech-to-intent pipelines were developed and evaluated: a two-stage system combining automatic speech recognition (ASR) with a large language model (LLM), and an end-to-end large audio–language model (LALM) architecture. The performance of Whisper-based ASR systems was evaluated across a wide variety of LLMs and several LALMs, comparing intent-classification accuracy, latency, and resource efficiency. The results indicate that optimized ASR + LLM configurations, particularly Whisper Turbo coupled with Phi-3.5-mini or Qwen 2.5-7B, outperform unified LALM approaches while maintaining substantially lower memory and inference costs. Also, the analysis shows that, the unified LALM models outperform the two-step integration of ASR + LLM in the same configuration, but at the cost of higher resource utilization, likely due to limited optimization for edge deployment. Overall, the findings provide initial evidence that modular ASR + LLM pipelines provide a more practical solution for real-time, on-device intent recognition in assistive robotics in German, offering an effective trade-off between performance and deployability on resource-constrained platforms.

In recent years, the use of voice assistants has rapidly grown. Hereby, above all, the user’s speech data is stored and processed on a cloud platform, being the decisive factor for a good performance in speech processing and understanding. Although usually, they can be found in private households, a lot of business cases are also employed using voice assistants for public places, be it as an information service, a tour guide, or a booking system. As long as the systems are used in private spaces, it could be argued that the usage is voluntary and that the user itself is responsible for what is processed by the voice assistant system. When leaving the private space, the voluntary use is not the case anymore, as users may be made aware that their voice is processed in the cloud and background voices can be unintendedly recorded and processed as well. Thus, the usage of voice assistants in public environments raises a lot of privacy concerns. In this contribution, we discuss possible anonymization solutions to hide the speakers’ identity, thus allowing a safe cloud processing of speech data. Thereby, we promote the public use of voice assistants.

The LAST MINUTE corpus comprises multimodal recordings (e.g. video, audio, transcripts) from WOZ interactions in a mundane planning task (Rösner et al., 2011). It is one of the largest corpora with naturalistic data currently available. In this paper we report about first results from attempts to automatically and manually analyze the different modes with respect to emotions and affects exhibited by the subjects. We describe and discuss difficulties encountered due to the strong contrast between the naturalistic recordings and traditional databases with acted emotions.