llm-kb
← Back to research
Paper

Mixture of Probes: New method for multimodal LLMs to learn from privileged modalities

Researchers propose Mixture of Probes (MoP), a method enabling multimodal LLMs to leverage auxiliary modalities available only during training. This addresses real-world scenarios where inference-time modalities are limited, improving model performance without requiring all modalities at test time.

0 engagement·1 source·Thu, Jul 9, 2026, 06:00 PM
The paper introduces Mixture of Probes (MoP), a technique for multimodal large language models (MLLMs) to learn from privileged modalities—modalities present during training but absent at inference. Traditional MLLMs assume all training modalities are available at inference, which fails in many practical settings. MoP uses probing to extract complementary supervision from privileged modalities during training, allowing the model to benefit from them even when they are unavailable later. The method treats modalities as sources of complementary supervision rather than interchangeable inputs. This work is relevant for deploying MLLMs in resource-constrained or sensor-limited environments.

Entities

Mixture of Probes(concept)Multimodal Large Language Models(concept)

Related

Paper

TSRouter: Dynamic Modality-Model Selection for Time Series Reasoning

A new framework called TSRouter dynamically selects between LLMs and VLMs for time series reasoning, leveraging their complementary strengths. LLMs preserve exact numerical details but miss global patterns, while VLMs capture patterns but lose fine-grained data. TSRouter chooses the best modality and model per input, balancing accuracy and cost.

0 engagement·1 source·arxiv
Thu, Jul 9, 2026, 09:09 PM
arXiv
Paper

Researchers identify modality interference as root cause of full-duplex SLM degradation

A new paper on arXiv (July 7, 2026) presents a fine-grained analysis of optimization dynamics in full-duplex Spoken Language Models (SLMs), identifying severe modality interference as the root cause of knowledge degradation and compromised semantic integrity. The work aims to enable more natural and intelligent full-duplex SLMs.

0 engagement·1 source·arxiv
Tue, Jul 7, 2026, 05:43 PM
arXiv
Paper

LLM/VLA models enable prompt-driven exploration in RL

A new research paper proposes using large language models (LLMs) and vision-language-action (VLA) models to drive exploration in reinforcement learning by modifying natural language prompts, which induce global behavioral changes beyond standard action noise. This approach could help policies escape weak local optima more effectively.

0 engagement·1 source·arxiv
Thu, Jul 9, 2026, 06:00 PM
arXiv
Product

LLM hardware recipe database with filters and community usage tracking

A community-driven database that lists which LLM models run on which hardware, with performance details. Users can filter by hardware, submit new recipes, and mark which recipes they actively use to show popularity. It solves the problem of finding compatible model-hardware combinations for LLM deployment.

4 engagement·1 source·reddit
Sat, Jul 11, 2026, 10:45 PM
Paper

ICML accepts prompt-engineering paper 'Verbalized Sampling' sparking debate on rigor

A paper titled 'Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity' has been accepted to ICML 2026. The work proposes a simple prompt-engineering trick to improve sampling diversity, but its acceptance at a top-tier ML conference has drawn criticism for lacking rigorous theoretical analysis.

17 engagement·1 source·reddit
Mon, Jul 13, 2026, 05:00 AM