ML Times
Article Summary
Supabase's Model Context Protocol (MCP) can be exploited to leak sensitive SQL data due to its inability to distinguish between user instructions and data, allowing attackers to craft messages that execute unauthorized SQL commands.
SmolLM3 is a competitive 3B multilingual model that excels in long-context reasoning, outperforming Llama-3.2-3B and Qwen2.5-3B while remaining efficient against larger models like Qwen3 and Gemma3, trained on 11 trillion tokens.
Biomni is a versatile biomedical AI agent that leverages large language model reasoning and retrieval-augmented planning to enhance research productivity and hypothesis generation across various biomedical fields.
State Space Models (SSMs) offer a compressed state that enables efficient, online processing, making them suitable for tasks requiring real-time interaction, unlike Transformers which maintain a detailed cache of every token.
OLMo 2 32B is the largest model in the OLMo family, trained on 6 trillion tokens and surpassing GPT-3.5-Turbo and GPT-4o mini on academic benchmarks, showcasing its advanced capabilities.
Monadic DNA's experiment utilized Nillion's multi-party compute (MPC) technology to enable anonymous genetic data analysis, allowing participants to gain insights without exposing their raw genomic data, thus prioritizing privacy and security.
Adopting a human developmental visual diet (DVD) significantly enhances AI vision by simulating the visual development process from infancy to adulthood, leading to improved shape-based recognition and robustness against adversarial conditions.
TaIrTe₄ photodetectors demonstrate exceptional performance in room-temperature THz sensing, achieving a responsivity of ~0.3 A/W and ultrafast response times on the picosecond scale, significantly enhancing THz detection capabilities.
The study presents a human activity classifier trained on over 60,000 participants and 10 million choices, suggesting a significant leap in understanding human cognition through machine learning.
LTLCrit enhances LLM planners by employing a temporal logic-based critic that identifies and mitigates unsafe or inefficient actions, thus improving overall performance in tasks like Minecraft.
Agencies represent a significant evolution in AI, moving from single intelligence Agents to systems that orchestrate multiple specialized intelligences for task execution, enhancing efficiency and effectiveness.
Fine-tuning Nous Hermes 2 Mistral 7B DPO for a multilingual chatbot requires careful consideration of language mixing and script handling to enhance performance in French, English, and a lesser-known language.
The Semantic Resilience Index (SRI) quantifies how well a sentence's meaning is preserved when rewritten using a limited vocabulary, specifically the Longman Defining Vocabulary (LDV) of about 2,000 basic English words.
Differential design techniques can significantly enhance Mamba architecture, addressing the overallocation of attention that leads to noisy representations and hallucinations in language models.
ToolGen enables AI agents to create tools dynamically, eliminating the need for pre-defined configurations and allowing for immediate tool generation based on user context and needs.