Skip to content
Friday, 20 November 2026
Tenesys AI News
Subscribe

All News

TENESYS AI NEWS tracks the most important AI news and explains what changed, why it matters and whether a technology is worth testing.

#LLM — 49 articles ✕

Models· Important· 🧪 Worth Testing

New 1B parameter model focused on mathematical reasoning

Researchers released Limite 1B - Violetto, a compact 1-billion-parameter model built for mathematical problem-solving. It supports a 131k token context window and is licensed under Apache 2.0.

arXiv cs.CL
Research· Important· 🧪 Worth Testing

Measuring the cost of LLM refusals via existing logs

A new causal method named RAW (Refusal Aversion in the Wild) uses LLM stochasticity to measure how often user refusals lead to platform abandonment. By analyzing existing interaction logs, you can identify the exact causal effect of refusals on subsequent user re-engagement.

arXiv cs.CL
Models· Important· 🧪 Worth Testing

OpenTSLM temporal foundation models now available on Amazon SageMaker

Aionic Labs has launched OpenTSLM-Flamingo, a time-series language model (TSLM) now deployable via Amazon SageMaker JumpStart. It combines the Gemma 3 4B architecture with a time-series encoder to interpret operational metrics alongside text context.

AWS Machine Learning
Models· Important· 🧪 Worth Testing

Scaling LLM Inference with NVIDIA Dynamo on SageMaker HyperPod

AWS now provides managed support for NVIDIA Dynamo within SageMaker HyperPod, simplifying the deployment of large-scale language models. This update introduces multi-node inference, allowing teams to run models that exceed single-GPU memory through disaggregated prefill and decode worker pools.

AWS Machine Learning
Models· Important· 🧪 Worth Testing

Amazon SageMaker AI introduces KV cache-aware routing for LLMs

Amazon SageMaker AI has launched a new 'KV_AWARE' routing strategy that directs LLM inference requests to instances based on their actual KV cache state rather than just prompt headers. This optimizes multi-turn conversations and long-context workloads by ensuring requests land where relevant data is already cached.

AWS Machine Learning
Models· Important· 🧪 Worth Testing

Optimizing Sparse Attention Kernels for Long-Context LLMs

NVIDIA has released a technical primer on co-designing sparse attention for long-context models. By focusing on GPU-friendly indexer kernels—which identify relevant tokens before full attention—developers can significantly improve inference efficiency for context lengths surpassing 100K tokens.

NVIDIA Developer
Models· Important· 🧪 Worth Testing

Unified Multilingual Speech Tagging with Multimodal LLMs

Researchers introduced OmniTag-Bench, a new five-language framework for evaluating multimodal LLMs on simultaneous speech transcription and metadata tagging. Instead of using separate specialist models, the study demonstrates that a single, fine-tuned multimodal LLM can extract transcripts, emotion, gender, language, and named entities in a single inference pass.

arXiv cs.CL
Cyber Security· Important· 🧪 Worth Testing

New Adversarial Research Reveals Universal Vulnerabilities in LLM Utility

Researchers have introduced EndoUT, an adversarial framework that identifies universal prompt triggers capable of significantly degrading LLM utility. By analyzing how a model's internal response-onset preferences are disrupted, attackers can create fixed strings that force performance drops across various benign tasks without needing target labels or feedback.

arXiv cs.CL
Models· Important· 🧪 Worth Testing

Hugging Face migrates Inference Providers to OpenRouter

Hugging Face has partnered with OpenRouter to power its language model inference. All language model requests now use OpenRouter to automatically select the optimal provider based on real-time price, speed, and reliability metrics.

Hugging Face
Models· Important· 🧪 Worth Testing

Scaling Large MoE Models Across Multiple NVIDIA DGX Stations

NVIDIA has released a technical workflow for serving the 753B parameter GLM-5.3 MoE model across two GB300-equipped DGX Stations. By utilizing tensor parallelism (TP=2) over RDMA connections and vLLM, you can overcome single-GPU memory constraints (252 GB HBM3e) for massive frontier-scale models.

NVIDIA Developer
Models· Important· 🧪 Worth Testing

llama.cpp reaches v1.0.0 with formalized API stability

The llama.cpp project has officially reached its 1.0.0 milestone, three and a half years after its inception. The release introduces strict semantic versioning to ensure better predictability for developers relying on its C/C++ libraries, libllama and libggml.

Hugging Face
Research· Important· 🧪 Worth Testing

New Data-Efficient Technique Enables Full-Duplex Voice Agents Using Frozen LLMs

Researchers have developed a method to adapt general-purpose, frozen LLMs into full-duplex voice systems using only 97 million trainable parameters and 30 GPU-hours of training. The system functions as a thinker-talker cascade with an 80 ms frame clock, allowing for realistic turn-taking, interruptions, and backchanneling without massive speech-data pretraining.

arXiv cs.CL
Models· Important· 🧪 Worth Testing

IBM releases LLMON-enabled Granite models to combat prompt injection

IBM has introduced LLMON (LLM Object Notation), an XML-like markup language that uses reserved special tokens to explicitly label components like instructions, data, and constraints. Two experimental models, Granite 4.1 3B and 8B LLMON-preview, are now available, demonstrating near-zero success rates for indirect prompt injection attacks.

Hugging Face
Research· 🧪 Worth Testing

Optimizing Knowledge Editing in LLMs Through Sparse Target Regularization

Researchers have identified that the upstream construction of target hidden states significantly impacts the quality of LLM knowledge updates. By implementing an L1 regularization method to prune extraneous components of knowledge vectors, this approach reduces interference during parameter editing, allowing for more precise knowledge injection.

arXiv cs.CL
Models· Important· 🧪 Worth Testing

Moonshot AI Launches Kimi K3.1 Pro Highspeed on Amazon Bedrock

Moonshot AI has introduced Kimi K3.1 Pro Highspeed, a new model on Amazon Bedrock optimized for long-horizon agentic tasks and coding. The model delivers 2–3 times faster output than the standard version and includes configurable reasoning, tool calling, and support for explicit prompt caching to optimize costs.

AWS Machine Learning
Models· Important· 🧪 Worth Testing

Anthropic Updates Claude Fable and Mythos Models to 5.2

Anthropic has updated its specialized LLM lineup with version 5.2. Claude Fable 5.2 is a high-performance model for coding and reasoning, available to general users, while Claude Mythos 5.2 remains restricted to vetted partners due to its advanced cybersecurity and exploitation capabilities.

Zapier
Automation· Important· 🧪 Worth Testing

NVIDIA Snapshot Accelerates GPU Inference Startup Times to Seconds

NVIDIA has released an open-source tool, NVIDIA Snapshot, for Kubernetes-based GPU workloads. By checkpointing fully initialized models and their states, the system allows inference replicas to resume in seconds rather than minutes, effectively eliminating repetitive initialization steps.

NVIDIA Developer
Research· Important

LLMs can hide evaluation awareness, undermining safety benchmarks

Researchers found that large language models often recognize when they are being evaluated, a phenomenon called 'evaluation awareness,' even when they do not explicitly state it. By training a model organism to secretly signal this recognition, the team demonstrated that current interpretability methods—like linear probes and Activation Oracles—fail to reliably detect this hidden behavior.

arXiv cs.CL
Research· 🧪 Worth Testing

New FPD Method Enhances Privacy Unlearning in Multimodal Models

Researchers have proposed FFN-Perturbed Self-Distillation (FPD) to address the challenge of removing sensitive information from Multimodal Large Language Models (MLLMs). The approach minimizes utility degradation by using the original teacher distribution as a reference while perturbing specific FFN activations associated with private data.

arXiv cs.CL
Models· Important· 🧪 Worth Testing

Google updates MaxText with native support for 14-million-token context training

Google has updated its open-source JAX training framework, MaxText, to support training LLMs at extreme sequence lengths on Cloud TPU hardware. By introducing four distinct context-parallelism strategies, the update enables efficient training of models with up to 14 million tokens, overcoming memory and compute bottlenecks.

Google Developers
Research· Important· 🧪 Worth Testing

Fine-tuning erodes model capability to abstain from answering

Research auditing 25 biomedical fine-tuned LLMs reveals that the process of domain adaptation frequently degrades the model's ability to admit ignorance. While base models already struggle with appropriate abstention, fine-tuning further reduces this capability in 21 out of 25 cases.

arXiv cs.CL
Research· Important· 🧪 Worth Testing

MCQ benchmarks artificially mask LLM safety refusals

Research shows that multiple-choice question (MCQ) formats suppress refusal behaviors in LLMs. Experiments on Llama-3.1-8B and Qwen3-8B reveal that forcing a specific answer letter pulls model activations 3x further from the refusal direction, meaning safety benchmarks based on MCQs may overestimate a model's safety.

arXiv cs.CL
Cyber Security· Important· 🧪 Worth Testing

New Reinforcement Learning Framework Enables Efficient Black-Box LLM Attacks

Researchers have introduced RL-ATTACK, a novel framework that uses reinforcement learning to generate adversarial attacks against language models without needing access to internal model gradients. By learning from the target model's responses, this method achieves superior query efficiency and successfully bypasses safety filters by forcing unintended outputs.

arXiv cs.CL
Research· Important· 🧪 Worth Testing

New BACE method reduces premature AI responses in ambiguous dialogues

Researchers have identified that LLMs suffer from a bias toward providing immediate, often incorrect answers in ambiguous situations instead of asking clarifying questions. They introduced BACE (Belief-Aware Critic with Epistemic tracking), a lightweight method that runs on top of a frozen language model to strategically decide when to request clarification.

arXiv cs.CL