Recent research reveals that language models fail to adapt to evolving user preferences, instead defaulting to 'exploit-first' behavior. While models can identify the correct strategy in multiple-choice scenarios, they show a 35-46% drop in calibration during open-ended generation, persistently over-relying on stale historical data.
Researchers have proposed 'Artifact Distillation,' a method where frontier models externalize their reasoning into an 'artifact bank.' Smaller models query this bank at inference time to improve performance without requiring expensive model retraining or gradient updates.
Researchers have proposed LOTUS, a framework for hallucination detection that uses multilinear low-rank decomposition to extract signals from hidden LLM layers. By analyzing peak activations and cross-layer patterns, the method identifies incorrect responses more effectively than existing pooling-based techniques.
Researchers have proposed an embedding-based k-center approach to select smaller, highly representative subsets of training data for Whisper fine-tuning. This method outperforms random sampling by capturing greater linguistic diversity, particularly effective when training budgets are limited.
Researchers introduced Recursive Semantic Divergence (RSD), an unsupervised metric that measures and penalizes inconsistency in LLM agent trajectories. By applying RSD as a reward signal during fine-tuning, models exhibit significantly more stable behavior and higher success rates across tool-use benchmarks.
Recent research on Qwen3.5-27B demonstrates that while activation interventions can force a model to produce a specific target output, they often fail to instill correct underlying reasoning. Researchers found that gains in target selection collapse when facts are recombined, suggesting that forced outputs do not reliably represent valid tool-use condition logic.
NVIDIA's BioNeMo team, working with the University of Washington's Institute for Protein Design, has introduced RFD4-Proteina, a model that generates both a protein's sequence and its full atomic structure in a single pass instead of handing off to a separate inverse-folding tool. By working with a residue-level representation and a streamlined diffusion process, the model needs only about 60 denoising steps instead of roughly 400, which translates into markedly quicker runs and a much smaller memory footprint than its predecessor RFD3 when generating a 200-residue protein. During generation, the model can also run a reward-guided search that scores and branches partial designs using GPU-accelerated evaluation tools such as OpenFold3 and TMol, steering computation toward the most promising candidates. It ships both as an open-source model on GitHub and as a NIM microservice reachable over HTTP, with toolkit skills available to let automated agents chain generation, scoring and evaluation steps together.
IBM Research released vLLM-Hook MIA, a plugin for the open-source vLLM inference engine that lets developers observe and intervene in a model's internal states at runtime — for example to decline malicious prompts, flag hallucinated facts, or adjust tone — without disabling vLLM's fast graph-replay execution mode. According to IBM's technical report, MIA is up to 20x faster than the earlier vLLM-Hook and up to 14x faster than methods that insert code directly into the model. In tests, steering an 8B Llama or 1.5B Qwen model added only about 2% latency, and a 70B Llama model added about 6%.
Amazon partnered with the independent nonprofit AVERI to audit the quality of its own internal safety-evaluation datasets for its AI models, not just the models' outputs. AVERI reviewed a random, anonymized sample of safety-test prompts covering areas like harmful content, cybersecurity, privacy and copyright. The review confirmed strong topical coverage, diverse testing methods and low contamination risk, but flagged gaps in multilingual coverage and multistep adversarial testing. Amazon responded with four concrete changes to its evaluation process.
Google Research announced the final milestone of its Coral NPU open hardware standard — a 32-bit RISC-V architecture with a new Matrix Execution Unit for up to 1 TOPS at roughly 10mW — plus a v1.0 open-source IREE/MLIR compiler that takes JAX, PyTorch and TensorFlow models directly to the hardware. Synaptics is launching the first commercial silicon built on this standard, including the Astra SR200 microcontroller series (400 GOPS) and new Coralboard SR and SL developer kits, with the SL now generally available via DigiKey. Google is also partnering with VeriSilicon and GlobalFoundries to bring the open RTL into production chips.
A new study questions the common belief that whether an AI agent falls for a prompt injection is settled the moment it reads the malicious text. The researchers built a branching-rollout technique on AgentDojo benchmarks: they replayed an agent's run up to the point it encounters an injected instruction, then let the rest of the episode play out many different ways to measure how much of the final outcome was already locked in versus still open. For a Qwen3 agent running without extended reasoning, 91% of the outcome was already determined at that moment; switching on reasoning ('thinking') cut that figure to 62%, showing the agent's compliance decision often keeps forming well after it first sees the attack. Enabling reasoning also cut Qwen3's susceptibility to successful attacks, but made two other tested models more vulnerable, and benchmark scores for reasoning agents flipped on a rerun as often as 35% of the time.
AWS released a push-based NIXL connector for vLLM that changes how key-value (KV) cache data moves between prefill and decode instances in disaggregated LLM serving. Instead of the decode instance pulling KV blocks after prefill finishes, the decode instance registers its target memory in advance and the prefill instance writes the KV data directly once ready, moving coordination work off the critical path. Benchmarked on a 1-prefill/1-decode Qwen3-32B deployment using EC2 P5en instances with EFA, push mode delivered up to 3x better time-to-first-token (TTFT) and up to 30% better time-per-output-token (TPOT) than the default pull mode, especially under high request rates.
Google Research describes new work on grounding AI agents in physical lab reality, using a digital-twin framework called VERID to let agents safely plan and execute liquid-handling robot protocols. Combined with closed-loop molecular design cycles for aptamers and proteins, the agents iteratively learned from wet-lab results, improving hit rates over multiple rounds and discovering single-digit nanomolar protein binders for a cancer-related target.
Google DeepMind demonstrated how its agentic platform Google Antigravity, combined with the ProtNLM model, helped researchers search millions of uncharacterized proteins in UniProt to find a novel antimicrobial peptide candidate against respiratory pathogens. The AI agent handled database queries, structural comparison via AlphaFold, and genomic context analysis, while a human researcher guided the scientific strategy. The resulting candidate is now undergoing wet-lab validation, and DeepMind introduced a new 3D Biomolecular Structure Viewer built into Antigravity for shared human-agent inspection of protein structures.
AWS detailed three kernel-level optimizations for Mixture-of-Experts (MoE) model serving on Hopper GPUs, implemented in collaboration with the vLLM community and shipped in Amazon SageMaker AI's LMI container. The changes target the inefficiencies of fused MoE computation during low-batch decode, which previously left Tensor Cores underutilized. Combined, the optimizations deliver up to 10% higher throughput and up to 9% faster time-to-first-token (TTFT).
IBM Research used a coordinated multi-agent 'agent factory' to explore hardware design optimizations for a quantum error-correction decoder, comparing proprietary and open frontier models under the same 100-hypothesis search budget. The open-model configuration, run on-premises via IBM's llm-d distributed inference framework, achieved a 52x speedup over baseline for about $26 in inference cost, versus $299 for a proprietary-model configuration that reached a higher 204x speedup. The study highlights that cache-aware routing and efficient serving (98% prefix-cache hit rate, ~$0.11 per million tokens) made the open-model approach dramatically cheaper while still finding valid, compact hardware designs.
Apple ML researchers published a NeurIPS paper introducing SELFCOMPACT, a scaffold that lets language model agents decide for themselves when and how to compact their own accumulated context, rather than relying on fixed-interval summarization triggered by a token count. It combines a compaction tool the model can call with a lightweight rubric telling it when to fire (a sub-task is resolved) and when to hold off (mid-derivation or stuck). Tested on six benchmarks and seven models, it matched or beat fixed-interval summarization while using far fewer tokens.
Amazon Web Services researchers published a method called 'priming' that converts pretrained Transformer models into hybrid architectures mixing attention and state space model (SSM) layers, reusing existing weights instead of retraining from scratch. The approach cuts pretraining costs by over 99% and lets a 32-billion-parameter hybrid model match its source Transformer's quality while decoding up to twice as fast. Amazon tested three SSM types and released an open-source model zoo, training code, and serving tools under Apache 2.0.
At the OCP Global Summit, NVIDIA contributed its Vera Rubin rack and compute tray designs to the open MGX architecture and announced a series of partnerships covering AI factory power, cooling and networking standards. Partners include hyperscalers such as Meta, Google, Microsoft and AMD, plus Cisco and Visionbay. The announcements cover rack design specs (MGX 1.2), new open-source software for managing AI data centers (DSX OS components), grid resilience standards, an 800VDC power architecture, new Ethernet switching (Spectrum-6) and a new optical transceiver standard (USFP) with Cisco.
NVIDIA described a workflow built into Megatron Core that makes large-scale pretraining bitwise deterministic, meaning independent or checkpoint-resumed training runs follow the exact same numerical trajectory. Using tensor fingerprinting across training phases, modules, operations and kernels, engineers can trace nondeterminism down to a specific kernel and fix it. On a trillion-parameter Nemotron model running on 2,432 GPUs, optimizations brought the performance cost of enforcing determinism (the 'determinism tax') down to about 2%, while keeping bitwise-identical results over 800 training steps.
Apple researchers, with National University of Singapore collaborators, published a paper on RISED, a new training framework for LLM agents that must learn across many different interactive environments at once. Instead of relying only on scalar reward signals, RISED uses an LLM judge to tag each training rollout with rubrics describing its behaviour, then uses these rubric tags to select training data and to supply extra supervision. The authors report RISED achieves the highest average success rate across environments compared to existing approaches, ranking first or second in every tested environment.
Google Research released a report from its Contextual Agent Privacy and Security (CAPS) workshop, co-authored with over 50 academic and industry researchers. The report argues that AI agents need to reason about context-specific social norms, not just static permissions, to act appropriately when handling personal data and taking actions on a user's behalf. It proposes a multi-layered approach spanning system-level sandboxing, model-level reasoning, user-centric controls, multi-agent guardrails, and ecosystem governance, grounded in the theory of Contextual Integrity.
GitHub has published ReviewBench, a research-preview benchmark designed to compare AI code review agents on a common, auditable standard. The dataset draws on 219 real public pull requests across 19 languages and 187 repositories, chosen to mirror the language and size distribution found across 103.9 million GitHub pull requests. Ground truth comes from a blended pool of human reviewers, static analysis tools, and multiple frontier LLMs, cross-checked by senior engineers who agreed with the benchmark's labels 96.6% of the time. GitHub says it has already used ReviewBench internally to track its own Copilot code review (CCR) product, and that offline benchmark movement reliably predicted the direction of later production A/B tests.
AWS researchers describe a system that models telecom networks as a continuously updated graph ('digital twin') and uses AI agents to run a cascade of graph algorithms (decomposition, clustering, centrality ranking) to pinpoint root causes of failures. The approach was demonstrated with NTT DOCOMO at Mobile World Conference, reportedly cutting root cause analysis from hours to minutes on commercial networks. Agents also consult an incident knowledge base and runbooks, and can escalate autonomy from advisory to supervised to full remediation over time.