{"slug":"large-language-2026-10-21","label":"Large Language","item_count":29,"day_count":13,"source_count":10,"first_seen":"2026-10-21T15:38:05+00:00","last_updated":"2026-11-09T17:58:38+00:00","generated_at":"2026-11-11T00:14:41.869679+00:00","sources":["anthropic_research","arxiv_agent_systems_research","arxiv_cs_ai","arxiv_cs_cl","arxiv_llm_reliability","hackernews_ai","huggingface_blog","infoq_ai_ml","search_cn_open_weight_labs","vllm_blog"],"days":[{"date":"2026-10-21","items":[{"title":"Automated Materials Data and Knowledge Extraction using Machine Learning and Large Language Models","url":"http://arxiv.org/abs/2610.29895v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"Over the past decade, substantial efforts have been devoted to developing machine learning (ML) methods for extracting materials data and, where possible, materials knowledge directly from the scientific literature. T...","why_it_matters":"Matches feed focus: agentic, evaluation.","sid":"16071b3c73317004","published":"2026-10-21T15:38:05+00:00","editor_note":"First member: a materials-science extraction survey, unrelated to the rest beyond the phrase."},{"title":"Self-Assessment Reinforcement Learning for Search-State Awareness in Retrieval-Augmented Large Language Models","url":"http://arxiv.org/abs/2610.30435v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Recent advances in search agents have enabled large language models to tackle complex tasks by retrieving and synthesizing external information. However, existing search agents often lack search-state awareness, i.e.,...","why_it_matters":"Matches feed focus: agent, eval.","sid":"276a486ae6eefe6b","published":"2026-10-21T20:46:24+00:00","editor_note":"Search agents trained via RL to know when they have retrieved enough."}]},{"date":"2026-10-22","items":[{"title":"DeepSeek vs Mistral Large 4 vs Cohere: 9x Price Gap [2026]","url":"https://news.google.com/rss/articles/CBMijwFBVV95cUxPN1hlSUJENkV6eFRCbEVzMl8zU3p4YjBTQ1dra0RTU2QyWnh3ZlVYZDB6cUNqdXNtUjFDZFZNYkpIVy1DV0F6S0tNdXpFZ3BkMWZLYXZaZ3VOUkczanNSS2RDUkZYWmhqUVZwRXhPaHB5LTc1dDVRM3IxSjBGNnl4djlEV2NCeV9uVWh5UWN5VQ?oc=5","source":"search_cn_open_weight_labs","type":"news","summary_1line":"DeepSeek vs Mistral Large 4 vs Cohere: 9x Price Gap [2026] tech-insider.org","sid":"590975aa217085b7","published":"2026-10-22T00:44:58+00:00","editor_note":"Search-sourced price comparison putting Mistral Large 4 against DeepSeek and Cohere."},{"title":"Spectral Smoothing and Spiking Perturbations Reconfigure Capability Profiles and Selective Task Performance in Large Language Models","url":"http://arxiv.org/abs/2610.30676v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Singular-value spectra encode important structure in transformer feed-forward projections, yet their causal role in downstream behavior remains underexplored. We study post-hoc spectral smoothing and spiking intervent...","why_it_matters":"Matches feed focus: eval.","sid":"8debe4581e047c88","published":"2026-10-22T03:42:07+00:00","editor_note":"Interpretability work on how spectral edits to feed-forward weights shift task performance."},{"title":"Parameter-Efficient Adaptation of Large Language Models to a Low-Resource Agglutinative Language: A Corpus, Datasets, and a Reproducible Benchmark for the Bashkir Language","url":"http://arxiv.org/abs/2610.31397v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Bashkir is a low-resource agglutinative language written in Cyrillic and remains poorly represented in modern LLMs. We present an open resource suite and a systematic benchmark of adaptation methods: the Bashkir Web C...","why_it_matters":"Matches feed focus: evaluation.","sid":"f07fa8132755e14c","published":"2026-10-22T14:59:55+00:00","editor_note":"Low-resource language adaptation benchmark for Bashkir."},{"title":"An Empirical Study of Recent Large Language Models for Knowledge Graph to Text Generation on Standard Benchmarks","url":"http://arxiv.org/abs/2610.31529v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Knowledge graph-to-text (KG-to-text) generation verbalizes a set of knowledge graph triples into fluent natural language. While fine-tuned encoder-decoder models have long been the standard approach, recent LLMs are b...","why_it_matters":"Matches feed focus: evaluation.","sid":"c888971c48a268c3","published":"2026-10-22T16:36:35+00:00","editor_note":"Empirical comparison of recent LLMs on knowledge-graph-to-text generation."}]},{"date":"2026-10-26","items":[{"title":"LongSearchEval: Evaluating Long-Horizon Web Search Capabilities of Large Language Models via a Trajectory-Aware Agent Framework with Failure-Node Localization","url":"http://arxiv.org/abs/2610.35150v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"Long-horizon web search requires LLM-based agents to integrate information across web pages through extended interactions under task-specific constraints. Evaluating this capability is important not only for measuring...","why_it_matters":"Matches feed focus: agent, evaluation.","sid":"5f97c21b490b1ce9","published":"2026-10-26T10:15:38+00:00","editor_note":"Long-horizon web-search eval that localizes the failing step in an agent trajectory."},{"title":"CANS: Capability-Preserving Neuron Steering for Large Language Models","url":"http://arxiv.org/abs/2610.35530v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Activation steering offers training-free and dynamic control of target behaviors in large language models, but existing methods can impair general capabilities and require domain-specific intervention designs. We intr...","why_it_matters":"Matches feed focus: eval.","sid":"d5cd3db9b5f4a99b","published":"2026-10-26T14:45:23+00:00","editor_note":"Activation-steering method that aims to keep general capability intact."},{"title":"Reva: Real-Time Vulnerability Detection through Contextual Probing of Large Language Models","url":"http://arxiv.org/abs/2610.35829v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"A vulnerability costs more to fix the later it is discovered. As coding agents generate code faster than developers can review it, software needs a first line of defense screening each new function in real time. LLM a...","why_it_matters":"Matches feed focus: agentic.","sid":"a2f964aa6fcbf7ea","published":"2026-10-26T17:50:09+00:00","editor_note":"Real-time vulnerability screening of agent-generated code via model probing."}]},{"date":"2026-10-27","items":[{"title":"CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models","url":"http://arxiv.org/abs/2610.36734v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Continual learning requires large language models to acquire new knowledge and capabilities while retaining what they have learned. Fine-tuning and replay support acquisition and retention, but deciding whether incomi...","why_it_matters":"Matches feed focus: eval.","sid":"200271de2a2f52aa","published":"2026-10-27T06:52:54+00:00","editor_note":"Continual-learning replay method that checks compatibility of incoming data."}]},{"date":"2026-10-28","items":[{"title":"Gold vs.~Translation: Are We Measuring Language, Knowledge, or Locality in Translated Benchmarks?","url":"http://arxiv.org/abs/2610.38788v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Translated versions of English-origin benchmarks such as MMLU are widely used to evaluate large language models in non-English languages, yet it remains unclear whether they measure the same constructs as natively created benchmarks. We frame this as a proxy validity question and","sid":"e826ccda9d048b84","published":"2026-10-28T19:56:32+00:00","editor_note":"Questions whether translated benchmarks like MMLU measure the same thing as native ones."}]},{"date":"2026-10-29","items":[{"title":"Day 0 support for Mistral Large 4 on vLLM","url":"https://vllm.ai/blog/2026-10-29-mistral-large-4","source":"vllm_blog","type":"news","summary_1line":"Day-0 vLLM support for Mistral Large 4 with sparse MLA, cross-layer KV sharing, DSpark speculative decoding, DCP/PCP long context and NIXL P/D on NVIDIA Blackwell.","sid":"37f6c0c263213efd","published":"2026-10-29T00:00:00+00:00","editor_note":"The launch's serving detail: vLLM day-0 support and the inference features it relies on."},{"title":"Welcome Le Chonk: Mistral Large 4 on Hugging Face","url":"https://huggingface.co/blog/mistral-large4","source":"huggingface_blog","type":"research","summary_1line":"Welcome Le Chonk: Mistral Large 4 on Hugging Face","sid":"b6b30743068c2627","published":"2026-10-29T00:00:00+00:00","editor_note":"Hugging Face confirms the weights are published on the Hub."},{"title":"Incentivizing Self-Learning in Large Language Models via Reinforcement Learning","url":"http://arxiv.org/abs/2610.39671v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Large language models (LLMs) draw on parametric knowledge and retrieval-augmented generation (RAG) to solve knowledge-intensive tasks. Yet problem solving in unfamiliar domains calls for tightly coupled knowledge acquisition and reasoning, beyond factual retrieval and aggregation","sid":"d571e973b6e6916d","published":"2026-10-29T09:21:26+00:00","editor_note":"Opens the memory sub-thread: RL to couple knowledge acquisition with reasoning."},{"title":"FIRME: Registry-Anchored Memory Graphs for Evolving, Heterogeneous Investor Knowledge","url":"http://arxiv.org/abs/2610.40482v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Financial AI workflows that reuse investor knowledge need persistent memory that integrates observations across sources and periods without losing identity or provenance. In a knowledge graph, recurring observations may use different surface forms, fragmenting relationship histor","sid":"e519641640234319","published":"2026-10-29T17:13:45+00:00","editor_note":"Registry-anchored memory graph for persistent, provenance-preserving agent memory."}]},{"date":"2026-10-30","items":[{"title":"ImplicitHop: Evaluating knowledge integration across context and memory","url":"http://arxiv.org/abs/2610.41169v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"The ability to integrate new information and deduce its consequences is central to learning, understanding, and reasoning in large language models (LLMs). Whether information arrives during training or in-context, mod...","why_it_matters":"Matches feed focus: agent, evaluation.","sid":"1d7852cb09b3031f","published":"2026-10-30T02:44:51+00:00","editor_note":"Eval of whether models integrate facts arriving in context versus training."},{"title":"Context as Infrastructure: Knowledge Compilation for Retrieval in Agentic Systems with Harnesses","url":"http://arxiv.org/abs/2610.41172v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Retrieval-augmented generation is becoming agentic. A search harness writes queries, reads results, reformulates, and decides when to stop. This largely solves accuracy: a harness with only BM25/grep and MiniMax M3 an...","why_it_matters":"Matches feed focus: agentic, harness, eval.","sid":"ce941c554e9fce8e","published":"2026-10-30T02:48:37+00:00","editor_note":"Argues agentic search harnesses solve accuracy and reframes context as compiled infrastructure."},{"title":"Reimagining Text-to-SQL: A Survey of Large Language Model Architectures and Strategies","url":"http://arxiv.org/abs/2611.00171v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"Text-to-SQL systems translate natural language questions into executable SQL queries, democratizing access to relational data. Early approaches relied on handcrafted rules and template matching, while subsequent neura...","why_it_matters":"Matches feed focus: agentic, evaluation.","sid":"e46c1ab3f7ca9939","published":"2026-10-30T13:45:49+00:00","editor_note":"Survey of LLM architectures for text-to-SQL."},{"title":"Prune Gradually, Adapt Sparsely: Efficient Pruning of Large Language Models at Extreme Sparsity","url":"http://arxiv.org/abs/2610.41968v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Weight pruning can reduce the storage and inference cost of large language models (LLMs), but preserving quality at high sparsity levels remains challenging. Existing methods face a quality-cost trade-off in this regi...","why_it_matters":"Matches feed focus: eval.","sid":"a8a5f23cd8e10643","published":"2026-10-30T16:17:43+00:00","editor_note":"Gradual pruning method targeting quality at extreme sparsity to cut inference cost."},{"title":"From Lookups to Lessons: Turning Retrieved Knowledge into Parametric Memory","url":"http://arxiv.org/abs/2611.00450v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Retrieval-augmented generation (RAG) helps large language models (LLMs) answer factual questions beyond their training data, but it adds storage and inference costs. Distilling previously retrieved information into pa...","why_it_matters":"Matches feed focus: eval.","sid":"0f8439521ab02185","published":"2026-10-30T23:05:50+00:00","editor_note":"Distills retrieved knowledge into weights to reduce RAG storage and inference cost."}]},{"date":"2026-11-02","items":[{"title":"Progress on provable inference for large language models","url":"https://www.anthropic.com/research/provable-inference","source":"anthropic_research","type":"research","summary_1line":"Sharing progress on a powerful type of cryptographic proof known as a zkSNARK, which is uniquely fit for provable inference.","sid":"43862ef771dfba65","published":"2026-11-02T00:00:00+00:00","editor_note":"Anthropic's progress report on zkSNARK proofs for verifiable inference."},{"title":"Article: Systemic Approach to Memory, Knowledge, and Context in Agentic AI Architectures","url":"https://www.infoq.com/articles/agentic-ai-context-memory-knowledge/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering","source":"infoq_ai_ml","type":"news","summary_1line":"The article argues that reliable agentic AI architecture depends on a context engine that selects, injects, compresses, and expires information before it reaches the model. It separates context, memory, and knowledge,...","why_it_matters":"Matches feed focus: agentic.","sid":"744a7a52285cfa40","published":"2026-11-02T11:00:00+00:00","editor_note":"Practitioner framing: a context engine that selects, compresses, and expires information."}]},{"date":"2026-11-03","items":[{"title":"Feeling of Knowing in Large Language Models","url":"http://arxiv.org/abs/2611.04834v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Post-training updates change what large language models (LLMs) can answer, but may leave their answerability judgments misaligned. We study this knowledge-judgment desynchronization through functional feeling of knowi...","why_it_matters":"Matches feed focus: evaluation.","sid":"94da3bfcbd46a769","published":"2026-11-03T13:04:46+00:00","editor_note":"Shows post-training can desync what a model knows from what it thinks it knows."}]},{"date":"2026-11-04","items":[{"title":"Reasoning over Non-Parametric Memory in Multilingual LLMs: A Cognitive-Based Analysis of Relational Knowledge","url":"http://arxiv.org/abs/2611.06715v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Retrieval-augmented generation (RAG) assumes that language models can reason over relations stated in their context rather than fall back on associations memorized during pretraining. We study this evidence-conditioned relational reasoning in isolation, with the facts supplied di","why_it_matters":"Matches feed focus: eval.","sid":"63ff637171c98d2c","published":"2026-11-04T16:11:26+00:00","editor_note":"Tests whether models reason over relations in context or fall back on memorized ones."}]},{"date":"2026-11-05","items":[{"title":"FedReason: Large Language Models as Weight-Space Meta-Aggregators for Heterogeneous Federated Learning","url":"http://arxiv.org/abs/2611.08383v1","source":"arxiv_cs_ai","type":"paper","summary_1line":"Federated learning servers combine client models with a fixed aggregation rule, and each rule in common use is designed for one kind of heterogeneity. We study whether a large language model (LLM) that inspects the client models can choose the rule itself, without any training on","sid":"416b05f86baa94c5","published":"2026-11-05T18:00:58+00:00","editor_note":"Uses an LLM to pick the federated-learning aggregation rule."},{"title":"Show HN: Thinker – Trim context and cache knowledge to extend agent usage limits","url":"https://thinker.sh/how-it-works","source":"hackernews_ai","type":"news","summary_1line":"I’ve been doing a lot of solo development lately and found myself hitting Claude token limits halfway through the week constantly. Even with a Max plan, I’d have to switch over to Codex or Cursor, and it would be a bi...","why_it_matters":"Matches feed focus: agent, harness, codex.","sid":"464b347f5551c3b2","published":"2026-11-05T19:22:23+00:00","editor_note":"Show HN tool that trims context and caches knowledge to stretch agent usage limits."}]},{"date":"2026-11-06","items":[{"title":"ROSE: Risk-Aware Orthogonal Subspace Navigation for Lifelong Knowledge Editing in Multimodal Large Language Models","url":"http://arxiv.org/abs/2611.09770v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Multimodal Large Language Models (MLLMs) require continuous factual updates to stay current, yet lifelong knowledge editing remains a daunting challenge. The primary obstacle is the severe Locality Erosion and Catastr...","sid":"5606760643e4651e","published":"2026-11-06T17:32:49+00:00","editor_note":"Lifelong knowledge editing method for multimodal models."}]},{"date":"2026-11-09","items":[{"title":"More Than Just Style: Benchmarking Personality Influence in Large Language Models","url":"http://arxiv.org/abs/2611.12392v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"In addition to the effects of explicit input prompts and instructions, large language model outputs are influenced by underlying behavioral preferences similar to human personality tendencies. These tendencies can aff...","why_it_matters":"Matches feed focus: evaluation.","sid":"9c88a41fbb756af3","published":"2026-11-09T13:29:45+00:00","editor_note":"Benchmark of how personality-like tendencies shift model outputs."},{"title":"Large Language Models as a Universal Searcher for Open-World Automated Machine Learning","url":"http://arxiv.org/abs/2611.12791v1","source":"arxiv_cs_cl","type":"paper","summary_1line":"Developing effective machine learning models requires complex manual tuning, while current Automated Machine Learning (AutoML) frameworks remain domain-specific and isolate model architecture, training hyperparameter,...","why_it_matters":"Matches feed focus: eval.","sid":"3b4f703ae0a8f385","published":"2026-11-09T17:58:38+00:00","editor_note":"Latest member: an LLM as a cross-domain AutoML searcher."}]}],"editorial":{"tldr":"This is not one story: the items share the phrase \"large language models\" and are mostly arXiv papers from Oct 21 onward. Two real threads sit inside it: the Mistral Large 4 launch on Oct 29, and a run of papers on how LLMs and agents handle memory, retrieval, and knowledge updates.","stale":false,"whats_new":"The Nov 9 additions are two arXiv papers, a personality-influence benchmark and an LLM-as-AutoML-searcher, so no new event has landed since Mistral Large 4 got day-0 vLLM and Hugging Face support on Oct 29.","why_it_matters":"The only deployable change here is Mistral Large 4 on vLLM; the memory and retrieval papers are research signal for agent context design, not shipped tooling.","take_for_builders":"If you serve open weights, test Mistral Large 4 on a current vLLM build before adopting it; treat the memory and retrieval papers as design input for your context layer, not drop-in components.","beats":[{"kicker":"BACKGROUND","tone":"neutral","headline":"Unrelated LLM papers share a phrase, not an event","summary":"Oct 21–28 arXiv work spans materials extraction, steering, continual learning, translated benchmarks, and a long-horizon web-search eval.","sids":["16071b3c73317004","276a486ae6eefe6b","8debe4581e047c88","f07fa8132755e14c","c888971c48a268c3","5f97c21b490b1ce9","d5cd3db9b5f4a99b","a2f964aa6fcbf7ea","200271de2a2f52aa","e826ccda9d048b84"]},{"kicker":"LAUNCH","tone":"launch","headline":"Mistral Large 4 ships with day-0 vLLM and Hugging Face support","summary":"vLLM lists sparse MLA, cross-layer KV sharing, speculative decoding, and P/D disaggregation on NVIDIA Blackwell.","sids":["590975aa217085b7","37f6c0c263213efd","b6b30743068c2627"]},{"kicker":"THEME","tone":"rising","headline":"Memory, retrieval, and knowledge updates become the densest sub-thread","summary":"Papers and practitioner pieces Oct 29–Nov 6 on RAG-to-parametric distillation, memory graphs, context engines, and knowledge editing.","sids":["d571e973b6e6916d","e519641640234319","1d7852cb09b3031f","ce941c554e9fce8e","0f8439521ab02185","744a7a52285cfa40","94da3bfcbd46a769","63ff637171c98d2c","464b347f5551c3b2","5606760643e4651e"]},{"kicker":"NOW","tone":"now","headline":"Scattered research continues, led by Anthropic's provable-inference update","summary":"Text-to-SQL survey, extreme-sparsity pruning, zkSNARK provable inference, federated aggregation, personality benchmarking, and AutoML search.","sids":["e46c1ab3f7ca9939","a8a5f23cd8e10643","43862ef771dfba65","416b05f86baa94c5","9c88a41fbb756af3","3b4f703ae0a8f385"]}],"open_questions":["Do independent benchmarks of Mistral Large 4 on vLLM confirm its long-context and throughput claims?","Does any provider publish Mistral Large 4 API pricing that confirms or overturns the reported price gap with DeepSeek and Cohere?","Does Anthropic ship provable inference (zkSNARK) as a product feature or customer-verifiable API option?","Does any agent framework adopt a separate context/memory/knowledge layer like the one InfoQ and the harness paper describe?"],"generated_at":"2026-11-11T00:13:42Z"}}