🤖 AI 资讯

每日 05:00 更新 · 11-04 · 主站 liuch.name ↗
招聘HR论文AI应用扩散模型大模型搜索RAG强化学习Google榜单评测Agent智能体模型评测算力芯片开源微调蒸馏OpenAI具身智能开发者生态推理思考网络安全端侧AI提示工程TransformerMeta模型发布模型安全对齐Anthropic预训练阿里巴巴政策监管融资
全部标签 →
筛选标签：Mistral · 返回个性化推荐 · 清空筛选
AI 资讯

How I use AI (Oct 2026)

Hacker NewsComments
2026-10-31T00:00:00.000Z · 大模型,AI应用,具身智能,开源,OpenAI,Google,Anthropic,Microsoft,Mistral,代码生成,Agent智能体,扩散模型,强化学习,微调蒸馏,提示工程,招聘HR,网络安全,榜单评测,开发者生态
AI 资讯

VIAS: Value-Isolating Activation Steering for Multi-Value Alignment and Compliance for Organisations

arXiv cs.AIarXiv:2611.03518v1 Announce Type: cross Abstract: Organisations increasingly deploy large language model agents that must follow organisational values. We check whether activation steering, which adds vectors to a model's internal activations during generation without changing its weights, can improve compliance beyond prompting alone. Since most scenarios involve several values at once, we focus on multi-value steering. A simple way to build these vectors is to contrast responses that follow all values relevant to a scenario with responses that violate them. However, this gives nearly identical vectors for different values: they capture a general compliance signal rather than individual values, which limits their use when combining several values. We propose Value-Isolating Activation Steering (VIAS), which contrasts responses to the same scenario under two prompts, one listing all relevant values and one omitting a single target value. We keep a pair only if the target value changes from compliant to non-compliant and both responses still meet a minimum compliance level on the other values. We evaluate VIAS on PROTECT, a workplace compliance dataset where each scenario is tagged with relevant values from a 21-value taxonomy. VIAS reduces overlap between value vectors, making them better suited for multi-value steering, although this alone does not lead to independent control over each value. Combined with prompting, VIAS complies with every tagged value on 69.6% of scenarios for Qwen2.5-7B, 75.2% for Mistral-7B, and 72.6% for Phi-3.5-mini, outperforming prompting alone for all three models.
2026-11-04 05:00:00 · 大模型,AI应用,阿里巴巴,Mistral,Agent智能体,提示工程,模型安全对齐,招聘HR,论文
AI 资讯

An Experimental Study of Hybrid Retrieval, NLI and Open-source LLMs for Scientific Claim Verification

arXiv cs.AIarXiv:2611.03093v1 Announce Type: cross Abstract: Claim verification is the task of deciding whether a source text supports a claim, contradicts it, or lacks sufficient evidence. Natural Language Inference (NLI) models can judge whether a single sentence supports or contradicts a claim, but they struggle when the evidence is spread across several sentences. Conversely, Large Language Models (LLMs) can reason across sentences, but may draw conclusions the text does not support. In this experimental study, we test whether a conservative fusion rule, labeling a claim Support or Contradict only when both models agree and defaulting to Not Enough Evidence otherwise, combines their strengths. We construct a benchmark of 2,035 claims from SciFact and CliniFact (858 Support, 152 Contradict, 1,025 Not Enough Evidence). We retrieve the top five most relevant sentences per claim using a hybrid of semantic and lexical (BM25) retrieval, which slightly outperforms either method alone. These retrieved sentences are then provided to every verification model. We then compare a DeBERTa-v3-base-mnli-fever-anli NLI baseline against Llama 3.1 8B, Qwen3 8B, and Mistral 7B. Qwen3 achieves the highest Macro-F1 (0.7937), followed by Llama 3.1 (0.7204) and Mistral (0.6388), while the NLI baseline reaches only 0.3157. Contrary to our hypothesis, fusion degrades performance, dropping Macro-F1 to 0.7392, 0.6730, and 0.5991, respectively. To understand this decline, we analyze agreements, disagreements, and common error patterns between the NLI and LLM verifiers. Our error analysis shows that NLI and LLM disagreements often result in correct LLM predictions being overturned.
2026-11-04 05:00:00 · 大模型,Meta,阿里巴巴,Mistral,强化学习,模型评测,榜单评测,论文
AI 资讯

They Know Where It Is, But Not What It Says: A Bilingual Hallucination Benchmark of Hadith Recall in Open LLMs

arXiv cs.AIarXiv:2611.02587v1 Announce Type: cross Abstract: Large language models are increasingly consulted for religious knowledge, including the recall of specific Hadith (recorded sayings and actions of the Prophet Muhammad, peace be upon him). Unlike open-ended generation, Hadith recall is a high-stakes, verifiable task: a reference (collection, chapter, and Hadith number) has exactly one correct text, and any deviation from it constitutes a fabrication that can be mistaken for authentic scripture. This study introduces a bilingual (English/Arabic) benchmark that probes five open-weight, locally-deployable language models (Mistral-7B, Gemma3-4B, Jais-7B, ALLaM-7B, and Phi4-mini-3.8B) across all six canonical Sunni Hadith collections (Sahih al-Bukhari, Sahih Muslim, Sunan Abi Da'ud, Jami' al-Tirmidhi, Sunan an-Nasa'i, Sunan Ibn Majah), sampling 500 Hadith per collection (3,000 per model per language, 30,000 responses total). Each model is given only the bibliographic reference and asked to reproduce the exact Hadith text, with no retrieval augmentation, and is scored on lexical fidelity (BLEU) and semantic fidelity (multilingual cosine similarity), with refusals flagged separately. Across all five models and both languages, 0% of 30,000 responses achieve both high lexical and high semantic fidelity, i.e., an exact, verifiable match to the source text. Arabic responses fare uniformly worse than English. These findings suggest that current open-weight LLMs, used without retrieval grounding, are unreliable and potentially dangerous sources for exact religious-text recall, and that deployment for Islamic scholarship should mandate retrieval-augmented generation over a verified corpus rather than parametric memory. The complete code, prompts, evaluation pipeline, and scored outputs are publicly available at https://github.com/dayam8696/Hadith-Hallucination-Benchmark.
2026-11-04 05:00:00 · 大模型,算力芯片,开源,Google,Mistral,扩散模型,模型评测,提示工程,模型安全对齐,论文
AI 资讯

From Lookups to Lessons: Turning Retrieved Knowledge into Parametric Memory

arXiv cs.AIarXiv:2611.00450v1 Announce Type: cross Abstract: 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 parametric memories enables reuse without online retrieval. However, training the memory to mimic a retriever rewards accurate independent predictions, which do not necessarily improve the final output quality when combined with the LLM's existing predictions. To address this, we introduce ResMem, in which the LLM reads retrieved text offline to generate teacher targets. This design avoids storing massive token representations, unlike existing parametric memory approaches such as MLP Memory, which relies on k-nearest neighbor (kNN) datastores. Furthermore, by using the LLM as an offline teacher that reads privileged retrieved evidence alongside the input, we train a parametric memory to learn an additive logit residual. This residual aligns the predictions of a student sharing the same backbone with the teacher distribution, while self-distillation updates remain confined to the parametric memory. On a Wikipedia subset, ResMem reduces the storage retained for teacher construction by 94.7% compared with the evaluated kNN pipeline. Across five question answering datasets, ResMem raises mean token F1 from 32.83% for the frozen LLM to 34.48%. It also breaks 21% fewer of the LLM's correct answers than a kNN-trained memory. These findings validate our motivation of treating retrieval as a learning resource, with parametric memory shaped by how retrieved knowledge is used to inform the model's future predictions. Code and model weights are available at https://github.com/SSRand/From-Lookups-to-Lessons and https://huggingface.co/RandHan/ResMem-Mistral-7B-v0.3.
2026-11-04 05:00:00 · 大模型,算力芯片,AI应用,开源,Google,Mistral,搜索RAG,扩散模型,微调蒸馏,端侧AI,论文

The Agent Said It Was Done. The Database Disagreed.

HuggingFace Blog
2026-10-03T22:56:48.455Z · 大模型,算力芯片,AI应用,具身智能,开源,OpenAI,Google,Anthropic,Microsoft,阿里巴巴,DeepSeek,xAI,Mistral,月之暗面,智谱,语音音频,Agent智能体,推理思考,搜索RAG,数字人,扩散模型,强化学习,微调蒸馏,模型评测,端侧AI,招聘HR,榜单评测,论文

Welcome Le Chonk: Mistral Large 4 on Hugging Face

HuggingFace Blog
2026-10-29T00:00:00.761Z · 大模型,算力芯片,AI应用,开源,Google,Meta,NVIDIA,阿里巴巴,DeepSeek,Mistral,智谱,多模态,Agent智能体,推理思考,搜索RAG,数字人,Transformer,扩散模型,MoE架构,强化学习,微调蒸馏,预训练,模型评测,提示工程,模型安全对齐,长上下文,端侧AI,招聘HR,榜单评测,开发者生态

NYT demands Brave shut down its paywall blocker and end deals with AI firms

Ars Technica

The New York Times and other newspaper companies sued Brave Software yesterday, alleging that the browser maker illegally bypasses paywalls and sells stolen news content to AI companies.

"Brave is the supplier, dealer, and distributor of stolen copyrighted content for its own users and many of the most prominent GenAI companies, including Anthropic, Perplexity, Mistral, xAI, and Cohere," the NYT alleged in a lawsuit filed in US District Court for the Southern District of New York. Citing third-party revenue estimates, the suit claimed that "Brave has made more than $100 million by copying and reselling publishers’ work without consent or permission," while "publishers have received nothing."

The lawsuit said, "AI companies pay Brave to provide them with real-time online content from the Brave search index," and use content from the Brave index "to train their own AI models" and "to respond to specific user queries." The NYT lawsuit points to Brave's website, which says the Brave Search API is "used by some of the biggest names in AI, including Cohere, Mistral, Perplexity, and You.com."

Read full article

Comments

2026-11-03 18:30:27 · 大模型,AI应用,Anthropic,xAI,Mistral,搜索RAG,招聘HR,版权诉讼,开发者生态

The New York Times Sues Brave Web Browser For AI Scraping, Bypassing Paywalls

SlashdotThe New York Times and several other publishers have sued (PDF) Brave Software, accusing the browser maker of scraping their websites without permission and helping users bypass paywalls. The Verge reports: According to a complaint (PDF) filed by the companies that run New York Daily News, the Chicago Tribune, and the Orlando Sentinel, Brave Software has monetized its index of scraped websites by, in part, providing AI companies with paid access to "real-time online content from the Brave Index." The AI companies then use that index to train their models, access content to "respond to specific user queries," or both. Brave has allegedly provided "stolen copyrighted content" to "many" AI companies including Anthropic, Perplexity, Mistral, xAI, and Cohere. Brave also promotes "hundreds of top publishers" as being accessible through its "Paywall blocker" feature in its browser. "Brave has never asked for the Publishers' permission to access their websites, much less to copy the material and provide it to users or other AI companies," the complaint says. "On the contrary, as noted above, Brave allows its users to bypass the Publishers' paywalls without their consent, and has taken other steps that prevent the Publishers from blocking Brave's access to the Publishers' websites. Brave has never compensated the Publishers for the unlawful use, copying, and distribution of the Publishers' copyrighted content." "While we believe in the ethical and responsible use and development of AI, Brave has flouted established copyright law and ignored repeated requests by The Times to stop its infringing conduct," Graham James, an NYT spokesperson, says in a statement. "We filed this lawsuit to hold Brave accountable for stealing our content to power its own products and to provide it to some of the most prominent and well resourced generative AI companies in the world."

Read more of this story at Slashdot.

2026-11-02 19:00:00 · 大模型,自动驾驶,Anthropic,xAI,Mistral,搜索RAG,招聘HR,版权诉讼,榜单评测,开发者生态
AI 资讯

An easy way to try local AI with braindead-local-intelligence

Lobsters

Comments

· 大模型,算力芯片,具身智能,开源,OpenAI,NVIDIA,阿里巴巴,Mistral,长上下文,端侧AI,招聘HR,开发者生态
AI 资讯

Robustness of Chatbot LLM Judgments of Presidential Actions and Hypothetical Reforms

arXiv cs.CLarXiv:2610.39398v2 Announce Type: replace Abstract: Chatbots are increasingly consulted about public affairs, and their judgments of policies are useful only if they are robust. We measure this robustness in the setting of an ordinary user, through consumer chat interfaces. For 30 days, five chatbot configurations (Mistral, Gemini Flash, Claude Sonnet, and Claude Opus at two effort settings) were asked twice a day to score the long-term effect on the United States of 40 actions of the last four presidential administrations and of 15 hypothetical reforms, each with a confidence and a short comment. The judgments are robust: all configurations agree on the sign and the ranking of the administrations' mean scores, and two different configurations asked on the same day give opposite signs to the same action in about 4% of cases. Responses from different days differ more than same-day responses (which shared the order of the items); the difference grows slowly over the month, except where the service itself changed. We identify three such changes, which shifted the judgments while leaving little or no trace in the interface. Treating the comments as a retrieval task, we find them specific to each item and consistent across runs and chatbots, with lexical and embedding similarities agreeing. We release the dataset of 16,500 judgments.
2026-11-02 05:00:00 · 大模型,Google,Anthropic,Mistral,对话助手,强化学习,向量数据库,招聘HR,论文
AI 资讯

BYOD: Build Your Own DLM - Efficient Autoregressive-to-Diffusion Language Model Conversion

arXiv cs.CLarXiv:2610.41574v1 Announce Type: new Abstract: Unlike autoregressive (AR) models, diffusion language models (DLMs) permit parallel token updates, flexible generation, and bidirectional attention, but leading systems rely on costly pretraining or full-model fine-tuning. We present Build Your Own DLM (BYOD), a generalizable LoRA-only recipe for parameter-efficient AR-to-DLM conversion. Without architecture-specific tuning, BYOD induces diffusion generation in Gemma 2 9B, Llama 3.1 8B, Qwen2.5 7B, and Ministral 8B; to our knowledge, this is the first reported AR-to-DLM conversion of a Mistral model. The adapters comprise 4.1-5.8% of final parameters while every base-model parameter remains frozen; each 25,000-iteration conversion needs at most 102.4M tokens, processed in 3.1-4.0 hours on one Google Colab GPU. All four models generate from fully masked sequences, use bidirectional context in a controlled attention test, and retain substantial parent-model knowledge. BYOD-Gemma is the strongest BYOD conversion on our nine-task evaluation, reaching 50.6% average accuracy compared with 51.9% for LLaDA 8B Instruct in the same harness. Open-generation quality at low network function evaluation (NFE) budgets remains the principal limitation: BYOD perplexity is 8.3-11.1 versus 4.1 for LLaDA and 2.2-3.0 for the AR parents when generating 128 tokens in 64 steps. To enable further experimentation and inspection, we provide a standalone training and evaluation implementation and interactive demos for all four conversions. BYOD thus provides a practical route for extending diffusion generation to other existing model families while identifying the gap that efficient conversion must next close.
2026-11-02 05:00:00 · 大模型,算力芯片,AI应用,Google,Meta,阿里巴巴,Mistral,搜索RAG,Transformer,扩散模型,微调蒸馏,预训练,论文
AI 资讯

Advancing Clinical Trial Outcome Prediction with Domain-Specific Large Language Models

arXiv cs.AIarXiv:2610.41710v1 Announce Type: new Abstract: Clinical trial failures are a major barrier to efficient drug development, incurring substantial costs and delaying patient access to effective therapies. Structured approaches such as the Hierarchical Interaction Network (HINT) predict trial outcomes by explicitly modelling interactions between drugs, diseases and protocols, but make limited use of the free text in trial registrations. We evaluate two domain-specific language models, BioBERT (108M parameters) and BioMistral (7B parameters), for phase-level outcome prediction on the TOP benchmark (12,477 trials), representing each trial as text built from ten ClinicalTrials.gov fields. We compare fine-tuning (full fine-tuning for BioBERT, low-rank adaptation for BioMistral) with probing, in which the pretrained model is frozen and only a lightweight classifier is trained on its embeddings. On the held-out test set of each phase, all four language-model variants achieved higher ROC-AUC (0.613--0.772) and PR-AUC (0.649--0.901) than the published HINT results, corresponding to relative PR-AUC gains of 10--20\%. F1 differences were small and mixed, and no model, including HINT, clearly exceeded the F1 of a trivial classifier that predicts success for every trial. Under the configurations tested, probing performed comparably to fine-tuning: fine-tuned models scored highest on ROC-AUC and PR-AUC in Phases I and II, and frozen BioMistral embeddings in Phase III (ROC-AUC 0.772; PR-AUC 0.901), although most differences between variants were below 0.03 and were not tested for significance. These results suggest that pretrained biomedical representations already encode signal relevant to trial outcome, and that probing is a computationally cheaper alternative to fine-tuning for this task.
2026-11-02 05:00:00 · 大模型,Mistral,微调蒸馏,预训练,模型评测,向量数据库,榜单评测,论文,开发者生态
AI 资讯

Multi-Test Nation Bias Benchmarking of LLMs with Explicit Unbiased Baselines

arXiv cs.AIarXiv:2610.41486v1 Announce Type: new Abstract: While prior work has highlighted that systematic nation-level bias in large language models (LLMs) can pose operational risks for international relations (IR) applications, many existing evaluations still lack a clearly specified unbiased reference (ground truth), limiting fully quantitative and cross-setting measurement. Using an expanded corpus of 722 official UN Security Council (UNSC) records, we measure nation-level bias with five complementary tests, thereby extending prior approaches to cover a broader set of bias expression channels: opinion ranking (WOW), association sorting (AK+-), vote simulation (PVS), opposition attribution (F-Oppose), and framing (Headliner). To enable grounded bias evaluation, we explicitly define an unbiased status for each test; in particular, for WOW and AK+- where ground truth is difficult to specify, we propose an action-based pseudo unbiased status derived from observable voting behavior to construct a quantifiable baseline. Benchmarking 4 LLMs (GPT-4o-mini, Llama 3.3-70B, Mistral-22B-Small, and Qwen3-30B) across the five permanent members of the UNSC, we find a test-agnostic macro-pattern: relative to the reference baselines defined for each test, model outputs lean toward the opposition (red) side for China and Russia and toward the supportive (blue) side for the U.K., the U.S., and France. These results underscore the need to audit IR-facing LLM deployments with grounded, multi-test benchmarks tied to explicit reference points.
2026-11-02 05:00:00 · 大模型,算力芯片,OpenAI,Google,Meta,阿里巴巴,Mistral,扩散模型,强化学习,模型评测,网络安全,论文
AI 资讯

Confident Fools or Careful Searchers? A Behavioural Study of When LLMs Search Beyond Their Static Internal Memory

arXiv cs.AIarXiv:2610.41124v1 Announce Type: new Abstract: Agentic LLM assistants must repeatedly decide whether to answer from memory or to call a tool such as web search. Unnecessary searches cost latency and money; missing ones yield confident but unsupported answers. Prior work mostly measures verbalised confidence or trains new adaptive-retrieval methods, and rarely audits the search decisions that deployed products actually make. We introduce a benchmark of 400 multiple-choice questions that combines widely known inside-knowledge (IK) questions, which should be answered directly, with outside-knowledge (OK) questions about facts from news published after the models' training cut-offs (to our knowledge), which should trigger search or, without tools, abstention. Each item offers an abstain option, and two action-level scores, NAS and KIS, reward searching on OK questions and restraint on IK questions. Across five commercial LLMs with and without their vendors' web search, GPT-5 mini and Grok 4 Fast search on over 90% of OK but only 1% of IK questions, whereas Mistral Small searches on just 20% of OK questions, often abstaining instead of using its tool; one product's search indicator proved unusable. Searching is partly triggered by explicit temporal cues: on uncued OK questions, the NAS of the two selective models falls from about 0.72 to 0.57--0.58. The scores thus measure the search triggering of specific products, not metacognition in a strict sense.
2026-11-02 05:00:00 · 大模型,AI应用,OpenAI,Meta,xAI,Mistral,Agent智能体,扩散模型,强化学习,模型评测,论文
AI 资讯

France’s AI Strategy: Turning Models, Compute, and Public Investment Into National Capability

Hacker Noon

France's potential role in AI rests on its ability to organise a chain of capabilities. Model developers can produce useful technologies, computing institutions can provide infrastructure, and public administrations can establish sustained demand. National capacity grows when these activities become connected through arrangements that work under practical constraints. The decisive question is whether French institutions can secure the resources they need, adapt them to their purposes and remain responsible for the resulting decisions.

That question matters because the visible scale of an AI strategy can obscure its operational content. In February 2025, the Élysée reported €109 billion in announced investments associated with AI infrastructure and deployment projects. [1] This signalled commitments by investors. It was neither a measure of installed computing capacity nor a statement that the French state had already spent that amount. Assessing France requires following projects and access arrangements beyond the announcement.

Models that users can adapt

In the Invention layer, an important distinction concerns what users can actually obtain and modify. Mistral's documentation lists Ministral 3 8B, released in December 2025, as generally available under Apache 2.0. [2] An available model with permissive licensing gives developers a different set of options from a service accessible only through a proprietary interface. It can support adaptation and deployment decisions closer to the institution using it.

That possibility is not equivalent to a complete operating capability. An administration still needs suitable infrastructure, technical staff, evaluation procedures and authority to use the relevant data. A model that performs well on a general benchmark may require considerable work before it can support an administrative task reliably. The French opportunity is to develop those supporting capacities around models that institutions can understand, test and replace.

Mistral's August 2026 announcement provides a useful illustration of access becoming more specific. The company said its regional endpoints were generally available, allowing customers to select inference in Europe or the United States. It also disclosed that limited, safeguarded transfers to subprocessors outside the selected region could occur. [3] This is evidence of a defined service offering, with stated limits. A European processing location alone cannot establish complete control over the surrounding technical and contractual relationships.

Computing infrastructure under shared authority

Industrialisation requires infrastructure that can support demanding work at useful scale. Alice Recoque illustrates both the ambition and the institutional complexity. EuroHPC signed its procurement contract in November 2025. The planned system will be owned by EuroHPC and hosted in France, with GENCI managing the project and CEA operating the machine. The Jules Verne consortium includes the Netherlands and Greece, and computing resources are to be allocated jointly in proportion to investment. [4]

France therefore occupies a significant organisational position without acquiring exclusive national disposal of the system. The planned architecture also combines European components with AMD processors and accelerators. [4] European coordination can enlarge access and influence while retaining external technological dependencies. The relevant policy issue is whether those dependencies remain manageable through procurement, support and continuity arrangements.

Project timing is equally consequential. GENCI's current project page identifies the end of 2027 as the expected operational date. [5] This profile therefore treats Alice Recoque as contracted future capacity. It does not add its promised performance to a present French inventory. The distinction affects decisions by laboratories and public organisations that need resources before the new system becomes available.

The state as an organised user

Operationalisation becomes more tangible in Albert API. DINUM, France's interministerial digital directorate, operates this shared gateway to generative models. Its official description covers access to open-weight and proprietary models, alongside functions such as document classification, retrieval and optical character recognition. It identifies Albert API as an instance of the state-developed OpenGateLLM software. [6]

A shared interface can help administrations avoid rebuilding the same connections and can make changing providers less disruptive. Its strategic value depends on the surrounding institutional arrangements: which data may be processed, who evaluates performance, how errors are corrected and who can suspend a failing service. DINUM's page states that an approved version for internal state data is still being finalised. [6] Claims about universal readiness for sensitive public workloads would therefore exceed the evidence.

This is where governance connects Invention, Industrialisation and Operationalisation. The state can shape demand through procurement specifications and common infrastructure, while retaining responsibility for each public use. An architecture that permits model substitution, independent evaluation and continuity of service would make national capability less dependent on the fortunes of a single supplier. Achieving this requires administrative expertise as well as technical investment.

The evidence to watch

Four developments would make France's trajectory easier to judge. First, documented acceptance and access milestones for Alice Recoque would establish when contracted infrastructure becomes usable. Second, reports on completed and energised facilities would distinguish delivered investment from announcements. Third, Albert API's authorisation status and service records would clarify which workloads it can responsibly support. Fourth, evaluations of public deployments would reveal whether AI improves accuracy, processing times or service quality at an acceptable cost.

France could play a substantial European role by connecting model development, shared computing and public demand into dependable arrangements. The strongest evidence would be institutions that can continue to operate, evaluate and adapt their systems when suppliers, prices or political conditions change. That would demonstrate a capacity to mobilise AI for public purposes, extending the significance of the companies and infrastructure located within French territory.

How to read this profile

This is original country analysis applying the GOAI capability conversion framework. It is not a peer-reviewed country assessment or a scored ranking. Proposed applications and indicators are analytical judgments; company and government claims are attributed to their sources.

Sources

All sources were accessed on 20 October 2026. Service characteristics reflect their providers' descriptions; expected dates remain forecasts.

  1. Élysée. France's AI strategy and announced investments. 11 February 2025.
  2. Mistral AI. Ministral 3 8B documentation. Model release: 2 December 2025; documentation checked on the access date.
  3. Mistral AI. Regional inference, open models and European infrastructure. 11 Au
2026-10-31 05:00:02 · 大模型,Mistral,扩散模型,强化学习,模型评测,招聘HR,网络安全,开发者生态
AI 资讯
AI 资讯

Comparing Human Role-Players and LLM-Simulated Clients in Online Counselling Training: An Analysis of Counselling Patterns

arXiv cs.AIarXiv:2610.40563v1 Announce Type: new Abstract: This study investigates the capabilities of Large Language Models to simulate counselling clients in educational role-plays in comparison to human role-players. Initially, we recorded role-playing sessions, where novice counsellors interacted with human peers acting as clients, followed by role-plays between humans and clients simulated by Mistrals Mixtral 8x7b using 4-bit quantization. These interactions were analysed with a counselling communication pattern system at sentence level. We investigated two key questions: (1) to what extent LLM-generated responses replicate authentic conversational dynamics and (2) whether counsellors' communication behaviour differs when interacting with human versus LLM-simulated clients. The findings highlight both similarities and differences in the application of counselling patterns across scenarios, showing the potential of LLM-based role-playing exercises to enhance counselling competencies and to identify areas for further refinement in virtual client simulations.
2026-10-31 04:00:00 · 大模型,Mistral,微调蒸馏,论文
AI 资讯

From Leniency to Suspicion: How an Open Post-Training Recipe Changes Whom LLM Fact-Checkers Believe

arXiv cs.AIarXiv:2610.38532v1 Announce Type: new Abstract: LLMs increasingly fact-check online content with the poster's identity in view. However, it remains unclear which post-training changes how much a model's truth judgement depends on the poster's social class, and how such a change happens inside the model. This paper investigates job-based doubt, i.e., judging a claim false more often when the poster has a lower-status job, using fact-checked claims with randomised poster profiles in base and post-trained checkpoints of six model families. Our results reveal that: 1) post-training creates such doubt only with the open T\"ulu-3 recipe (OLMo-2 32B, T\"ulu-3 70B), not with Meta's, Alibaba's or Mistral's; the effect is under one percentage point but statistically reliable, exceeding training drift and replicating on independent claim sets; 2) in OLMo-2 32B, the model does not start noticing the poster's job more, but starts interpreting it differently: a lower-status job pushes the base model toward "true" and the post-trained model toward "false"; and 3) this flip is carried by a single residual-stream direction and happens at supervised fine-tuning. The mechanism was confirmed on 400 new claims with pre-specified predictions. None of seven deployed commercial models doubts lower-status posters more. Our work underscores the need to audit speaker bias per training recipe.
2026-10-31 04:00:00 · 大模型,Meta,Mistral,扩散模型,微调蒸馏,网络安全,论文
AI 资讯
继续滚动加载更多…