60
Smart search ranks by meaning as well as keywords (one row per story, last 45 days).
OpenArch – PyTorch implementations of modern LLM architectures
OpenArch provides PyTorch reference implementations of modern LLM architectures for developers and researchers.
OpenArch is a GitHub project offering PyTorch implementations of modern large language model architectures. The repository attracted 43 points and 7 comments on Hacker News. It targets developers and researchers who want readable, runnable versions of current LLM architectures.
15
KindaRails2Shell threatens Ruby on Rails apps (CVE-2026-66066)
Critical CVE-2026-66066 in Rails' Active Storage/libvips allows unauthenticated arbitrary file read and possible RCE; active exploitation now observed.
CVE-2026-66066 (KindaRails2Shell), discovered by Ethiack researchers and independently by RyotaK of GMO Flatt Security, lets attackers upload crafted files that exploit libvips' handling of specialty formats to read arbitrary files, including process environment secrets, potentially escalating to RCE. Default Rails 7.0+ setups using Active Storage with the vips processor are affected before versions 7.2.3.2, 8.0.5.1, and 8.1.3.1; fixes shipped July 29, 2026, with VIPS_BLOCK_UNTRUSTED as a partial mitigation. Proof-of-concept exploits circulated after disclosure, and VulnCheck updated that it observed active exploitation originating from a single French IP establishing C2 to a host in Israel. Akamai deployed WAF rules, but experts stress patching and credential rotation over filtering alone.
82
PRs NOT Welcome: How Top AI Open Source Projects Are Managing Thousands of Contributors
Top AI open source projects like Vercel, Astro, Flue, and tldraw are restricting external PRs and using agent-based software factories to triage, fix, and review contributions.
Several prominent AI-native open source projects are closing or limiting external pull requests, largely because submissions are often AI-generated. Vercel built a multi-agent software factory for its AI SDK (over 20 million weekly npm downloads) that now authors 25-35% of merged PRs and closes 70-80% of issues. Astro adopted similar auto-triage automation, Fred Schott created the Flue framework with automatic PR-to-issue conversion, and tldraw automatically closes external PRs.
32
60
NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness
NeoHorse-1 introduces agentic post-training with intelligent routing that lifts agent benchmark scores at 4B and 9B scales, prototyping recursive self-improvement.
NeoHorse-1 is a family of agent-native models trained through agentic post-training: routing-harness logs (predicted capability demand, service tier, interaction) become structurally validated training data organized into a three-stage SFT curriculum plus routing-guided on-policy distillation. Capability-guided allocation converts evaluation feedback into the next training mixture, closing an evaluation-selection-update loop. Post-training raises the macro-average from 58.94 to 64.87 at 4B and from 65.60 to 69.04 at 9B across eleven agent, tool-use, coding, and instruction-following benchmarks. The authors position it as a prototype of harness-mediated recursive self-improvement.
42
Attackers Exploit Critical Langflow and Rails Flaws in Credential
VulnCheck reports active exploitation of critical Langflow CVE-2026-0768 and Rails CVE-2026-66066 for credential harvesting, with detections rising to 360.
VulnCheck observed active exploitation of CVE-2026-0768 (CVSS 9.8) in Langflow and CVE-2026-66066 'KindaRails2Shell' (CVSS 9.5) in Ruby on Rails, with detections rising from 50 on August 30, 2026 to 360 by September 1. The Rails flaw allows unauthenticated arbitrary file reads, leaking secret_key_base, Rails master key, database passwords, cloud credentials and API tokens, ultimately enabling RCE; the patch still leaves the variation-key Marshal deserialization RCE gadget functional. Observed chains include a Python credential harvester with SimpleHelp remote access via CVE-2026-5027, and weaponization of CVE-2025-3248 to enlist hosts into an XMR mining botnet after disabling auditd. More than 7,100 exposed vulnerable Ruby on Rails instances and over 15,000 successful exploitation attempts across three Langflow flaws were recorded.
88
Harnessing LLMs for Automating BOLA Detection
Unit 42's BOLABuster methodology uses LLMs to automate detection of broken object-level authorization vulnerabilities, uncovering flaws in Grafana, Harbor, and Easy!Appointments.
Palo Alto Unit 42 details BOLABuster, a methodology combining large language models with heuristics to automate detection of broken object-level authorization (BOLA) flaws, which traditional fuzzing and static analysis struggle to find. The approach uses LLM reasoning to understand application logic, map endpoint dependency relationships, and generate and interpret test cases. It found CVE-2024-1313 in Grafana, CVE-2024-22278 in Harbor, and 15 CVEs in Easy!Appointments. The team is continuing to hunt for BOLAs in open-source and internal projects.
45
60
TokenRhythm/NeoHorse-1-4B — new model trending #30 on Hugging Face
TokenRhythm releases NeoHorse-1-4B, an Apache-2.0 agentic fine-tune of Qwen3.5-4B claiming +5.93 benchmark macro-average gain.
NeoHorse-1-4B is a roughly 4B-parameter text-only causal language model post-trained by TokenRhythm from Qwen/Qwen3.5-4B for agent harnesses, tool use, coding, and instruction following. It applies routing-guided curriculum SFT and routing-guided on-policy distillation over execution trajectories as an early prototype toward recursive self-improvement (RSI). The release reports a 64.87 macro average across ten benchmarks versus 58.94 for Qwen3.5-4B (+5.93) and is distributed under Apache-2.0, trending #30 on Hugging Face.
28
Risky Bulletin: Dutch intel services to get extensive new powers
Netherlands proposed a bill granting AIVD and MIVD expanded warrantless tapping, faster hacking powers, and forced data disclosure, citing Russia, China, and Iran threats.
The Dutch government introduced a bill greatly expanding surveillance powers of intelligence agencies AIVD and MIVD, allowing up to one year of tapping without pre-approval and simplified hacking operations against 'foreign adversaries'. Agencies could compel Dutch companies or citizens to provide data under threat of charges, share data with the private sector, and oversight bodies would merge into a new CTT board. The bill follows similar overhauls in Ireland, Germany, and France after Russia's invasion of Ukraine. The newsletter also reports Moonwell hacked for $8.7M, a Cosmos EVM bug exploited for ~$3M, ShinyHunters listing McKesson with claimed hundreds of millions of records, and a pro-Kremlin DDoS claim against Norway's government network.
42
The Frontier AEO Tracker: What Astra Chooses (and every other frontier model, and what you can do about it)
Latent Space launches an AEO tracker scoring 7 frontier models' product recommendations across 161 categories, revealing generational bias flips.
Latent Space built a tracker measuring Answer Engine Optimization by running 6 prompt variations across 7 frontier models with search enabled over 161 product categories, scoring first choices, alternatives, mentions, and anti-recommendations. It found 28 categories with a universally dominant primary choice and observed soft biases, such as models favoring their own lab's coding agents. Analysis of Anthropic's Sol→Astra and Opus→Fable generations showed newer models consulting fewer sources and being less likely to change answers when questions are paraphrased.
35
τ^τ-Bench: An Environment for End-To-End, Realistic Agent Construction
New τ^τ-bench tasks coding agents with building deployable customer-service agents; best config, Claude Opus 5, passes only 23.9% of simulations.
Researchers introduce τ^τ-bench, an end-to-end benchmark where a developer agent must build a complete customer-service agent from real business records, a client with requirements, a production API, an inherited codebase, and cost/model limits, then is scored by deploying it against held-out simulated users. Across 53 tasks in four domains, the strongest configuration, Claude Opus 5 under Claude Code, passes just 23.9% of evaluation simulations versus an 82.2% expert-authored reference ceiling. Failure modes mirror those of human developers: shallow queries instead of deep record comprehension, almost no client communication, and shipping the first architecture that runs rather than experimenting.
50
Generative Marketing Mix Modeling: A Causal Inference Framework Linking GEO and GEM to Business Impact
Paper proposes Generative Marketing Mix Modeling to causally estimate Generative Engine Optimization and Marketing effects on business outcomes.
The authors develop GMMM, a causal inference framework for measuring how often users see and notice a firm's name in generated answers, which standard marketing data ignore. For GEO it combines repeated generated answers with question counts, shares of generative-system usage and notice probabilities; for GEM it uses sponsored placement records with notice probabilities. The framework compares expected business responses under alternative treatment sequences, establishes identification conditions, and is evaluated on simulated product-recommendation answers in English and Japanese.
20
Same Trajectory, Contradictory Rewards (ROBORMBENCH): Paraphrase Fragility in Vision Language Reward Models
New ROBORMBENCH benchmark shows vision-language reward models can flip robot success/failure judgments when goal instructions are paraphrased.
The authors show that paraphrasing the instruction alone can substantially change progress scores from VLM reward models, even flipping identical robot trajectories between failure and success. ROBORMBENCH comprises 2,390 real-robot trajectories with ground-truth progress labels and 21,673 verified paraphrases covering lexical, syntactic, and action-goal rewrites. Instability is widespread across proprietary and open-source VLMs, grows with more divergent rewrites, and is not reliably reduced by scale or explicit reasoning, while trajectory-grounded dedicated reward models are markedly more stable.
35
Show HN: LLM Attention Visualization
A developer released a browser-based tool that visualizes which past tokens influence each LLM output token using aggregated, value-weighted attention scores.
A Show HN project presents a React application built on Transformers.js that renders per-token attention influence by aggregating attention weights scaled by value-vector magnitudes across all attention heads and layers. To expose internal tensors, the author instrumented the ONNX computation graph, hosted a modified model on Hugging Face, and pre-generated prompts to avoid long model downloads in the browser. Demos with a 600-million-parameter model show how verbatim copying draws heavily on source tokens and how single outputs blend information from multiple phrases.
28
Show HN: Ordewell – turn one goal into an ordered plan of coding-agent tasks
Ordewell is an open-source tool that decomposes a single goal into an ordered plan of coding-agent tasks, posted on Hacker News.
Ordewell, shared as a Show HN project on GitHub, converts one high-level goal into an ordered plan of tasks for coding agents to execute. The post received 40 points and 29 comments on Hacker News. It focuses on task planning and orchestration for autonomous coding agents.
15
Quoting Rick Brewster
Paint.NET added a clean-room Direct2D rewrite for WINE, largely written by Anthropic's Claude and described as unreviewed 'vibe coded' code.
Rick Brewster says Paint.NET now ships a from-scratch, reverse-engineered Direct2D implementation (PaintDotNet.Windows.Direct2D1.Managed.dll) used under WINE via a /wine flag, since Direct2D was never completed well enough there. He credits the Claude coding assistant with writing most of the code, calling it largely 'vibe coded' and not thoroughly reviewed. Simon Willison shared the quote as an example of shipping AI-assisted systems code in production software.
15
Benchmark Radar: A Living Database and Search Engine for AI Benchmarks and Evaluation
Benchmark Radar provides a living searchable database of 1,283 AI benchmark records and 12,916 score observations drawn from 37 daily discovery sources.
Benchmark Radar combines daily discovery of benchmark papers, repositories, datasets, and releases from 13 direct connectors and 24 first-party feeds into a searchable catalog with model card mentions and score histories. The catalog contains 1,283 source records drawn from 4 benchmark catalogs plus 12,916 numeric observations on 790 records. The release includes a web dashboard with leaderboard, Pareto frontier of score versus usage, saturation and trend views, daily feeds, a CLI, and reproducible analysis. The paper audits the full catalog and examines benchmark saturation and limits of score comparisons.
25
From Parameters to Answers: How LLMs Retrieve and Use Their Internal Knowledge
Interpretability study traces how Qwen, Llama, and Gemma route query information and internal knowledge across layers when answering questions.
Researchers used layerwise interventions on hidden states to separate query-routing signals from target knowledge in language models, testing Qwen, Llama, and Gemma on country-continent questions with varied answer types. A pair-conditioned request direction strengthens before interventions alter downstream knowledge, opening a causal window while answer-supporting content is still forming. Trajectories differ by model: Gemma shows a partially overlapping mid-layer routing profile, while Llama has no sustained routing-effect window under the same gates.
25
27.5KB language-agnostic WebGPU syntax highlighter
A developer released gpu-lexer, a 27.5KB language-agnostic syntax highlighter that uses a tiny WebGPU model to label code tokens in the browser.
gpu-lexer splits source into words, whitespace, and symbols, then a small WebGPU model uses local and whole-file context to assign nine token classes, working on languages never seen in training. On held-out files, 12.57% of token labels differ from Shiki, though this measures agreement with Shiki rather than objective correctness. In benchmarks against Shiki 4.4.3, Prism.js, Highlight.js, Sugar High, and Starry Night, it highlighted 10 concatenated copies of three.min.js (5.56M characters) about 10x faster on an Apple M4 Pro in Chrome 152. The author frames it as an experiment, not a grammar-equivalent highlighter.
15
Augustinian BabyLM: What Ostensive Definition Can and Cannot Teach a Small Language Model
Study shows visually grounded token embeddings in a small masked LM persist through training and improve object-property knowledge, but escape standard BabyLM benchmarks.
The paper implements ostensive definition for a small DeBERTa masked language model trained on 10M words, seeding visually grounded tokens with embeddings derived from labeled image regions before training. Visual initialization leaves a persistent, seed-replicated advantage on object-property knowledge (COMPS) and a corpus-tailored Visual-Property Swap benchmark covering color, material, size, and shape, but has no effect on most BabyLM grammar benchmarks. Synthetic grounding of previously unseeded words causally transfers the advantage to exactly those words.
20
A Three-Layer Caching Architecture for Low-Latency LLM Web Search on Commodity CPU Hardware
OreoLook's three-layer Redis caching architecture cuts redundant LLM calls and embedding work for CPU-hosted web-search answer generation.
The paper describes a three-layer caching architecture for OreoLook (formerly lixSearch), an open-source LLM answer engine: a Redis session context window with Huffman-compressed disk overflow, a semantic query cache matching rephrasings via embedding cosine similarity, and a URL embedding cache deduplicating embedding computations. Deployed on a single 8-vCPU Intel Cascade Lake server with 30 Hypercorn workers across three containerized replicas, it achieved an 89.3% aggregate Redis keyspace hit rate, 0.1 ms read latency, and 1.38 MB memory overhead. An LRU eviction daemon migrates idle sessions to disk and rehydrates them for resumption hours or days later.
34
Ask HN: How do you manage skills files?
A Hacker News thread debates whether agent skill files are worth managing, citing 2–4x output-token reductions on flagship models in one company's testing.
Commenters argue skills are stored prompts that help less-technical users compensate for weak prompting, while one participant reports company testing found skills reduce flagship-model output tokens by roughly 2–4x, a gap growing with newer models. Others note skills can bundle reusable scripts and inline commands for deterministic context building, and that harnesses now execute backticked commands before the agent sees the skill. Some argue improving model capability makes downloadable skills redundant.
15
Ask HN: Anyone still coding like 2021? Where do you work?
Hacker News users debate coding without LLMs, with one developer fired for refusing AI tools and others describing daily hand-coding practice to counter skill atrophy.
An Ask HN thread collects experiences of developers who still write code without LLM assistance. One contributor says he was fired for political reasons after refusing to use LLMs despite adequate stated performance, and observes fewer job ads now require LLM use. Others describe starting each day with a LeetCode problem or 30-60 minutes of hand-coding to stay sharp, contractual bans on AI-generated code for a government-adjacent embedded product over unresolved copyright issues, and inconsistent corporate policies where ChatGPT or Codex use flip-flops between allowed and blocked while a CIO mandates 70-80% AI-generated code next year.
12
Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities
Google open-sourced Mantis, an Apache-2.0 modular skills toolkit that lets AI coding agents find, reproduce, and patch vulnerabilities with sandboxed verification.
Google released Mantis on GitHub under Apache 2.0 as a stack-agnostic set of slash-command skills that chain through the full vulnerability lifecycle: mining version history, building threat models, filtering findings, reproducing bugs in gVisor or network-disabled VMs, assembling exploit chains, patching, and scoring residual risk from 1 to 10. It runs with Gemini CLI, Antigravity CLI, the Google ADK, or comparable agent frameworks, and a supervisor skill (/mantis-meta-agent) can drive the whole loop. Google says the design targets the sub-7 percent true-positive rate of naive AI code scanning, and that its hierarchical summary tree cuts token overhead by over 85 percent. The toolkit is deployable for local and internal evaluation but not yet recommended for production.
55
Distill Globally, Adapt Locally: Reasoning Distillation and Product-Type Test-Time Training for Scalable Trade-Up Recommendation
A distillation framework compresses LLM reasoning into a 15.5M-parameter trade-up recommendation model reaching AUC 0.941 with product-type test-time training.
The paper targets trade-up recommendation, which identifies higher-quality alternatives that preserve customer purchase intent. A retrieval-augmented few-shot LLM teacher generates labels and rationales that supervise a compact embedding-pair classifier; at inference the 15.5M-parameter student uses only two precomputed 768-dimensional embeddings with no LLM calls. On 8,352 annotated pairs, label-only training scored AUC 0.912, reasoning distillation reached 0.924, and product-type test-time training lifted it to 0.941 with average precision 0.940. The distilled student is roughly 5,000x faster and 10,000x cheaper than direct LLM inference on a 100K-pair proxy catalog.
25
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Hugging Face published a tutorial on training and finetuning multi-vector embedding models using the Sentence Transformers library.
Hugging Face's blog walks through training and finetuning multi-vector embedding models with Sentence Transformers. Multi-vector approaches store multiple vectors per document to support late-interaction retrieval. The post is a practical guide for developers building retrieval pipelines with the library.
20
The Router Within: Eliciting Native Skill Routing from a Frozen LLM
Gavel reads native skill-routing signals from a frozen LLM's forward passes with two linear maps, beating retrieve-and-rerank pipelines by up to 21.9 points on Qwen3-32B.
Gavel (Glance And Verdict from a frozen LLM) elicits skill routing from a frozen agent LLM using two trained linear maps that read mid-layer states, keeping all skill text out of context. A glance step scores the full library against compact per-skill banks built in one forward pass at installation; a verdict step resumes shortlisted skills' forward passes and fuses likelihood and yes/no judgments as a product of experts. It transfers zero-shot to three public benchmarks plus SkillTraj, a new benchmark of 372 simulated agent trajectories. On Qwen3-32B it beats progressive disclosure and retrieve-and-rerank pipelines adding 1.2B–16B external parameters by up to 13.4 points on written tasks and 21.9 when skills are needed mid-rollout.
40
Graph Machine: Towards Better Pretraining via Edges
Researchers propose Graph Machine, an O(n)-state sparse architecture that replaces 75% of Qwen3-0.6B dense layers with only slight loss change.
The paper introduces the Graph Machine (GM), an architecture that maintains an O(n)-sized state accessed through sparse, dynamic routing via pointer-like edges updated differentiably by a referral mechanism resembling pointer chasing. The authors replaced 75% of dense Transformer layers in Qwen3-0.6B with GM sparse layers and pretrained from scratch on 15.7B tokens. Retrieving 2 of 4,096 tokens per KV head in each sparse layer degrades loss only slightly, while retrieving 4 marginally improves loss over the dense baseline.
30
MetroLLM-Bench: Evaluating Language Models as Transit Kiosk Runtimes
MetroLLM-Bench is a 955-case benchmark testing language models as transit kiosk tool-calling runtimes across six real metro systems.
The benchmark covers 37-414-station metro systems and eleven task categories including routing, fare calculation, disruptions, accessibility, and adversarial input, with 14 deterministic and 8 semantic scoring components. Of 26 models from six vendors, a PEFT-tuned 4B Qwen 3.5 student scored 91.3 on Tier 1, exceeding GPT-5.6 (90.6/90.0), while Muse Glimmer 30B led the composite ranking. A deterministic rule-based baseline reached 84.6, and PEFT gains over base models shrank from +7.03 points at 2B to -0.91 at 27B.
30
Everything in Moderation: Per-Domain Coverage Optima and Alignment-Resistant Domain Gaps in Multi-Domain Mid-Training
Controlled mid-training experiments on Qwen3-8B-Base find each domain has a 10-40% coverage optimum and domain gaps survive alignment SFT.
Using Qwen3-8B-Base (with a 4B replication) across five semantically rule-disjoint KOR-Bench domains, the authors train 30 data allocations spanning the five-domain simplex at five seeds each. All five domains show interior optima in the moderate 10-40% coverage band, and domain gaps persist after a fixed-budget compensatory SFT pass, which raises 116/120 cells yet bridges 0/240 pairs at a 5% threshold. Zero coverage collapses mid-training-only accuracy, though a FineWeb-Edu-only control shows the collapse is partly generic drift. The results argue mid-training data composition requires principled design rather than reliance on later alignment.
30
Can Foundation Models Moderate Online Content? Evaluating Instruction- vs. Example-Driven Policy Operationalization
ModerationBench shows foundation models can nearly triple Bluesky's moderation F1 (0.60 vs 0.22), with instruction- and example-driven guidance performing comparably.
Researchers built ModerationBench, a new benchmark of 4,000 manually annotated in-the-wild posts from Bluesky, to test whether foundation models can reliably operationalize content moderation policies. They systematically compare instruction-driven guidance (reasoning from policy precepts) with example-driven guidance (generalizing from precedents) for Vision-Language Models. Both paradigms achieve comparable peak effectiveness, and foundation models nearly triple the F1 of Bluesky's deployed moderation system on Random Posts (0.60 vs 0.22).
28
Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
Hugging Face details building and using multi-vector late-interaction embedding models with Sentence Transformers for retrieval workloads.
Hugging Face published a guide on multi-vector, late-interaction embedding models (ColBERT-style) supported through Sentence Transformers. The post covers how practitioners can build and use these models for retrieval and RAG pipelines. It is a developer tooling and technique write-up, not a security advisory.
42
Technical Manual for a Toolkit for Measuring Contextual Individuation in Transformer Language Models
An open methodology toolkit measures whether transformer language models contextualize fixed word forms across domains using bridge forms and layer-wise silhouette analysis.
The manual documents an open toolkit built around 'bridge forms' - identical written words recurring across two or more subject domains with a different sense in each - to test whether transformer language models individuate word occurrences by context beyond the embedding layer. It covers declarative specification of bridge forms, Wikipedia corpus acquisition, occurrence localization, layer-wise representation extraction, domain-pairwise silhouette measurement, and visualization, justifying each choice against failure modes such as sense contamination and subword-tokenization misalignment. It is a methodological and implementation reference and reports no empirical results.
14
Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context Learning
Researchers show IICL structural jailbreaks generalize to Google Gemini, lifting attack success to 80-100% on harm and financial benchmarks; non-English prompts attenuate it.
The study red-teams two Google Gemini models with Involuntary In-Context Learning (IICL), a structural jailbreak reframing harmful requests as the final cell of a data-labeling task. IICL lifts attack success from at most 6.7% to 80-90% on HarmBench and 97-100% on financial abuse (FinProof), an order of magnitude above prior results on OpenAI's GPT-5.4. Against a compounding hypothesis, forcing IICL output into Spanish, Hindi, or Arabic attenuates the attack in 11 of 12 conditions, attributed to a 'relevance curse' producing lower-quality harmful content in lower-resource languages. Findings replicate under an independent non-Google judge (Cohen's kappa 0.86 over 377 paired verdicts).
58