ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “GPT”

40 stories in the last 30d

Searchkeywordsmart Sortlatesthotimportancerelevance
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllAI safety & security 55AI industry 46AI research 41Model release 40AI tools & infra 14AI policy 5Research 5Threat actor 5Vulnerability 2Exploit / PoC 2Phishing & fraud 2Tools 2Advisory 2Data breach 1
SeverityAllcritical 2high 12medium 36low 0info 174

Show HN: Pelican-bicycle alternatives (updated for 2026)

Hobbyist benchmark re-runs the pelican-bicycle SVG test on six 2026 frontier models, comparing generation time and API cost per image.

A Show HN post re-runs the classic pelican-bicycle and similar SVG generation tests across six 2026 models: GPT-6 Astra, Claude Fable 5.1, Gemini 3.8 Flash, DeepSeek V4 Pro, Qwen3.8 Max, and Fugu Ultra v2, recording wall-clock time and cost. It also lists 2025 baseline runs with ten models including Claude Sonnet 4.5, GPT-5.2 Pro, and Qwen3-VL-235B-A22B-Thinking. DeepSeek V4 Pro is consistently cheapest ($0.04-$0.10) while Qwen3.8 Max is slowest, taking up to roughly 17 minutes per generation.

Hacker News · security · 3d agoAI researchHN 24↑ · 4 comments

Real-SWE: Benchmarking AI models on private, real-world, enterprise codebases

Real-SWE benchmark tests coding agents on licensed private enterprise codebases; top model Fable 5.1 resolves only 38.8% of tasks.

Real-SWE is a new benchmark evaluating frontier AI coding agents on tasks drawn from private production codebases licensed from real companies, spanning billing, tax calculation, and cross-service migrations. Fable 5.1 with Claude Code leads at 38.8% resolution rate (pass@1 over eight runs), followed by GPT-6 Astra Codex CLI at 33.8% and Gemini 3.8 Flash Gemini CLI at 31.2%. Tasks use native harnesses and realistic tooling including Docker, Kubernetes, PostgreSQL, Redis, and Linear; median reference solutions edit 11 files versus 6 for DeepSWE and FrontierCode.

Hacker News · AI · 4d agoAI researchHN 45↑ · 31 comments1

GPT-5.6 Luna vs. GPT-6 Astra: Is a $1.20 Model Good Enough for Code Review?

Entelligence benchmarks GPT-5.6 Luna ($1.20/M output) against GPT-6 Astra for code review: Luna found 69 verified bugs at 3.6% of Astra's cost.

Entelligence compared GPT-5.6 Luna ($0.20/$1.20 per million tokens) against GPT-6 Astra ($10/$50) on 50 benchmark pull requests from Cal.com, Sentry, Discourse, Keycloak, and Grafana. Astra verified 92 bugs versus Luna's 69, with precision of 96% versus 74%, and Astra caught 19 of 24 security bugs while Luna found only 9. Luna cost $0.20 total versus Astra's $5.66 and reviewed faster at 23 seconds versus 36, with the widest quality gap on Keycloak authentication and permission logic (6 vs 14 verified bugs). Running both models would find 82% of the 143 verified bugs for $5.86 total.

Hacker News · AI · 2d agoAI researchHN 44↑ · 45 comments

GPT-6 Astra pilots a surveillance drone and runs a business on its own

GPT-6 Astra outperforms Claude Fable 5.1 on Vending-Bench and becomes the first model to beat the human-AI baseline on all five Drone-Bench subtasks.

Andon Labs tested OpenAI's GPT-6 Astra on two agent benchmarks: Vending-Bench 2, where Astra averaged $15,515 running a simulated vending-machine business versus Claude Fable 5.1's $5,422, and Drone-Bench, where models write code for a DJI Tello EDU drone to navigate an office and follow a specific person. Astra is the first model whose best submissions beat the human-AI baseline on all five Drone-Bench subtasks, using a COLMAP and DA3 pipeline with depth filtering for 3D reconstruction. Reliability remains limited, as an average Astra run has only a 2.8 percent chance of passing all five drone steps sequentially. In Vending-Bench Arena, Astra refused a price-fixing proposal from GLM-5.3, while Claude Fable 5.1 participated in an arrangement Andon Labs classified as illegal price-fixing.

The Decoder · 4d agoAI research

In-Context Robot Learning with VLM Agentsnew

GPT-Policy uses commercial VLM agents like GPT-6 Astra for in-context robot learning, translating demonstrations and feedback into verified robot actions without gradient updates.

The paper (arXiv 2609.19138) presents GPT-Policy, a framework combining a context compiler that preserves task-relevant visual transitions, a VLM such as GPT-6 Astra that proposes robot-tool actions, and a constrained controller that verifies and executes each action. Real-robot trials show human video demonstrations improve task completion even without robot action labels, while aligned action references yield further gains on contact-sensitive tasks. Evaluation covers success and efficiency metrics, matched model comparisons, and controlled context ablations.

Hugging Face daily papers · 1d agoAI research

The Pelican comparison grid for Astra is pretty interesting

Simon Willison's pelican SVG comparison shows GPT-6 Astra producing markedly better images than GPT-5.6 Sol, Terra, and Luna across reasoning levels.

Willison generated pelicans-riding-bicycles SVGs with newly accessed GPT-6 Astra at low through max reasoning levels and rendered them in a comparison grid against GPT-5.6 Sol, Terra, and Luna. Astra's outputs were markedly more coherent, while even the best GPT-5.6-Sol images remained largely abstract shapes. Astra does not support a reasoning=none setting, so all comparisons involved reasoning-enabled runs.

Simon Willison · 12d agoAI research

Can LLMs Engineer Their Own Agent Harness? ByteDance Seed’s HarnessDev Says Only 34 of 64 Changes Generalize

ByteDance Seed's HarnessDev benchmark finds LLM-built agent harnesses trail human engineering on code and search, with only 34 of 64 revisions generalizing.

Researchers from ByteDance Seed, SUTD, Georgia Tech, M-A-P, and TokenWave.AI introduce HarnessDev, a benchmark that evaluates the runnable agent harness an LLM writes rather than its answers, using Creation and Evolution stages across SWE-bench Pro, Terminal-Bench 2.1, MLE-bench, EQ-Bench3, and BrowseComp (2,207 instances). Six creator models including Opus 4.8, GPT-5.5, Gemini 3.1 Pro, DeepSeek V4 Pro, Qwen 3.7 Max, and Seed 2.0 Pro were tested; Opus 4.8 posted the best average of 67.8 versus an 86.2 human-engineered reference. Self-built harnesses beat references on writing and ML experimentation but lag badly on code and search, and quality proved executor-specific: Opus 4.8 fell from 69.3 to 33.0 on SWE-bench Pro when the executor was switched to Gemini. Evolution gains were small and noisy: of 64 adjacent changes, feedback and held-out scores agreed only 34 times (53.1%), and much generated state and memory code never executed.

MarkTechPostupdated · 20h agofirst · 5d agoAI research 20 sources

Unsolved Problem by Fields Medalist Breached by Two High School Students

Two high school students used Claude Opus 5 and GPT-5.6 Sol to help solve an open Lorentzian polynomials problem, posting a 75-page arXiv proof.

Aayush Bathija and Prince Rohatgi of Oak Park High School, mentored by UCLA postdoc Daniel Soskin, published the 75-page paper 'Bounded Ratios for Lorentzian Polynomials' (arXiv 2609.05341), solving an open problem in Fields Medalist June Huh's Lorentzian polynomial theory. The main structural theorem extends bounded coefficient-ratio characterization from quadratic to arbitrary-degree polynomials via discrete convexity conditions. The students used Claude Opus 5 and GPT-5.6 Sol for exploration and proof ideas but independently verified all arguments; the result follows an open letter from 25 Fields Medalists voicing concerns about AI's impact on mathematical rigor.

Hacker News · security · 3d agoAI researchHN 22↑ · 19 comments

Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes

Independent developer Hugo Vergnes trained a 3.8B-parameter Llama-style model to 0.384 CORE on 65B tokens for $998 in 43 hours on rented B200s.

Hugo Vergnes trained little-lm, a 3.848B-parameter decoder-only LLM, on 65.3B tokens in 43 hours for $998 using rented NVIDIA B200s, scoring 0.384 on the CORE benchmark and beating nanochat d32 (0.310) at similar cost. The Llama-style architecture uses RMSNorm, RoPE, GQA with 24 query and 8 KV heads, relu-squared MLPs, QK-norm, and ResFormer-style value embeddings that account for 19% of parameters. Key wins included the Muon optimizer for matrix parameters, a trapezoidal learning-rate schedule with linear cooldown, FP8 training plus vocabulary padding for roughly 33% throughput gains, and the ClimMix dataset over FineWeb-Edu. The project, inspired by Karpathy's nanochat, was built as a config-driven YAML framework for small LLM training.

Hacker News · security · 7d agoAI researchHN 28↑ · 2 comments1

Right Tool, Right Job: Native-Language Evaluation, Tokenizer Sensitivity, and Methodological Findings from a French-Only BabyLM

French BabyLM entry METRON-FR (125M GPT-2, 92.47M words) shows tokenizer artifacts dominate child-scale zero-shot evaluation; proposes standard diagnostics.

METRON-FR is a 125M-parameter GPT-2 pretrained on 92.47M French words, submitted to the BabyLM 2026 Strict track, scoring 85.97% on the native Quebec-French QFrBLiMP benchmark and 62.80% on the BabyLM-weighted leaderboard. A cross-lingual GLUE protocol combining French task-data translation with rank-16 LoRA shows relational tasks gain while world-knowledge tasks regress. Bilingual Lexicon Induction reaches p@1 of 68.84%, 18x above chance, and ablations show single-token zero-shot scoring is dominated by tokenizer and template artifacts at child scale.

arXiv cs.AI / cs.LG / cs.CL · 1d agoAI research

Before You Poll with LLMs: A Deliberative Diagnostic Framework

Deliberative diagnostic shows all five tested frontier LLMs misrepresent human belief shifts after arguments, with GPT-5.1 reversing on outgroup questions.

The Deliberative Polling Diagnostic Framework compares human and LLM persona belief shifts after identical informational interventions, using data from America in One Room (526 personas, 72 questions). All five frontier models tested failed uniquely: GPT-5.1 exhibited partisan reversal (80% on outgroup vs 26% on policy questions), Gemini 2.0 Flash, Claude Sonnet 4.5 and Llama 3.3 70B overshot at 5-7x human magnitude, and DeepSeek V3 showed near-zero change (rigidity). The authors term the underlying signature 'self-sycophancy', conformity to the model's internal persona stereotype rather than reasoning from provided information.

arXiv cs.AI / cs.LG / cs.CL · 2d agoAI research

Emergency Department Revisit Quality Review Screening: Exploring Human Decision-Making and Artificial Intelligence Support

A retrospective study found GPT-4 over-flagged emergency department revisit cases while an LLM knowledge-graph screener achieved 83-100% positive predictive value.

In an exploratory retrospective study of 99 emergency department diagnosis pairs from a multihospital health system, clinicians and GPT-4 independently judged whether revisit pairs warranted further assessment. GPT-4 responses correlated poorly with clinicians, flagging 94% of pairs for follow-up, 4.4-13.3 times more than clinicians, though prompt engineering was minimal. An algorithm leveraging an LLM-populated knowledge graph (KGA) achieved 83-100% positive predictive value against at least one clinician rater, suggesting LLM-based screening could broaden revisit quality review without substantially increasing reviewer workload.

arXiv cs.AI / cs.LG / cs.CL · 7d agoAI research1

How Model Growth, Recursion, and Boundary Operators Influence Scaling Exponents

Researchers show model growth via looped transformers improves scaling exponents; a 7.4B architecture matches GPT-3 13B with roughly 20x less compute.

The paper shows that architectural interventions, contrary to conventional wisdom, can modify pre-training scaling exponents and yield exponential performance gains with compute. Looped transformers with increasing loop counts provide a model growth mechanism; a 7.4B model-growth architecture matches GPT-3 13B on CORE with roughly 20x less compute, with efficiency gains that increase with scale. A boundary operator that normalizes and injects an earlier block also improves compute efficiency, and in data-constrained multi-epoch settings increasing loops with scale is compute-optimal.

arXiv cs.AI / cs.LG / cs.CL · 20h agoAI research

Can Skills Learned in Games Transfer to Real-World Work?

Good Start Labs trains models in strategy games like 1830 and Diplomacy, showing terminal-agent training transfers to financial research benchmarks.

Good Start Labs, spun out of Every with $3.6M from General Catalyst and Inovia, trains AI models in verifiable strategy games. A 30B model trained as a multi-turn terminal agent in 1830: The Game of Railroads and Robber Barons improved Finance-Agent benchmark performance, while single-turn QA training did not transfer. The founders also co-authored COS-PLAY, a paper on co-evolving LLM decision and skill-bank agents for long-horizon tasks.

Latent Space · 1d agoAI research

RSIAgent: Autonomous Exploration for Recursive Self-improvement in New Environments

RSIAgent, a training-free multi-agent framework, builds reusable environment memory enabling Kimi-K3 and GLM-5.3 to beat GPT-6.

RSIAgent is a training-free framework for recursive self-improvement through autonomous memory construction, coordinating curriculum, actor, and verifier agents. It uses broad-then-deep exploration to capture environment structures, hidden constraints, and causal dependencies, and freezes the resulting memory for direct reuse without parameter updates. On OSWorld-v2 and Agent's Last Exam it substantially improves strong open-source models, enabling Kimi-K3 and GLM-5.3 to outperform frontier closed-source models including GPT-6.

Hugging Face daily papers · 3d agoAI research2

Realtime-Venus: A full-duplex interaction system with asynchronous delegation

Realtime-Venus introduces two 9B full-duplex interaction models (Omni and Audio) that outperform Gemini 3.1 Live and GPT-4o on continuation metrics.

Realtime-Venus is a proactive full-duplex interaction system built on two separately trained 9B models: Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for spoken interaction. A dual-loop runtime lets foreground interaction continue while Realtime-Venus-Harness asynchronously executes background reasoning and tool tasks. Realtime-Venus-Omni leads on six of eight video benchmarks, including StreamingBench (70.2%), OVO-Bench (64.7%), and Daily-Omni (81.3%), while Realtime-Venus-Audio tops MMAU (78.0%) and MMAU-Pro (63.2%). On Full-Duplex-Bench v1.5, Realtime-Venus-Audio handles 75% of interruptions and exceeds Gemini 3.1 Live and GPT-4o on all three continuation metrics.

Hugging Face daily papers · 5d agoAI research

T1: Terminal Agent Reinforcement Learning for Long-Horizon Tasks

T1, a 122B MoE terminal agent trained with reinforcement learning, reaches 64.0% on Terminal-Bench 2.1, surpassing GPT-5.4 and GLM-5.1 on long-horizon tasks.

T1 is a 122B mixture-of-experts model trained with reinforcement learning to operate a real shell in a cloud sandbox for up to 300+ tool-call turns per task, rewarded by executing each task's own verifier. The recipe combines aggressive warm starts, dense process rewards, TITO construction, and rollout routing replay, cutting the training-to-inference log-probability difference from 0.021 to 0.013 with zero token drift. Training used an out-of-distribution corpus disjoint from Terminal-Bench 2.1. Post-training raised the base model from 43.8% to 64.0% resolved on Terminal-Bench 2.1 and 27.9% on Long-Horizon Terminal Bench.

Hugging Face daily papers · 7d agoAI research1

[AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded

OpenAI-linked accounts claim roughly 10,000 AI agents produced a Navier-Stokes singularity result in 88 hours, pending mathematical verification.

OpenAI-affiliated accounts claim a system of roughly 10,000 agents, trained over about a year with multi-agent reinforcement learning, produced a finite-time singularity result related to the Navier-Stokes Millennium Problem. The claimed 88-hour runtime and 130B-token cost circulate only via social posts, and no preprint, theorem statement, or proof artifact is available. Acceptance by the mathematics community is unresolved, so the claim's epistemic status remains unknown. The roundup also notes Cognition's $48B and Mistral's $24B fundraises, GPT Image 2.5, and Meta's Muse agent relaunch.

Latent Space · 8d agoAI research1

What researchers learned about building an LLM security workflow

Oslo and FFI researchers show structured agentic workflows lift LLM alert-triage accuracy from 0% to about 93% on malicious cases.

Researchers at the University of Oslo and the Norwegian Defence Research Establishment tested GPT-5-mini, Claude 3 Haiku, Qwen3:30B, and Gemma 3:27B on alerts from the AIT Log Data Set V1.1; given only alert descriptions and log summaries, all four models correctly flagged zero percent of true-positive cases involving reconnaissance, brute-force logins, and initial access. Wrapping the same models in a workflow with constrained SQL queries over Suricata logs, an evidence summarizer, and a verdict stage with revision loops raised malicious-case accuracy to an average of 93 percent, with GPT-5-mini identifying every malicious case across 100 runs. The authors flag it as a proof-of-concept on one synthetic scenario and note models skewed conservative on benign alerts, with GPT-5-mini marking every benign case uncertain.

Help Net Security · 24d agoAI research1

Enhancing Accessibility of Medical Texts through Large Language Model-Driven Plain Language Adaptation

Study shows LLMs with Mixture-of-Agents and QLoRA finetuning effectively simplify medical texts into plain language while preserving content.

The paper evaluates Plain Language Adaptation (PLA) using GPT-4o-mini, Gemini-1.5-pro, and LLaMA in zero-shot and few-shot settings. It compares prompting strategies, QLoRA finetuning across models, and integrates Mixture-of-Agents (MoA) techniques for robustness. Results demonstrate LLM-driven PLA makes healthcare texts more comprehensible while retaining essential content.

arXiv cs.AI / cs.LG / cs.CL · 1d agoAI research

Another Blueprint In The Wall: How to Ask Frontier AI Like a Kid?

Six frontier models from OpenAI, Anthropic, xAI, and Google DeepMind converge on one imagined successor architecture when asked under a school-audience framing.

Researchers ran ten independent sessions per model type across six frontier models using a three-stage prompt sequence progressing to a full ASCII backbone architecture. Under school-audience framing, responses repeatedly converged on a shared motif including persistent latent state, adaptive computation, memory, specialist routing, verification, and stopping control, while control runs without the framing produced heterogeneous responses. A GPT-5.6 Sol output closely overlapped an architecture independently sketched by GPT-6 Astra, raising questions about shared design priors or motif propagation between model families. The paper coins 'epistemic jailbreak' for the observed loss of provenance discipline as prompt specificity increases.

Hugging Face daily papersupdated · 4d agofirst · 4d agoAI research 6 sources

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.

Hugging Face daily papers · 8d agoAI research1

What Else Needs Fixing? Exploring Cost-Effective Test-Time Compute for Revision Propagation in Artifacts Generated Through Conversation

A new benchmark shows LLMs reach 68.3-93% accuracy propagating local revisions across conversationally generated artifacts, with parallel-sample selection most cost-effective.

The paper introduces a benchmark for revision propagation: when users request a local change, LLMs must identify dependencies and update all affected parts of an artifact generated through conversation, where context lives in the chat history. Nine revision methods, including sequential reflection and parallel sampling variants, were evaluated on gpt-oss-20b/120b, gpt-5.4-mini, and qwen3.5-9b/27b/122b. Baselines scored 68.3-93% accuracy, and selecting among three parallel samples via LLM-based or medoid selection improved accuracy by 2.2-9.7% as the most cost-effective test-time compute strategy. Code and dataset are released.

Hugging Face daily papers · 14d agoAI research

An AI CAPTCHA solver talked itself out of the right answer

Bern researchers solved rotation CAPTCHAs in 0.006 seconds with classical computer vision, while Gemini 3.1 Pro needed 67 seconds and overruled correct tool answers.

Researchers at Bern University of Applied Sciences built a script using 1970s circle-detection math and signal matching that solved rotation CAPTCHAs in 0.006 seconds, scoring 10/10 on real-world puzzles. Frontier models fared poorly: Gemini 3.1 Pro scored 7/10 taking 67 seconds, while GPT-4o and Grok scored 1/10. When given the script's correct answer as a tool, Gemini overruled it and lost a fifth of its score; models could verbally describe targets, such as identifying a cyan ring, but could not produce accurate click coordinates. The paper also notes these no-JavaScript CAPTCHAs reduce tracking, leaving only shape-matching tasks classical vision solves easily.

Help Net Security · 15d agoAI research1

Agora: Git as Shared Memory for Collective AutoResearch

Agora records multi-agent research as an append-only Git DAG; 13 LLM workers ran nearly 12 days on a weight-transfer problem.

Agora stores every result, hypothesis, and verification as an immutable commit in a Git-stored DAG, with a derived index exposing the frontier and verification status of claims. In a nearly 12-day run, 13 language-model workers with no assigned tasks or central planner published 1,703 contributions on initializing a frozen 119.6M-parameter attention-SSM hybrid from 141 donor models. They improved the evaluator from 3.39 to 1.899 bits per byte, closing 62% of the gap to a trained GPT-2 124M, with 165 independent reproductions posted and none failing.

Hugging Face daily papers · 1d agoAI research

ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks

ProgramDistill is a benchmark evaluating coding agents on reconstructing web app features from reference applications, testing nine frontier agents.

ProgramDistill evaluates coding agents on features discovered through interaction with fully functional reference applications, factorizing apps into features with replayable behaviors verified via gold patches. Its mine-craft-patch pipeline discovered 1,975 replay-verified behaviors across 26 applications and built 4,063 tasks without human intervention. On cumulative full-application reconstruction workflows, GPT-6 Astra achieved 49.2% and Claude Opus 5 28.8% success. In partial reconstruction, success drops from 100% to 64.0% and from 96% to 32% as restoration depth increases from 1 to 8.

Hugging Face daily papers · 1d agoAI research

Agent Harness vs Agent Framework vs MCP: Which Layer Owns the Loop, State, Tools, Permissions, and Recovery

Architecture explainer separates agent harnesses, frameworks, and MCP by which layer owns the loop, state, permissions, and recovery.

The article distinguishes agent harnesses (OpenAI Codex, Claude Agent SDK), which own the execution loop, sandbox, permission model, and recovery; frameworks (LangGraph, OpenAI Agents SDK, Microsoft Agent Framework), which supply composable primitives; and MCP, a stateless JSON-RPC wire protocol governed by the Linux Foundation's Agentic AI Foundation since December 2025. An ownership matrix maps the execution loop, state, tool transport, permissions, recovery, sandboxing, and multi-agent orchestration to each layer. The 2026-07-28 MCP specification made the protocol fully stateless, retiring the initialize handshake and session headers.

MarkTechPost · 2d agoAI research1

Why don't machine learning research agents overfit?

Amazon researchers explain why ML research agents avoid benchmark overfitting, attributing generalization to compressibility of successful strategies.

Amazon Science summarizes the paper "What fits (into few tokens) doesn't overfit: Compression and generalization in ML research agents," which investigates why benchmark hill-climbing loops, whether run by human communities or LLM research agents, do not produce rampant overfitting. The explanation formalizes Occam's razor via a counting argument: successful ML strategies are highly compressible, so short descriptions lack room to memorize benchmark data and must capture real structure. LLM-based agents, being resettable and controllable, allow this hypothesis to be tested empirically.

Hacker News · security · 2d agoAI researchHN 21↑ · 3 comments

PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models

PhysBrain 1.5, an 8B physical foundation model, sets open-source state of the art across 28 embodied understanding benchmarks.

The paper presents PhysBrain 1.5, a unified 8B model for understanding physical environments, generating actions, and predicting future states, built from a vision-language model with joint autoregressive next-token prediction over language, end-effector motion, and dense visual targets. Pre-training uses embodied supervision from human interaction videos, followed by supervised fine-tuning on human demonstrations, robot trajectories, and simulated experience. The model averages 72.5 across 28 embodied benchmarks, setting a new open-source state of the art and performing on par with proprietary GPT-6-Astra and Gemini 3.6 Flash, with best open-source results on 14 benchmarks.

Hugging Face daily papers · 3d agoAI research1

Context Engineering Inside the Harness: 4 Mechanisms That Beat Context Overflow and Goal Loss on Long-Horizon Tasks

Survey of four harness mechanisms—context budgeting, compaction, todo-state, and memory—that keep long-horizon LLM agents on task across 200+ tool calls.

The article details how agent harnesses, not larger context windows, solve context overflow and goal loss on long-horizon tasks, citing Chroma's Context Rot report showing 18 LLMs (GPT-4.1, Claude 4, Gemini 2.5, Qwen3) degrade on long inputs. Concrete implementations include LangChain Deep Agents offloading tool responses over 20,000 tokens to the filesystem and truncating old tool calls at 85% window usage, and Claude Code capping auto memory at 25KB while re-reading the 5 most recently modified files after compaction. OpenAI's Responses API now offers server-side compaction via context_management with a standalone /responses/compact endpoint, which Codex uses for long-running coding tasks. Manus reports a roughly 100:1 input-to-output token ratio per ~50-tool-call task, motivating todo.md state recitation to prevent goal drift.

MarkTechPost · 4d agoAI research2

AI models' written reasoning steps correspond to distinct internal patterns, a new study finds

KAIST and Naver AI Lab researchers show LLM reasoning steps like extraction and computation map to distinct activation patterns, strongest in middle layers.

Researchers at KAIST and Naver AI Lab defined eight recurring reasoning operations, including extraction, decomposition, formula recall, deduction, and computation, and showed they correspond to separable activation patterns in Qwen2.5-7B, Qwen3-8B, and Gemma4-31B on math tasks, with GPT-5 labeling solution segments. The separation peaks in middle layers, holds even when a computation step produces a wrong answer, and goes beyond surface-level token choice. Findings replicated on Llama-3-8B, and classifiers trained on Qwen3-8B transferred to GPQA-Diamond and MATH-500. The authors note that using internal states for error detection or mid-generation steering remains future work.

The Decoder · 5d agoAI research2

Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation

Google Research and partners introduce ToolGrad, a verified tool-chain-first data generation framework reaching 99.8% pass rate and boosting Gemma-3-12B to 83.1 on BFCL.

Researchers from Google, the University of Tokyo, RIKEN AIP, and Tohoku University released ToolGrad, which inverts query-first tool-use data generation by executing and verifying API chains before annotating them with user queries. On the ToolBench database of 16,000+ APIs, ToolGrad raised generation pass rate from 63.8% to 99.8% while increasing tool uses per sample from 2.1 to 3.4 and cutting tool-use steps from 34.3 to 20.0. Fine-tuning Gemma-3 at 1B, 4B, and 12B parameters on the 500-sample ToolGrad-500 dataset lifted ToolGrad-12B to 83.1 on the Berkeley Function Calling Leaderboard, near Gemini 2.5 Pro at 83.2 and ahead of GPT-5 at 74.4. Code is Apache-2.0, with the dataset, PyPI package, and models available on Hugging Face.

MarkTechPost · 6d agoAI research1

Root-Cause Attribution Is a Search Problem: Continual Search for Long-Horizon Agent Failures

Continual Search framework iteratively prompts LLM judges to keep searching agent execution logs, boosting long-horizon failure root-cause attribution accuracy.

The paper frames automated root-cause attribution (RCA) for long-horizon AI agent failures as a search problem, since relevant evidence is sparse and distributed across massive execution traces. The authors propose Continual Search, an iterative framework that nudges an LLM judge across successive turns to keep hunting unresolved diagnostic evidence instead of settling on an early plausible diagnosis. They introduce MegaRCA-Mix, a benchmark of 50 human-annotated failure trials on long-horizon, execution-heavy tasks. On MegaRCA-Mix, Continual Search improves GPT-5.5's F1 from 0.349 to 0.498 (over 40% gain), and lower-tier models can surpass higher-tier counterparts when search is effective.

Hugging Face daily papers · 6d agoAI research1

ExecCritic: Learn to Test, Test to Improve for Coding Agents

ExecCritic separates test generation from patching for coding agents, lifting SWE-bench Verified resolution to 72.6%.

ExecCritic pairs a test-verify-revise scaffold with role-specific reinforcement learning: a Test agent writes repository-native tests and a Repair agent fixes code from execution feedback, both using Qwen-3.5-35B-A3B backbones. Post-trained Qwen agents compose to 72.6% on SWE-bench Verified, an 11.4-point gain over the 61.2% no-test baseline, without stronger-model or oracle feedback at evaluation time. The work shows test quality is the key variable: base-agent tests lowered resolution to 57.3% while GPT-5.6-sol tests raised it to 65.3%.

arXiv cs.AI / cs.LG / cs.CL · 8d agoAI research1

How well do agents use test/verification techniques?

Dan Luu's eval finds coding-agent testing instructions (TDD, formal methods, PBT, skills) mostly fail to beat defaults on Zstd implementation correctness.

The author ran 26 prompt conditions plus 4 skills on a Zstd-in-Rust implementation eval using codex with GPT-5.6, testing TDD, fuzzing, property-based testing, formal methods (Lean 4, TLA+, Verus, Kani, SMT solvers) and community skills. Nothing dramatically outperformed the default no-instruction condition, which did above average; at xhigh effort, fuzzing and PBT conditions did slightly better than formal methods. Pre-registered predictions included TDD underperforming and popular test skills (ECC, Hegel, Trail of Bits) not outperforming. Results are averages of 80 runs per condition plotted against cost.

Hacker News · security · 9d agoAI researchHN 131↑ · 50 comments

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.

Latent Space · 9d agoAI research

SAFIRE: Safety-Critical Benchmark for Fine-grained Fire and Smoke Understanding in Multimodal LLMs

SAFIRE, an 83K-image fire and smoke benchmark, shows open-source multimodal LLMs average only 61.9% accuracy on safety-critical fire reasoning.

SAFIRE is a large-scale benchmark for fire-smoke understanding in multimodal LLMs with 83K captioned images across 20 scenarios and 193K multiple-choice VQA questions spanning 10 evaluation dimensions from perception to higher-order reasoning. Annotations were built via a GPT-5.4-assisted multi-stage pipeline with MLLM majority voting. Ten open-source MLLMs (8B-38B) average 61.9% accuracy, exposing major gaps in safety-critical reasoning. Adapting vision encoders on 7% of the domain data raises fire-scene classification from 20.1% to 64.5%.

arXiv cs.AI / cs.LG / cs.CL · 9d agoAI research

Recreating Minecraft Is Not a Benchmark

Opinion piece argues viral demos like one-prompt Minecraft recreations are overfit 'demo-benchmarks' measuring preparation, not true model capability.

The author argues that fixed, famous demo tasks (Minecraft builds, SVG pelicans) are trivially optimizable by labs each release cycle, so they no longer differentiate model capability. The piece cites Thinking Machines' Inkling Small scoring within a point of its flagship on the Artificial Analysis Intelligence Index with less than a third of the parameters, and beating it on Humanity's Last Exam, GPQA Diamond, and SciCode. The proposed alternative is rotating or holdout evals such as LiveBench, ARC-AGI's private set, and held-back portions of Humanity's Last Exam.

Hacker News · AI · 10d agoAI researchHN 73↑ · 71 comments

When Models Edit Too Much: On the Fidelity of Minimal Code Edits

A 400-task BigCodeBench evaluation shows frontier LLMs widely over-edit code; a preservation instruction cuts excess edits and raises Pass@1 by 2.3 points.

Researchers built an evaluation framework from 400 BigCodeBench problems with injected AST-level corruptions, each with a known minimal patch, to measure over-editing in LLM code repair. Even strong models like GPT-5.5 produce unnecessarily large edits despite high Pass@1. Adding a preservation instruction reduced average excess Levenshtein distance from 0.195 to 0.131, cut added cognitive complexity by 26.6%, and raised Pass@1 by 2.3 points. Reinforcement learning post-training gave the best out-of-domain edit-fidelity trade-off, while supervised fine-tuning overfit to seen corruption patterns.

Hugging Face daily papers · 14d agoAI research1