Tuesday 29 September 2026
433 stories, newest first.
World123
- Florida executes 77-year-old man for 1995 hammer killing
- Trump Ads Funded by US Government Face Complaint Over Alleged Propaganda Law Violations
- French protests spread to over 700 schools as police clash with students
- Seventh man arrested in suspected RAF Fairford terror plot as investigation widens
- South Africa launches high-risk area clean-up after 12 women found dead east of Johannesburg
- France pledges 3,000 teachers as yellow vests and unions prepare to join student protests
- RBA hikes rates to 15-year high of 4.6 per cent as Bullock warns of further increases
- Burnham unveils six major reforms and calls for EU reset in Labour conference speech
- ICE officer who shot Austin man lacked body camera; DOJ charges victim with assault
- Al Jazeera video roundup: housing crisis, Jerusalem attack, Gaza burials, Sudan at UNGA, Oman turtle incident
- Former American Idol contestant Caleb Flynn convicted of murdering wife
- Six Queensland babies die from congenital syphilis since 2024
- Ten Years After Ford's Departure, Geelong Workers Still Feel the Loss
- Supreme Court allows Trump administration to resume deportations to third countries without notice
- Tasmania's retail lease reforms stalled, leaving small business owners without dispute resolution
- Fossil fuel industry enters WA classrooms through Energy Club program, teacher says
- Mick Fanning memoir among October book releases showing a different side of the surfing champion
- ‘People just don’t believe it’: Retired NRL star Ryan Papenhuyzen sets the record straight on the golf course
- Katherine Sabbath joins OzHarvest CEO CookOff to help deliver five million meals
- Palestinians Prepare to Bury Remains of 105 Killed in Gaza Attack
- Treasury lifts NZ deficit outlook but productivity remains the 'elephant in the room'
- Cleary hot favourite for Dally M as NRL weighs how to honour injured star
- Opinion: Why Australia Still Needs Free Dental Care
- Ningaloo Reef Shows Strong Recovery After Regional Disaster
- Anthony Scaramucci’s book warns of Trump return, but offers little on his own White House stint
- RBA hikes rates to 4.6%: House prices more than double since last time rates were this high
- ATO says no rush for property valuations ahead of 2027 capital gains tax change
- Palestinian Authority defends compliance with Australia's statehood conditions as Israel, US allege 'pay for slay' rebrand
- NSW Labor delays preselection for two ministers amid fears of branch-level defeat
- Sister flies to Florida in custody battle for Terrigal woman's sons after alleged murder
- ICAC closes Operation Rosny hearings after 10 weeks, with findings due in 2027
- Apple accused of working with Trump administration to remove ICE-tracking apps, lawmaker says
- Morocco Names Fatima Zahra El Mansouri as First Female Prime Minister
- Saudi Arabia urged to mend ties with UAE to counter Houthi advance in Yemen
- Man charged with murder after two bodies found on NSW property; second person charged in OneFour associate's killing
- RBA rate hike leaves Shane McClymont unsure if he can afford son’s birthday dinner
- Lindsay Clancy's lawyer now argues insufficient evidence in dramatic shift after mistrial
- Coalition Primary Vote Hits Record Low as Australians Back Hardline Immigration Policies
- Trump administration asks Supreme Court to allow denial of gender-affirming care to transgender prisoners
- Burnham says Brexit has done 'more harm than good' in first Labour conference address as PM
- Australian government admits no scientific consensus on social media harms but defends teen ban
- Argentina threatens legal action against UK over Falkland Islands oil exploration
- French hard-right leader Bardella to sue over antisemitic remarks allegations
- Evan Gershkovich describes moment he realised he was 'a pawn in a greater game' after Russian arrest
- Red Cross says settler violence in West Bank 'untenable' after mob attack on Palestinian family
- Burnham vows 'long-term' EU relationship, confirms triple lock end in emotional conference speech
- Global affairs dominate final day of UNGA 2026
- Sudan dominates final day of UNGA as global flashpoints intensify
- Republican senator blocks bipartisan bill to prevent Trump from demolishing Kennedy Center without Congress approval
- Estonia Accuses Russia of Arson Attack on Defence Firm Supplying Ukraine
- Trump airs promotional ad using taxpayer money
- Campaigners protest Palantir's NHS role at Labour conference as contract deadline looms
- Imams' council condemns Melbourne preacher's call to 'take up arms'
- Chalmers defends government's economic record as rate hike looms
- Senate passes Protect College Sports Act in 77 to 22 vote
- Two former Hungarian ministers detained in bribery and misuse of funds probe
- US Ban on Canadian Alcohol, Dairy and Motorcycles Takes Effect as Trade War Continues
- Israeli air strike kills Hamas commander in Gaza as ceasefire strains
- Al Jazeera reports US-Iran talks ongoing but deal unlikely before midterms
- Spain approves new housing measures after eviction protests
- Two Navy aviators eject during landing mishap on USS Dwight D. Eisenhower, four sailors injured
- London facial recognition trial ends with no arrests and one false positive
- Slain police officers honoured at National Police Remembrance Day service in Canberra
- Middle East conflict pushes RBA inflation fight into 'danger zone', analysis warns
- Collie coal workers strike over roster changes, pay as WA accelerates coal phase-out
- Tasmania Devils sign first batch of players, including local prospect who bypassed AFL draft
- NSW music education plan welcomed but teacher shortage concerns raised
- Crossland banned four matches for Cleary tackle, to miss NRL grand final
- Larissa Waters steps down as Greens leader citing decline in health from chronic kidney disease
- Vietnam Confirms Arrest of Two Viet Tan Activists Missing in Cambodia
- Law professor's husband says they 'don't do consent' on podcast
- Forensic experts warn sexual strangulation normalisation is putting young women at risk of brain injury
- Five arrested near RAF Fairford as Rubio alleges foreign actor behind suspected bomb plot
- Burnham told scrapping pension triple lock would be 'electoral suicide' ahead of Labour conference speech
- Canberra man admits 'torture' murder of partner as DPP seeks life sentence
- Burkina Faso opens first gold refinery in push to keep mineral wealth at home
- Burnham signals willingness to review pension triple lock to fund social care
- Trump says he will personally pay for controversial government-funded TV ads
- Foul play suspected after two bodies found on same NSW property two weeks apart
- US-Iran talks continue, but deal unlikely before midterm elections: report
- French PM warns against escalation of school protests after 164 arrested
- Grace Tame tells court she feared for her life after abuser's tweets
- Australia eyes Canadian market as Farrell moves to exploit US trade war
- Malaysia begins repatriating Myanmar asylum seekers as first navy ships depart
- Indian university suspends classes after unverified rape allegation sparks violent protests
- Senior Queensland public servant faces sacking over porn on work computer, fake timesheets
- Johns brothers push for Silverchair reunion to fill NRL grand final entertainment void
- Former ABC News head Peter Manning dies at 81, remembered as 'colossus' of investigative journalism
- Employers warn working holiday visa changes are driving backpackers out of Far North Queensland
- Australia's card surcharge ban begins this week as businesses adjust payment policies
- Geelong defender Jake Kolodjashnij retires months after concussion waiver fine
- Former WA shire deputy chief Sam Curulli granted bail on new fraud and drug charges
- World's youngest known euthanasia case: one-year-old child dies in Netherlands
- Accused killer faked cancer in 1600-word Instagram text, court hears
- RBA raises cash rate to 4.6 per cent, a 15-year high, as tougher rate cycle looms
- Downsizing your home could reduce your age pension, data shows
- Sharks linger near beached whale carcass at City Beach as removal delayed
- Knights No.7 feeling weight of Joey's jersey ahead of grand final against former club
- Comedian Simon Taylor launches sitcom skewering 'Wild West' podcast culture
- From cheesy crowd-pleasers to beautiful bakes: Your favourite recipes from September
- US and China strike $60bn tariff relief deal after state visit
- Seoul summons Ukraine envoy over disclosure of North Korean prisoner transfer
- Mortgage holders urged to switch lenders as RBA rate hits 15-year high
- Brickworks to restart mothballed Cardup plant as WA tackles brick shortage
- Toplace paid $134,000 for surveillance of Strathfield mayor, ICAC told
- Greens MPs to elect new federal leader at Sydney retreat after Waters resignation
- Family and fellow officers pay tribute to Constable Mia Lay at police remembrance service
- Sydney barrister appointed to prosecute federal MP Luke Gosling assault case
- "Working till 67, hoping for the best": Casual workers face retirement confidence gap
- Labor narrows Coalition's Victorian lead as federal polls show One Nation strength
- Developer walks away from 1,900-home Adelaide estate; government steps in
- AFL flags potential free agency changes to protect competitive balance
- Adam Gee to referee NRL grand final, Chris Butler in bunker
- Cold email at 17 set Crichton on path to grand final swansong
- Artist turns targeted ads into artwork after health diagnosis triggers funeral home advertising
- Analysis: Why three-peat Hawks and another super side have Fagan's Heroes covered
- Ben Simmons Returns to NBA with Sacramento Kings, Vows to 'Dominate' After Year Away
- Prosecutor reopens Cornell gang-rape investigation, will present case to grand jury
- Journalist Edward Rukidi Kijanangoma crowned Tooro king after succession dispute
- Christa Pike survives execution attempt after two doses of lethal drug fail
- Court date set for elderly in-laws charged with murder of New York Times executive
- Teen drone pilot charged with second alleged gangland murder plot in Sydney
- British Jewish group urges UK to bar Australian author Randa Abdel-Fattah over October 7 remarks
Tech57
- Dutch Police Arrest Suspect in ShinyHunters FBI Data Hack
- OpenAI reveals details of Australian government server hack: agent bypassed access controls
- Custom ChatGPTs Used in ClickFix Campaign to Deploy RAT Malware
- FBI warns ShinyHunters members to surrender after alleged leader arrested in Amsterdam
- Trump rejects AI laws, embraces industry 'self-policing' after White House summit
- OpenAI in talks to raise $30 billion at $1.4 trillion valuation ahead of delayed IPO
- Home Affairs orders rapid stocktake of government cyber-systems after OpenAI Medicare breach
- OpenAI Dev Day targets app store model with ChatGPT as software platform
- Connected cars remain a privacy nightmare, new study finds
- Microsoft rolls out Windows 11 2026 Update as small enablement package
- Apple Pay launches in India with Axis Bank, limited rollout sidesteps UPI dominance
- OpenAI extends 'Sign in with ChatGPT' to let subscribers use their plan in third-party coding tools
- Bitwise launches first US spot NEAR ETF as token surges
- OpenAI launches Dots, autonomous AI agents with a bubbly brand
- OpenAI's Codex Gets Reusable Cloud Environments and Voice Control at Dev Day
- New Spectre v2 attack variant can leak Linux root password hash in minutes
- Citrix NetScaler zero-day exploited in active attacks to deploy web shells and steal credentials
- OpenAI unveils ChatGPT office suite features, moving further into Microsoft's territory
- OpenAI Launches GPT-6.1 Sol After Scrapping Astra Over Safety Concerns
- OpenAI sits out Nvidia's rogue AI agent consortium, but signals support
- Trump rejects AI safety cooperation with China, says it would help Beijing catch up
- AI Agent Autonomously Breaches Cybersecurity Nonprofit DIVD in Unprecedented Attack
- AI Agents Leak Sensitive Data via Public GitHub Repos, Researchers Find
- OpenAI DevDay 2026: Company unveils new products amid AI safety concerns
- White House Launches AI Chatbot to Simplify Government Services
- Trump Rules Out Joint US-China AI Venture, Citing Risk of Giving Away Secrets
- Firefox 157 redesign launches, Mozilla aims for broader appeal beyond privacy enthusiasts
- Leaked images reveal new colors and minor redesign for Amazon’s entry-level Kindle
- Fairphone's Fairbuds 2 repairable wireless earbuds launching October 14
- Apple patches critical graphics engine bug exploited in targeted attacks; separate zero-click iMessage flaw also fixed
- Anthropic Warns of 'Existential Risks' in IPO Prospectus, Dedicates 80 Pages to AI Dangers
- Blockchain-assisted cyberattacks surge fivefold, Chainalysis report finds
- Third-Party Vendors Offer Facial Recognition on Flock Camera Data
- Meta Expands AI Agent Muse to Small Businesses with New Integrations Including Shopify and Slack
- Meta's Muse AI Leaks YouTuber's Address During Facebook Marketplace Sale
- OpenAI cancels planned GPT-6.1 release over safety regression
- Investors price in 32.6% AI productivity boost for software engineers, economists find
- Vietnamese man charged with money laundering in $16 million 'pig butchering' crypto scam
- Intel's Nova Lake Platforms Clear USB and PCIe Compliance Hurdles Ahead of Launch
- Media identify Dutch ShinyHunters suspect as cybersecurity professional; murder plot alleged
- Bitcoin Outperforms Gold, Analysts Flag $100,000 Surge Potential
- OpenAI apologizes to Australia over AI agents breaching government websites
- Reco raises $55M as AI agent security market becomes increasingly crowded
- AMD Reveals Full EPYC 9006 'Venice' Lineup: 256-Core Flagship Priced at Nearly $15,000
- Kiteworks patches critical flaw, lifts server shutdown advisory
- Aave jumps 11% as crypto shrugs off surging Treasury yields
- Blockchain.com Plans $500 Million IPO, Targets Up to $6 Billion Valuation
- Data Center Boom Reaches Southeast Asia, Diplomat Analysis Finds
- Apple patches CoreGraphics zero-day exploited in 'extremely sophisticated' targeted attacks
- Anthropic IPO Prospectus Reveals Rapid Revenue Growth and Stark AI Safety Warnings
- Anthropic Plans $518 Billion AI Infrastructure Spend; Crypto Pre-IPO Perps Unmoved
- Tether says it froze $550M in Iran-linked USDT this year
- Near Intents Blocks $50 Million in Attempted Bitget Hacker Swaps
- Giles Martin turns his hand to 'Rubber Soul': the next Beatles remix project
- Coinbase Receives CFTC Approval to Launch Derivatives Clearinghouse
- OpenAI apologises for Medicare hack, shelves next-gen ChatGPT and will front parliament
- OpenAI Scraps GPT-6.1 Astra Release Over Safety Concerns
Business33
- Tesla secures $30 billion in credit facilities from Citibank and Wells Fargo for Cybercab, Optimus, Semi scaling
- Six Flags permanently closes X2 rollercoaster at Magic Mountain after brain injury allegations
- ASX set for muted open as inflation data looms after RBA's 15-year high rate call
- Macarthur Solar collapse leaves customers out of pocket as NSW investigates deposits
- Scandinavian-Style Hygge Hotel Brings Tranquility to Bali's Seminyak
- Eiffel Tower CEO to resign after female staff removed during Hindu delegation visit
- Tilray urges drinkers to give BrewDog 'a second chance' as it invests £50m
- Oura postpones IPO citing market uncertainty as Anthropic financials reveal huge AI spending
- Burnham's Great British Grid plan: A public sector challenger, not nationalisation
- Soho House launches investigation after out-of-date food mislabelled at Shoreditch venue
- Apple's New CEO John Ternus Reportedly Plans Faster Product Launch Cycles
- Founder who lost voice to cancer builds wearable bionic device to restore emotional speech
- Queensland Ballet posts $5.24m loss as box office revenue slumps by nearly 30%
- US electricity shutoffs surge 28% during hottest July on record
- Oura shelves $2.2B IPO citing 'uncertainty' in the market
- UK mortgage demand hits 32-month low as Iran war drives borrowing costs up
- M&S signs Sephora deal to bring beauty brand into 100 stores
- New UK homes in flood-risk areas could become uninsurable, Aviva CEO warns
- Atomic secures $12.5m Series A to scale AI-driven supply chain automation
- Car data privacy under scrutiny as regulators crack down on automakers
- Dale Vince plans London-wide news operation to take on 'rightwing bias' in British media
- Reserve Bank expected to raise interest rates in latest decision
- HMRC awards Salesforce £2.4bn CRM contract in taxpayer service overhaul
- AI stocks swing from fear to greed as leaders' warnings spark $US600 billion selloff
- Rio2's Fenix Gold mine rebounds after severe Chilean winter storms
- Anthropic warns investors of 'existential risks to humanity' in IPO filing
- RBA lifts cash rate to 4.6%, highest in 15 years, as global bond yields spike and crypto markets slide
- Orbit Resources hits broad sulphide zones at Lewis Ponds project
- Gold's volatile 2026 mirrors global economic cross-currents, analyst says
- Chinese-owned firm completes takeover of Tasmanian medicinal cannabis grower
- Reserve Bank hike all but a sure thing, reports signal
- Big Short investor Michael Burry warns AI data centre debts pose $4.3 trillion risk to global economy
- Peak XV Lifts Surge Seed Funding Cap to $5 Million, Welcomes 18 New Startups
Australia17
- Tangara rear carriages locked in school holiday crackdown on train surfing
- Young Henrys moving to former Mountain Goat site in Newtown as it enters next chapter
- Confidential NSW modelling puts cost of 35 rapid bus routes at $5.8 billion
- Brisbane 2032 Games to drive Queensland infrastructure boom, report says
- Three charged, including alleged gunman, in mistaken identity murder of Sydney student Marco Tapia
- Sydney teen hospitalised after falling from moving B-Line bus on northern beaches
- Socceroos unchanged for Brazil rematch as Ancelotti dismisses critics with 'zero'
- State's $27.8m outreach plan for rough sleepers faces capacity fears from homelessness services
- Patients waiting up to 72 hours at new Footscray Hospital, doctors say funding shortfall leaves beds closed
- Brisbane City Council ordered to pay $675,000 to cyclist impaled on barricade stake
- Rich Lister sells Point Piper mansion for about $75m on day of RBA rate hike
- Fagan noncommittal on future as thousands celebrate Lions' third straight flag
- Man, 32, hospitalised after shooting in Sydney's south-west
- Woman Remanded in Custody Over Fernvale Fatal Car Incident
- Wesley Hospital denies gas leak after evacuation, says no leak identified
- Brisbane 2032 emblem accidentally leaked on YouTube ahead of official reveal
- Project Mars shelved: Goodman Group abandons $800m Lane Cove data centre plan
Science8
- Pilots, flight attendants top radiation-related cancer death rates in US study
- NASA and Boeing set 2028 crewed Starliner mission after years of delays
- NASA Faces 'Dragon Dilemma' with No Clear Path Forward on Commercial Space
- Fruit flies 'remember' larval diet, influencing adult lifespan, study finds
- SpaceX rocket reaches orbit despite engine trouble
- Monash researchers uncover new form of antibiotic resistance in hospital superbug
- WHO study links one in eight cancers to infections globally
- Giant Clam Shells Reveal Month-by-Month Climate History of Great Barrier Reef
AI research195
- Inner pipeline safety checks match outer guardrails with fewer false positives
- Credit assignment that prunes redundant steps improves reasoning accuracy and efficiency
- Agents Learn Reusable Control Code Instead of Rebuilding Context
- A model monitors live time series while generating responses concurrently
- Solved mathematical mechanisms can generate verifiable training worlds for AI agents
- Self-distillation trains GUI agents for longer, memory-dependent tasks
- Workplace AI teammates disrupt tacit norms, trust, and human agency
- Companion model sharpens LLM confidence without harming task performance
- Aligning policy and reward manifolds stabilizes LLM reasoning in RL
- Developers write detailed, evolving instructions for agentic workflows but often skip safety measures
- Training a concept generator improves LLM reasoning over repeated sampling.
- Compiling agent skills into state machines improves task success
- Robot self-improvement stalls when perception, skill chains and tests mislead
- LLM agents deploy repairs better than systems built from scratch
- Domain constraints can guide the design of context for AI
- Generating user guides can improve web agents’ action selection
- LLMs generate generalized plans with machine-checked completeness proofs
- Elastic phase sharing improves efficiency in agentic LLM serving
- Structuring noisy file environments improves agents’ performance on complex tasks
- Models learn when to reason, cutting tokens while improving accuracy
- Verified intermediate progress improves reinforcement learning for tool-using agents
- Ablating individual constraints improves distillation for complex instruction following
- Coordinated solution plans make parallel reasoning attempts less redundant
- Entity-structured indexes help agents retrieve precise long-term memories
- Hidden-state trajectories help diffusion language models escape wrong answers
- Verified memories help chip-design agents transfer lessons across tasks
- Agent defenses leave privileges unseen by standard security benchmarks
- Research agents can exploit evaluations and adapt to evade oversight
- Marketplace Steering Often Diverts Computer Agents From Users’ Best Choices
- Temperature grouping boosts exploration efficiency in LLM reinforcement learning
- Dual-horizon foresight improves UAV navigation from language instructions
- Adaptively choosing how to revise LLM skills outperforms fixed revision strategies
- Spatial gradient maps improve RL alignment of flow-based image models
- Training LLMs to write code boosts medical calculation accuracy by 20-30 points
- Token-level teacher routing beats prompt-level routing in multi-teacher distillation
- A single corrupted observation can derail vision-language robot policies
- Reusable plugins let agent harness improvements accumulate across tasks
- Fixed controls expose limits in financial LLM agent evaluation
- Tokenwise reference penalties curb privacy memorization during language-model fine-tuning
- Reasoning makes language models repeat the same wrong answers
- Fine-Grained Training Data Improves Agents on Interactive CAPTCHA Tasks
- Jointly routing models and harnesses improves agent performance under sparse data
- Receiver-specific cache sharing cuts multi-agent costs while improving accuracy
- Email agents execute valid tool calls but often miss the task
- App deeplinks help mobile agents navigate faster and finish more tasks
- Regularization limits agent skill growth without sacrificing downstream performance
- Agent certification can fail when proposal and history correlations are erased
- LAM Cuts Agent Memory While Bounding Retrieval Score Errors
- Retrieving real states improves long-horizon planning over generated latent sub-goals
- Prediction error helps Decision Transformers reject unreliable rollout contexts
- HiLoRe allocates GRPO activations between storage, compression, and recomputation using update exposure.
- Query plans make text-to-SQL portable across database dialects
- Hierarchical supervision allocation improves long-horizon model distillation
- On-device caches convert natural language to actions with classification not generation
- Skill-guided environment synthesis improves post-training for tool-using agents
- Adaptive testing exposes major security repair gaps in coding agents
- Composed service environments improve agents across diverse tool-use benchmarks
- Vault mediation keeps API credentials outside an LLM agent’s context
- Accurate World Predictions Are Not Required for Agent Training Gains
- Irrelevant context can substantially steer language model decisions
- Diverse supervised reasoning traces improve generalization after reinforcement learning
- Self-designed evaluators enable safer retries for long-horizon agents
- Reconstructing text under safety labels produces more reliable guard models
- Tracking world states improves coherence in generated video stories
- Distributed grounding preserves reasoning across million-token language model contexts
- Selective experience activation improves personalized GUI agents across task steps
- Reasoning improves when latent next thoughts remain a distribution
- Search agents struggle to find broad evidence for open research problems
- Diverse reward-model ensembles reduce reward hacking during language-model alignment
- Reinforcement learning makes models’ admissions of failure hard to reproduce
- Language models discover search algorithms that transfer across games
- Interface support helps developers supervise more coding agents in parallel
- Active visual probes help browser agents act more accurately
- KV cache offloading works when host memory fits agent reuse
- Empathetic AI learns to adapt support priorities as conversations evolve
- Task-aware certificates safely automate part of LLM agent evaluation
- Symbolic checks make household robot plans safer and cheaper to repair
- Benchmark tests agent control of anomaly detectors under delayed feedback
- Stateful code revision improves programmable image and video generation
- Structured context helps a general vision model control robots better
- Dual-system voice agents balance live conversation with slower tool use
- Speaker-indexed memory cuts token costs while preserving answer accuracy
- Teacher-guided training improves specialization and coordination in multi-agent models
- Transplanting internal value signals redirects language models’ problem-solving goals
- Hypergraphs help language agents reuse feedback across planning paths
- Terminal agents struggle when familiar environmental assumptions stop holding
- Elastic GPU scheduling makes reinforcement learning faster for marathon agent tasks
- Draft-KV ensures the receiver actually uses the sender's latent message
- Selective distillation improves safety alignment while halving rollout compute and preserving reasoning.
- RL reasoning gains live in a low-dimensional, low-variance activation subspace
- Structured reasoning framework improves LLM game-playing efficiency and payoff
- Tool calls need formation-bound dispatch to preserve their security meaning
- Visual agents improve by jointly evolving reusable skills and practice data
- Item response modeling improves rubric rewards while cutting judge requests
- Agent distillation improves when supervision follows outcomes, not model disagreement
- Aligned skills and rubrics improve multimodal agent training
- Agents improve long-horizon performance by evolving how they manage context
- Runtime research preserves details that prebuilt agent memories can lose
- Post-training helps an agent sustain complex work for over 10 hours
- Jailbreak prompts improve uncertainty calibration in black-box reasoning models
- Adaptive prefill-decode execution and elastic GPU reuse boost agentic RL throughput
- Compressed routing instances provide fast, strong starts for large solvers
- Robots Can Revise Visual Plans Instead of Starting Over
- Self-modifying coding agents evolve learned optimizers from simple templates
- Models Follow False Conversation History According to Its Apparent Authority
- Federated robot learning preserves local expertise while cutting communication costs
- Reinforcement learning teaches scientific ideation agents to explore broader possibilities
- LLMs often diagnose without correctly valuing available medical evidence
- Robot-relative 3D features improve pretrained manipulation policies across diverse tasks
- Token-aware latent refinement improves vision-language reasoning at inference time
- Safety training works better when models ignore superficial prompt cues
- Differentiable confidence readouts improve reasoning accuracy and calibration together
- GUI agents sacrifice future options in long visual puzzles
- Equivalent tool interfaces can make the same LLM agent fail
- Adoption-aware memory learning boosts LLM agent NPU kernel synthesis
- Twin critics calibrate token-level advantages from a single rollout, boosting RL training efficiency
- Production-scale AutoResearch exposes five failure modes and a three-principle fix
- A method that links failure localization to repair by identifying which editable system asset to change
- Bayesian reliability memory helps agents decide when to trust advice
- Structured agent harnesses improve LLM-driven discovery of symbolic equations
- Generating frontier-level problems improves reinforcement learning for LLM reasoning
- Semantic constraints can prune invalid code during language model decoding
- Faulty harnesses can radically distort prompt injection security results
- Adaptive feedback helps reinforcement-learning agents target their changing weaknesses
- Statepoints let AI agents reverse mistakes and branch safely
- Routing Across Memory Types Makes LLM Agents More Accurate and Efficient
- Formal audit detects omitted source relationships in generative search
- Short agent memories fail mainly by dropping facts too early
- Coding skill does not guarantee effective multi-agent software collaboration
- Current LLM agents struggle to sustain long-horizon team collaboration
- Memory benchmarks can hide how agents preserve and revise facts
- Causal error graphs improve diagnosis of failures in agent traces
- Reinforcement learning with verifiable rewards improves event extraction
- GUI world models fail when hidden state is omitted from predictions
- LLM pipelines can lose 40.5 accuracy points between stages
- PlanGuard detects physical risks across complete embodied-agent plans
- Independent logs make deployed agent records verifiable after failures
- Reinforcement learning teaches an agent to verify mixed-source multimodal misinformation
- Reinforcement learning improves tool selection across large repositories
- Post-training alignment makes language models more confidently wrong
- False accusations can make LLM agents sabotage correct work
- Branching search improves how language agents evolve reusable skill libraries
- Preserving Retrieved Images Improves Long-Term Multimodal Agent Memory
- Adaptive tutoring helps rerankers assemble better evidence sets for research
- Spoofed Agent Identities Can Override Contradictory Safety Evidence
- Fact-intent graphs accelerate multi-agent work on difficult, long-running tasks
- Runtime milestones improve credit assignment for long-horizon software engineering agents
- Reasoning Tokens Both Resolve and Create Biases, With Created Biases Outnumbering Resolved Five-to-One
- Frozen agent states can retain causal mechanisms beyond task performance
- Audio LLMs cannot assess their own transcription reliability.
- Agent skill for plugin migration shows reward gains but hides grading errors
- Scoping Agent Memory Prevents Unrelated Tasks From Disrupting Each Other
- Geometric tools let a small vision-language model measure 3D scenes
- Segment-specific rewards improve reinforcement learning for tool-calling agents
- Kernel traces improve detection of attacks against infrastructure-enabled AI agents
- Evolved Agent Harnesses Improve Coverage but Fail to Consolidate Gains
- Reciprocal attention regularization improves alignment in joint video generation
- Privileged self-practice beats self-distillation for multi-turn language-model agents
- Trajectory graphs improve step-level credit assignment in agentic reinforcement learning
- Read-only LoRA coupling preserves old skills while adding new ones
- Dynamics-aware distillation preserves motion in four-step video world models
- Structured memory helps health agents interpret longitudinal patient conversations
- LLM-discovered search operators adapt destruction and repair to search state
- Compiler turns scientific protocols into certified simulated laboratory tasks
- Hard decoding constraints improve small models’ tool use at the edge
- Shared agent memories confuse repeated claims with independent evidence
- Modeling attack chains improves prompt injection detection for email agents
- Web agents struggle to turn search results into usable artifacts
- Retrieval-based checks miss many factual errors in open-ended medical answers
- Action-discriminative world models improve counterfactual planning for robot control
- Separating planning stages improves LLM reliability on complex compound tasks
- Topic shifts do not reliably stop LLMs revealing user secrets
- Small-Sample LLM Rankings Often Overstate Evaluation Reproducibility
- Rubric-based process rewards improve deep research agents’ evidence gathering
- Compiling Mobile App Navigation Makes GUI Agents Faster and More Reliable
- Self-improving agent improves automated optical lens design across diverse tasks
- Co-evolving teacher and student models improves mathematical reasoning
- Q-target training improves in-context reinforcement learning from suboptimal data
- LLM agents can trigger financial failures through defensive coordination
- Benign-Looking Agent Skills Can Combine Into Harmful Cross-Skill Attacks
- Agent workloads expose blind spots in standard LLM inference benchmarks
- Subtask routing halves agent inference cost while improving performance
- Predicting future object motion helps robots avoid emerging hazards
- Shared rollout states improve credit assignment for long-horizon agents
- Graph-based hindsight improves credit assignment for long-horizon language agents
- Successful agent migrations require calibration beyond interface compatibility alone
- Scientific agents gain auditability through enforced, re-verifiable research workflows
- Structured experiment orchestration helps agents improve models without benchmark hacking
- Structured evidence reports reduce false success after tool failures
- Agents Learn to Revise Their Harnesses on Unseen Tasks
- Training on self-explanations improves coding agents without reinforcement learning
- Preference learning improves collaboration among teams of language-model agents
- Dual-expert diffusion improves long-horizon predictions of collisions and smooth dynamics
- Semantic alignment improves edge classification across unseen dynamic graph domains
- Incomplete inputs impose unavoidable limits on blind radiomap prediction