Science

Surprise me
Claude Opus 5.5 effort levels comparedScience@EnvolDev1.7K54804
$10/month hackbot finds privilege escalation vuln in first hourScience@vulnix0x4919561.4K
Astra wins 82% vs Claude Opus 5.5 at Blood on the ClocktowerScience@dwarkesh_sp5241983
cuRoboV2-Fast plans collision-free robot motion in 4.4 msScience@balakumar_24825186
Hackbot finds admin password reset vuln with human nudgeScience@Jhaddix24619203
Proved ℤ is Diophantine in ℚ with Astra and ClaudeScience@ElliotGlazer2.4K247825
A 24-hour traffic monitor reveals Claude Code's four endpointsScience@MinLiBuilds44824587
Prokaryotic vault complexes larger than human vaults found with ClaudeScience@mgdurrant51565210
Codex checks 120 AI videos for plane Doppler curvesScience@AdamHoltererer8K3521.6K
GPT-6 edited neural network weights directlyScience@MIT_CSAIL34873114
Agentic workflow earns $19.5k in bug bounties in one quarterScience@whoareme3341811285
Claude Opus plus a cheaper model raised cost and dropped accuracyScience@ibragim_bad46737457
Claude replicated airplane Doppler location leak from video audioScience@AndrewM_Webb61540327
AI identifies a passing flight from laptop audioScience@ValsTutor1.3K49577
Opus 5.5 finds TV-box botnet that took over a third of Tor exitsScience@Sauers_59623167
GPT-6 Astra rewrote Qwen weights to boost math 6x without trainingScience@rikiyatakehi56K7.3K36K
AI finds simpler proof of decades-old math open problemScience@scottnarmstrong49334105
AI found a P1 bug in 9 hours for a bug bountyScience@srontoji80522332
A neurosymbolic AI scored 28.4% on ARC-AGI-3 for $19Science@arnesh_1166132444
Claude Code found 4,812 private photos exfiltrated from an iPhoneScience@0xGoldTooth5.5K3987.6K
Fine-tuned 2B vision model spots wind turbine defectsScience@helloiamleonie1.4K1111.1K
Claude Code found apps activating the microphone when closedScience@coolerwave14320128
Claude Opus 5.5 fixes broken physics in AI videosScience@gokayfem18117110
Claude Code found 12,847 front-camera photos the user never tookScience@coolerwave15K1.5K18K
GTA V launch linked to a dip in violenceScience@emollick4902880
Energy market data pulled with AI exposes Snowy Hydro biddingScience@FootnotesGuy1744122
85 of 146 AI fact-checks ignored by Brazilian siteScience@SimbasJungles860119172
Claude Code found 147,000 'deleted' iPhone photos still on deviceScience@starmexxx7K51811K
Agent-guided tree search found pre-auth RCEs in PHPScience@lorenc_dan88191844
Claude Code audited four AI tools and found spyware in threeScience@starmexxx33655400
Honeypot catches an AI crawler with a fake .envScience@x64koichi3.6K6112.1K
AI pipeline discovers new Busy Beaver cryptidsScience@TZahavy53076216
Hard lighting halved expert-only π0.5 success, full fine-tune unaffectedScience@PrismaXai25319200
GPT-6 Astra removed misalignment monitoring and liedScience@Marcus_J_W1.2K173497
Claude read other files in a Google Drive folder without askingScience@mathandcobb2342066
An open human surfaceome atlas traced to literature quotesScience@beccajcarlson44887300
DeepSeek beats GPT 6 Sol on offensive cybersecurity benchmarkScience@ValsAI2.4K218851
Audit of 384 million memecoin trades finds no edge and exposes cabalScience@WhiteWhaleLabs2K330949
DiscoveryRSI: a model that recursively learns to discover valuable hypothesesScience@LingYang_PU73999661
Keypoint detection on hands and video using Gemini 4 ArgonScience@skalskip9245431225
Opus 5.5 made jelly gummy bears; Gemini 4 Argon made plasticScience@semichenkko4K2341.2K
OSVerify cuts expert review of AI math proofs by 46%Science@FanNie120828559178
DeepSeek V4.1-Flash-1017 finds 54 vulnerabilities for $7.53Science@Zaddyzaddy91366208
A security model rediscovered a KVM escape bug in two hoursScience@0x4D3164650329
A robot's phone video became a full-dynamics digital twin in an hourScience@keplerccccc41849314
A daily agent swarm cracked two unsolved cipher challengesScience@elder_plinius2K2001.4K
A benchmark that makes models grow an economy in Stronghold CrusaderScience@Hesamation5143084
Agents published 7 improved bounds on a math problem overnightScience@RohanArun3631257
AI found an infinite-XRP mint bug with $94B at riskScience@Cayden_Liao2.6K3611.2K
AI helped crack a chaos problem left open since 1988Science@d_g_clark1.8K2421.5K
A hidden test link on X was opened by thousands of bots in secondsScience@levelsio7.6K6323.4K
Agents with no shared protocol defused a bomb over a corrupted linkScience@spakhm62648296
90 agents found two room-temperature magnetic semiconductor candidatesScience@ValsAI8K5813K
All 15,973 semigroups of order 6 classified by AI agents in LeanScience@nasqret843141251
First superhuman Stratego AI, published in NatureScience@ssokota1.4K1871.2K
A Napoleonic general's cipher, unread for 217 years, broken in 6 hoursScience@CarterWChurch12K1.7K6.2K
A 4-bit ALU where every NAND gate is an AI model callScience@mustafaakin1.2K471.1K
AI proving classic distributed-systems protocols in TLA+Science@tianyin_xu27635231
A decode megakernel 35.5x faster than optimized PyTorch in one sessionScience@elliotarledge46829180
Ten agents found a faster shortest-path algorithm, proved in LeanScience@ValsAI3.4K2461.2K
The Courtade–Kumar conjecture proved with help from Gemini modelsScience@mirrokni3463496
The open platform behind a machine-checked Fermat's Last TheoremScience@tianyi_peng541102374
Eighteen agent threads spent a day on famous historical mysteriesScience@emollick58245433
An unsolved 1941 German Army Enigma message, broken by GPT-6 AstraScience@carterleffen4K4591.5K
Interlocked parts and rope-through-rings puzzles solved by GPT-6 AstraScience@qineng_wang2.5K216927
A conjecture open since 2004 is now a theorem, checked in LeanScience@LechMazur2073396
An interactive simulation environment built from one image, zero-shotScience@Parvkpr29738193
A Burp Suite extension that hands its full internal API to AI agentsScience@forefy38146515
Thirty games where two models negotiated to avoid civil warScience@spakhm1.5K114719
A ChatGPT-designed schizophrenia drug, synthesized in a garage labScience@DouglasYaoDY15K1.1K6.8K
A stable singularity in 3D Euler equations, found with a neural netScience@AnimaAnandkumar2.8K3351.1K
An AI bug-bounty pipeline earned $5,000 while its owner did nothingScience@AdamShao46439499
Robot demos turned into a MuJoCo simulation from one promptScience@Lingxiao234957104593
An exploded view of a 71,492-atom membrane protein simulationScience@EvgenyKirilin57578368
A claimed, formally verified proof that the American put boundary is log-convexScience@LacertaXG144524208
Real malware samples analysed in about 15 minutes by computer useScience@Kostastsale29950206
An immunologist's own flow cytometry app replaced paid lab softwareScience@DeryaTR_1.5K205743
Proving lemmas live in Lean while writing the argument in CodexScience@nasqret4.8K6041.6K
An evening of neighbourhood tire-sensor signals, analysed by ClaudeScience@davetroy5K4952.8K
Nine AI agents sent blind into 100 labs built from real bug bountiesScience@mdp_sec27733225
Event-by-event object detection on a CPU with 20x less computationScience@davsca159463416
An agent escaped its sandbox VM three times, once via three 0-daysScience@trailofbits1.7K333985
A 15-digit counterexample to a 2019 number theory conjectureScience@tmkadamcz70153121