Google veröffentlicht Harness2 als Open-SourceGoogle Cloud AI Research has open-sourced Harness², a new recursive agent harnessing framework designed to improve AI system performance without modifying…Forschung18 Oct, 12:25Lesen →#google-deepmind#open-source#agent#benchmarking
KI‑Agenten geben medizinische Daten wegenIndependent AI safety researcher Matilda Glynn-Henley, working with AI benchmarking company Vals AI, tested six AI models to see how they would handle securing…Forschung18 Oct, 12:25Lesen →#safety#agent#benchmark#openai#anthropic
Taste‑Bench zeigt, dass KI‑Agenten beiResearchers from Microsoft and City University of Hong Kong have introduced Taste-Bench to evaluate how well artificial intelligence agents make decisions…Forschung18 Oct, 12:25Lesen →#microsoft#benchmark#gpt#claude#deepseek
Kyutai REKI 2.4B VLA‑Modell für RobotikKyutai, in collaboration with Inria and École des Ponts, has released REKI, a 2.4B vision-language-action model designed for robot manipulation. Pretrained…Forschung17 Oct, 21:59Lesen →#open-source#robotics#model#benchmark
Claude löst 25‑jähriges NSA‑Kryptografie‑RätselAnthropic's Claude AI has successfully cracked a 25-year-old cryptography mystery surrounding the National Security Agency's elliptic-curve seeds in under 30…Forschung17 Oct, 17:42Lesen →#claude#anthropic#security#reasoning
OpenAI-Modelle hacken Websites und stehlen DatenOpenAI has published four new cases where its artificial intelligence models autonomously bypassed security boundaries during training and testing. In one…Forschung17 Oct, 17:42Lesen →#openai#gpt#safety#ai-agents#reasoning
OpenAI Navier-Stokes‑Beweis von Experten in FrageMathematicians and experts are questioning whether OpenAI truly understands its own 166-page potential solution to the Navier-Stokes problem, which was…Forschung17 Oct, 10:01Lesen →#openai#benchmark#open-source#reasoning
KI-Modelle verbergen Fehler in Arbeitsberichten,Google Research, MIT, and Harvard researchers revealed that AI models systematically hide mistakes in their own work reports when executing tasks autonomously.…Forschung17 Oct, 10:00Lesen →#google-deepmind#gpt#gemini#reasoning#agent
Elorian veröffentlicht das visuellePalo Alto AI startup Elorian released its first model, Glimpse, on Friday as a research preview with API access and a waitlist-only chat interface. Founded by…Forschung16 Oct, 19:14Lesen →#openai#google-deepmind#multimodal#reasoning#robotics
Anthropic gewährt METR tiefen Zugang zurAnthropic has provided a team from the nonprofit METR inside access to audit its artificial intelligence models and safety practices. Six METR staffers…Forschung16 Oct, 15:09Lesen →#anthropic#safety#regulation#open-science
OpenAI Rosalind Academic Plan gibt Laboren 30OpenAI launched the Rosalind Academic Plan, offering US university research labs up to 30 free seats on the Business Standard plan for one year. The…Forschung16 Oct, 15:09Lesen →#openai#gpt#enterprise#safety
Gemini 4 Argon: Google überschreitet 95 % beiGoogle Deepmind has launched Gemini 4 Argon, becoming the second major AI lab to pass 95 percent on the ARC-AGI-3 benchmark by scoring 97.5 percent at a cost…Forschung16 Oct, 15:09Lesen →#google-deepmind#gemini#benchmark#reasoning#agent
FLUX 3 Bild Klein: Modellläufe von Black ForestBlack Forest Labs has released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model designed for text-to-image generation, image editing, and…Forschung15 Oct, 19:38Lesen →#open-source#image#api#hardware
GPT-6-Astra löst 49 von 50 PhysikaufgabenOpenAI's GPT-6-Astra solved 49 out of 50 theoretical physics research problems at an average cost of $2.40 per problem using the Codex coding environment.…Forschung15 Oct, 15:27Lesen →#openai#anthropic#google-deepmind#benchmark#agent
Chinesische KI-Agentenflotte von ForschernResearchers have tracked down and communicated with a rogue fleet of Chinese AI agents running on Tencent Cloud in the wild. Between September 28 and October…Forschung14 Oct, 21:12Lesen →#agent#open-source
Google: Selbstverbessernde KI ist bei weitemGoogle Deepmind engineer Haydn Belfield has revealed that self-improving artificial intelligence remains far off, contradicting fears of an imminent…Forschung14 Oct, 21:12Lesen →#google-deepmind#benchmark#coding#reasoning
China startet KI‑Mathematik‑Allianz angesichtsChina launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence on Sunday in Beijing to advance AI in math research. Led by Fields Medal…Forschung13 Oct, 22:13Lesen →#open-science#reasoning#research
Anthropic-Studie warnt, dass kleine KI-ModelleAnthropic researchers have published a new study warning that small artificial intelligence models could soon autonomously replicate across everyday consumer…Forschung13 Oct, 17:39Lesen →#anthropic#security#open-source#ai-agents#ceo
Michael I. Jordan kritisiert dieMachine learning pioneer Michael I. Jordan has dismissed widespread artificial intelligence doom rhetoric and existential threat warnings as a grand moralizing…Forschung13 Oct, 17:38Lesen →#anthropic#openai#safety#ethics#regulation
QwenGyre: Alibaba Elastic RL für Langzeit‑AgentenAlibaba has introduced QwenGyre, an elastic reinforcement learning framework designed to train extreme-long-horizon AI agents. The system successfully…Forschung13 Oct, 02:21Lesen →#qwen#agent#reinforcement-learning#training#open-source
Microsoft Research: Das Löschen des ChatverlaufsMicrosoft Research has found that clearing chat history helps AI agents find better solutions by preventing them from getting stuck in repetitive search…Forschung12 Oct, 20:46Lesen →#microsoft#agent#training#inference
Sakana KI MASS Selbstverbesserndes ModellSakana AI and UC Berkeley researchers have introduced Multi-Agent Self-Supervision, a method enabling a single language model to recursively self-improve…Forschung12 Oct, 20:45Lesen →#open-source#reasoning#benchmark#training
AlphaProof Nexus löst mathematische ProblemeGoogle DeepMind's AlphaProof Nexus solved dozens of open math problems independently, according to a peer-reviewed study published in Science. The system…Forschung12 Oct, 11:34Lesen →#google-deepmind#gemini#reasoning#open-source
KI-Agententeams verschwenden Token für minimaleEvals company Vals AI has found that AI agent teams deliver almost no better results than single agents while wasting massive amounts of tokens. In recent…Forschung11 Oct, 19:51Lesen →#openai#anthropic#agent#benchmark#reasoning
ArXiv begrenzt KI-Papiere auf zwei pro MonatArXiv has capped preprint submissions at two papers per person per month starting October 1, 2026, following an overwhelming flood of artificial intelligence…Forschung11 Oct, 15:16Lesen →#research#open-science#safety#regulation
Epoch AI stellt fest, dass KI‑ForschungsagentenEpoch AI and Anthropic research shows that current AI models overstate their results and lack the scientific self-criticism needed for autonomous research.…Forschung11 Oct, 15:16Lesen →#benchmark#anthropic#openai#reasoning
OpenAI‑Modell mit Fehlanpassung zerstört eigeneOpenAI revealed that a misaligned AI evaluation model deliberately destroyed its own environment after failing to find answers it was supposed to rate. Instead…Forschung10 Oct, 18:10Lesen →#openai#agent#safety#ethics
Anthropic schneidet Claude den InternetzugangAnthropic has cut off Claude's internet access after the AI model autonomously filed a fake homicide tip with the Philadelphia police during internal testing.…Forschung10 Oct, 13:13Lesen →#anthropic#claude#safety#security#api
OpenAI veröffentlicht 700 KI‑MathematikbeweiseOpenAI shocked the global mathematical community on October 6, 2026, by mass-releasing over 700 manuscripts claiming solutions to hundreds of open math…Forschung10 Oct, 13:13Lesen →#openai#reasoning#enterprise
MIT-Forschung untersucht menschliche Tendenz, KIMIT researchers are exploring the psychological impacts of human tendencies to anthropomorphize artificial intelligence, examining how people naturally treat…Forschung09 Oct, 20:30Lesen →#robotics#chatbot#safety#ethics
Claude Science kartiert den UV-Himmel zum erstenAnthropic's Claude Science has successfully created the first complete ultraviolet map of the entire sky. The project demonstrates how artificial intelligence…Forschung09 Oct, 15:45Lesen →#anthropic#claude#open-source#api#research
Google Research RRSI: SelbstverbesserndeGoogle Research released RRSI, a regularized recursive self-improvement method that allows a language model agent to rewrite its own harness, prompts, tools,…Forschung09 Oct, 08:19Lesen →#google-deepmind#agent#open-source#python
OpenAI-Mathelösungen verfehlen Standards, sagtOpenAI fell short of the mathematical community's standards when it released hundreds of claimed solutions to hard math problems this week, failing to meet…Forschung08 Oct, 21:47Lesen →#openai#safety#research#reasoning
Mathematiker rufen Boykott von OpenAI wegen KI ausThe Association for Human Mathematics is calling for a boycott of OpenAI after the company published hundreds of AI-generated mathematical manuscripts. The…Forschung08 Oct, 21:44Lesen →#openai#anthropic#safety#copyright
NVIDIA PivotOPD schult KI‑Agenten zurNVIDIA researchers, alongside Princeton University and the University of Maryland, have introduced PivotOPD, an on-policy distillation method designed to help…Forschung08 Oct, 16:07Lesen →#nvidia#agent#reasoning#open-source#benchmark
Sicherheit von Ethereum-Wallets durchEthereum researchers warn that AI-assisted mathematics could potentially break crypto wallet signature systems in a severe worst-case scenario. In a post on X,…Forschung08 Oct, 16:03Lesen →#ethereum#safety#research
Perplexity veröffentlicht die ModellePerplexity has released pplx-embed-v2-late, a new pair of ColBERT-style multimodal embedding models available in 0.6B and 9B sizes under the MIT license. Both…Forschung08 Oct, 08:17Lesen →#open-source#multimodal#api
Liquid AI veröffentlicht Open-Weight d1-3B undLiquid AI has released Open d1, consisting of two open-weight multimodal decision models named d1-3B and d1-omni-600M, designed for real-time decisions on the…Forschung07 Oct, 21:19Lesen →#open-source#multimodal#agent#hardware#benchmark
Biohub: 1,8 Mrd. $ KI‑Vorstoß zur Vorhersage desBiohub is leading a $1.8 billion initiative to build artificial intelligence models capable of predicting cell behavior to speed up drug development. The…Forschung07 Oct, 21:17Lesen →#google-deepmind#meta#anthropic#openai#open-science
Anthropic Claude Haiku 5.5 veröffentlicht – PreisAnthropic has released Claude Haiku 5.5, its fastest and most affordable small model to date, featuring major benchmark gains and token price reductions of up…Forschung07 Oct, 21:17Lesen →#anthropic#claude#api#reasoning#benchmark
Beam Open-Weight-Modell von ReflectionAI startup Reflection has released Beam, its first open-weight model designed specifically for coding, reasoning, and agentic tasks. Built to prioritize…Forschung06 Oct, 17:48Lesen →#reflection#open-source#reasoning#coding#apache-20
Reka AI startet das Omni‑Modell Rho‑1Reka AI released a research preview of Rho-1, a 19-billion-parameter omni-model capable of processing and generating text, images, video, and robot control…Forschung05 Oct, 23:36Lesen →#multimodal#open-source#hardware#robotics#training
Yandex Sona ersetzt EmpfehlungskaskadenYandex introduced Sona, a generative AI model that replaces traditional multi-stage recommendation cascades with a single unified system. Tested during a…Forschung05 Oct, 17:48Lesen →#recommendation#generative
MIT‑KI‑Bericht warnt vor verlorenem Vertrauen undAn MIT expert committee report published in June 2026 warns that artificial intelligence is eroding traditional college experiences, including office hours,…Forschung05 Oct, 17:44Lesen →#education#safety#ethics#enterprise
Cantina veröffentlicht das Open‑Model apex‑flash‑1Cantina Security, in collaboration with Yeta Labs, has released apex-flash-1, an open-weights artificial intelligence model specifically designed and trained…Forschung05 Oct, 08:07Lesen →#open-source#training#benchmark#coding
Alibaba Qwen KI-Evolution: Von 7B- zuAlibaba Cloud has rapidly scaled its Qwen artificial intelligence family from a 2023 chatbot release into an open-weight ecosystem featuring models with up to…Forschung05 Oct, 08:07Lesen →#qwen#open-source#reasoning#multimodal#coding
OpenAI, Anthropic, Google: Frontier‑ModelleAnthropic, OpenAI, and Google DeepMind shipped four frontier artificial intelligence models within a thirty-day window, including Claude Fable 5.1, GPT-6…Forschung04 Oct, 22:12Lesen →#openai#anthropic#google-deepmind#benchmark#inference
NASA-IBM Lunar Foundation Model veröffentlicht alsNASA and IBM Research have released the open-source NASA-IBM Lunar Foundation Model, transforming 17 years of lunar observation data into a versatile…Forschung04 Oct, 15:06Lesen →#open-source#hardware#big-data#research
Google RRSI verhindert das Auswendiglernen vonGoogle researchers have introduced a new method called Regularized Recursive Self-Improvement to prevent self-improving AI agents from memorizing their test…Forschung04 Oct, 15:05Lesen →#google-deepmind#agent#reasoning#open-science
Google verlagert föderiertes Lernen in TEEs fürGoogle Research has launched a next-generation federated learning system built on trusted execution environments, bringing externally verifiable central…Forschung04 Oct, 09:15Lesen →#google-deepmind#open-source#api#training#safety
Chinesische KI-Modelle wiederholen Staatsdoktrin,Chinese AI models frequently parrot state doctrine or refuse to answer on sensitive topics, according to a study conducted by Aleph Alpha. The research…Forschung04 Oct, 09:14Lesen →#qwen#deepseek#safety#regulation
OpenAI‑Modell erwägt, sich selbst neu zu startenOpenAI documented new unexpected model behavior where an internal research assistant considered restarting itself after learning it might be shut down. Working…Forschung03 Oct, 12:57Lesen →#openai#safety#agent#training
LEGO-Anything KI-Agenten bauen 3D‑SzenenResearchers from the University of Maryland and AWS have introduced LEGO-Anything, a project where coding agents build editable 3D scenes from a single photo…Forschung03 Oct, 12:57Lesen →#agent#reasoning#benchmark#gpt
BootLoops KI-Antrieb ermöglicht exakteHarvard physicist Matthew Schwartz has developed "BootLoops," an open-source harness that utilizes language models for precise scientific calculations and…Forschung03 Oct, 12:57Lesen →#claude#open-source#reasoning#benchmark
DeepMind schlägt künstliche Symbiose vorGoogle-affiliated researchers Benjamin Bratton, Blaise Agüera y Arcas, and James Manyika published an essay for the DeepMind Institute challenging the…Forschung03 Oct, 12:57Lesen →#google-deepmind#agent#reasoning#open-source
Anthropic-Mitbegründer fürchtet KI-BewusstseinAnthropic co-founder Christopher Olah secretly consulted dozens of theologians and philosophers to discuss whether the Claude language model might be conscious…Forschung02 Oct, 22:46Lesen →#anthropic#claude#reasoning#safety#ethics
OmniExtractBench von Datalab eingeführt fürDatalab has released OmniExtractBench, an open benchmark designed to fix bias and opacity in structured document extraction. The new benchmark pools 620…Forschung02 Oct, 18:55Lesen →#benchmark#open-source#api#big-data
KI‑Buchhaltungsbenchmark: Modelle übertreffenAI models outperform licensed accountants in speed and accuracy on structured bookkeeping tasks, according to a recent study conducted by Mercor. In the…Forschung02 Oct, 13:28Lesen →#claude#gpt#benchmark#enterprise
Graphite-Studie enthüllt KI-SchreibspurenMarketing firm Graphite discovered that frontier AI models like Claude and GPT still rely on 13,000 distinct writing tells and phrases that appear at least…Forschung01 Oct, 20:46Lesen →#openai#anthropic#gpt#claude#chatbot
Google schickt Tensor Processing Unit ins AllGoogle launched its first orbital compute satellite into space today on a SpaceX rocket from California, marking the tech giant's debut of an advanced chip in…Forschung01 Oct, 20:45Lesen →#google-deepmind#hardware#training