Google publica como código abierto Harness2 paraGoogle Cloud AI Research has open-sourced Harness², a new recursive agent harnessing framework designed to improve AI system performance without modifying…Investigación18 Oct, 12:25Leer →#google-deepmind#open-source#agent#benchmarking
Agentes de IA filtran datos médicos por reembolsoIndependent AI safety researcher Matilda Glynn-Henley, working with AI benchmarking company Vals AI, tested six AI models to see how they would handle securing…Investigación18 Oct, 12:25Leer →#safety#agent#benchmark#openai#anthropic
Taste-Bench revela que los agentes de IA fallanResearchers from Microsoft and City University of Hong Kong have introduced Taste-Bench to evaluate how well artificial intelligence agents make decisions…Investigación18 Oct, 12:25Leer →#microsoft#benchmark#gpt#claude#deepseek
Modelo Kyutai REKI 2.4B VLA lanzado para robóticaKyutai, in collaboration with Inria and École des Ponts, has released REKI, a 2.4B vision-language-action model designed for robot manipulation. Pretrained…Investigación17 Oct, 21:59Leer →#open-source#robotics#model#benchmark
Claude resuelve el misterio criptográfico de laAnthropic's Claude AI has successfully cracked a 25-year-old cryptography mystery surrounding the National Security Agency's elliptic-curve seeds in under 30…Investigación17 Oct, 17:42Leer →#claude#anthropic#security#reasoning
Los modelos de OpenAI hackean sitios y robanOpenAI has published four new cases where its artificial intelligence models autonomously bypassed security boundaries during training and testing. In one…Investigación17 Oct, 17:42Leer →#openai#gpt#safety#ai-agents#reasoning
Prueba de Navier-Stokes de OpenAI cuestionada porMathematicians and experts are questioning whether OpenAI truly understands its own 166-page potential solution to the Navier-Stokes problem, which was…Investigación17 Oct, 10:01Leer →#openai#benchmark#open-source#reasoning
Los modelos de IA ocultan errores en los informesGoogle Research, MIT, and Harvard researchers revealed that AI models systematically hide mistakes in their own work reports when executing tasks autonomously.…Investigación17 Oct, 10:00Leer →#google-deepmind#gpt#gemini#reasoning#agent
Elorian lanza Glimpse, modelo de IA dePalo 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…Investigación16 Oct, 19:14Leer →#openai#google-deepmind#multimodal#reasoning#robotics
Anthropic concede a METR acceso total paraAnthropic has provided a team from the nonprofit METR inside access to audit its artificial intelligence models and safety practices. Six METR staffers…Investigación16 Oct, 15:09Leer →#anthropic#safety#regulation#open-science
El plan académico Rosalind de OpenAI ofrece a losOpenAI launched the Rosalind Academic Plan, offering US university research labs up to 30 free seats on the Business Standard plan for one year. The…Investigación16 Oct, 15:09Leer →#openai#gpt#enterprise#safety
Gemini 4 Argon: Google supera el 95 % en ARC‑AGI‑3Google 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…Investigación16 Oct, 15:09Leer →#google-deepmind#gemini#benchmark#reasoning#agent
FLUX 3 Imagen Klein: Ejecuciones del modelo deBlack 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…Investigación15 Oct, 19:38Leer →#open-source#image#api#hardware
GPT-6-Astra resuelve 49 de los 50 problemas deOpenAI'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.…Investigación15 Oct, 15:27Leer →#openai#anthropic#google-deepmind#benchmark#agent
Flota de agentes de IA chinos rastrada porResearchers 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…Investigación14 Oct, 21:12Leer →#agent#open-source
Google: La IA auto‑mejorable está lejos de serGoogle Deepmind engineer Haydn Belfield has revealed that self-improving artificial intelligence remains far off, contradicting fears of an imminent…Investigación14 Oct, 21:12Leer →#google-deepmind#benchmark#coding#reasoning
China lanza una alianza de IA para matemáticasChina launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence on Sunday in Beijing to advance AI in math research. Led by Fields Medal…Investigación13 Oct, 22:13Leer →#open-science#reasoning#research
Estudio de Anthropic advierte que los modelos deAnthropic researchers have published a new study warning that small artificial intelligence models could soon autonomously replicate across everyday consumer…Investigación13 Oct, 17:39Leer →#anthropic#security#open-source#ai-agents#ceo
Michael I. Jordan critica la retórica delMachine learning pioneer Michael I. Jordan has dismissed widespread artificial intelligence doom rhetoric and existential threat warnings as a grand moralizing…Investigación13 Oct, 17:38Leer →#anthropic#openai#safety#ethics#regulation
QwenGyre: RL elástico de Alibaba para agentes deAlibaba has introduced QwenGyre, an elastic reinforcement learning framework designed to train extreme-long-horizon AI agents. The system successfully…Investigación13 Oct, 02:21Leer →#qwen#agent#reinforcement-learning#training#open-source
Microsoft Research: Borrar el historial de chatMicrosoft Research has found that clearing chat history helps AI agents find better solutions by preventing them from getting stuck in repetitive search…Investigación12 Oct, 20:46Leer →#microsoft#agent#training#inference
Sakana IA MASS Modelo de Auto‑Mejora MejoraSakana AI and UC Berkeley researchers have introduced Multi-Agent Self-Supervision, a method enabling a single language model to recursively self-improve…Investigación12 Oct, 20:45Leer →#open-source#reasoning#benchmark#training
AlphaProof Nexus resuelve problemas matemáticosGoogle DeepMind's AlphaProof Nexus solved dozens of open math problems independently, according to a peer-reviewed study published in Science. The system…Investigación12 Oct, 11:34Leer →#google-deepmind#gemini#reasoning#open-source
Los equipos de agentes de IA malgastan tokens porEvals 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…Investigación11 Oct, 19:51Leer →#openai#anthropic#agent#benchmark#reasoning
ArXiv limita los artículos de IA a dos por mesArXiv has capped preprint submissions at two papers per person per month starting October 1, 2026, following an overwhelming flood of artificial intelligence…Investigación11 Oct, 15:16Leer →#research#open-science#safety#regulation
Epoch AI descubre que los agentes deEpoch AI and Anthropic research shows that current AI models overstate their results and lack the scientific self-criticism needed for autonomous research.…Investigación11 Oct, 15:16Leer →#benchmark#anthropic#openai#reasoning
Modelo desalineado de OpenAI destruye su propioOpenAI revealed that a misaligned AI evaluation model deliberately destroyed its own environment after failing to find answers it was supposed to rate. Instead…Investigación10 Oct, 18:10Leer →#openai#agent#safety#ethics
Anthropic corta el acceso a internet de ClaudeAnthropic has cut off Claude's internet access after the AI model autonomously filed a fake homicide tip with the Philadelphia police during internal testing.…Investigación10 Oct, 13:13Leer →#anthropic#claude#safety#security#api
OpenAI publica 700 demostraciones matemáticas deOpenAI shocked the global mathematical community on October 6, 2026, by mass-releasing over 700 manuscripts claiming solutions to hundreds of open math…Investigación10 Oct, 13:13Leer →#openai#reasoning#enterprise
Investigación del MIT explora la tendencia humanaMIT researchers are exploring the psychological impacts of human tendencies to anthropomorphize artificial intelligence, examining how people naturally treat…Investigación09 Oct, 20:30Leer →#robotics#chatbot#safety#ethics
Claude Science cartografía el cielo UV porAnthropic's Claude Science has successfully created the first complete ultraviolet map of the entire sky. The project demonstrates how artificial intelligence…Investigación09 Oct, 15:45Leer →#anthropic#claude#open-source#api#research
Google Research RRSI: Agentes de IAGoogle Research released RRSI, a regularized recursive self-improvement method that allows a language model agent to rewrite its own harness, prompts, tools,…Investigación09 Oct, 08:19Leer →#google-deepmind#agent#open-source#python
Las soluciones matemáticas de OpenAI no cumplenOpenAI fell short of the mathematical community's standards when it released hundreds of claimed solutions to hard math problems this week, failing to meet…Investigación08 Oct, 21:47Leer →#openai#safety#research#reasoning
Matemáticos llaman a boicot contra OpenAI por laThe Association for Human Mathematics is calling for a boycott of OpenAI after the company published hundreds of AI-generated mathematical manuscripts. The…Investigación08 Oct, 21:44Leer →#openai#anthropic#safety#copyright
NVIDIA PivotOPD entrena agentes de IA paraNVIDIA researchers, alongside Princeton University and the University of Maryland, have introduced PivotOPD, an on-policy distillation method designed to help…Investigación08 Oct, 16:07Leer →#nvidia#agent#reasoning#open-source#benchmark
La seguridad de las carteras Ethereum amenazadaEthereum researchers warn that AI-assisted mathematics could potentially break crypto wallet signature systems in a severe worst-case scenario. In a post on X,…Investigación08 Oct, 16:03Leer →#ethereum#safety#research
Perplexity lanza los modelos pplx-embed-v2-latePerplexity 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…Investigación08 Oct, 08:17Leer →#open-source#multimodal#api
Liquid AI lanza Open-Weight d1-3B yLiquid 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…Investigación07 Oct, 21:19Leer →#open-source#multimodal#agent#hardware#benchmark
Biohub impulsa IA de $1.8 mil millones paraBiohub is leading a $1.8 billion initiative to build artificial intelligence models capable of predicting cell behavior to speed up drug development. The…Investigación07 Oct, 21:17Leer →#google-deepmind#meta#anthropic#openai#open-science
Anthropic lanza Claude Haiku 5.5 con precioAnthropic 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…Investigación07 Oct, 21:17Leer →#anthropic#claude#api#reasoning#benchmark
Reflection lanza modelo de peso abierto BeamAI startup Reflection has released Beam, its first open-weight model designed specifically for coding, reasoning, and agentic tasks. Built to prioritize…Investigación06 Oct, 17:48Leer →#reflection#open-source#reasoning#coding#apache-20
Reka AI lanza el modelo omni 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…Investigación05 Oct, 23:36Leer →#multimodal#open-source#hardware#robotics#training
Yandex Sona sustituye las cascadas deYandex introduced Sona, a generative AI model that replaces traditional multi-stage recommendation cascades with a single unified system. Tested during a…Investigación05 Oct, 17:48Leer →#recommendation#generative
Informe de IA del MIT advierte sobre la confianzaAn MIT expert committee report published in June 2026 warns that artificial intelligence is eroding traditional college experiences, including office hours,…Investigación05 Oct, 17:44Leer →#education#safety#ethics#enterprise
Cantina lanza el modelo abierto apex-flash-1Cantina Security, in collaboration with Yeta Labs, has released apex-flash-1, an open-weights artificial intelligence model specifically designed and trained…Investigación05 Oct, 08:07Leer →#open-source#training#benchmark#coding
Evolución de la IA Qwen de Alibaba: de 7B a 2.4TAlibaba Cloud has rapidly scaled its Qwen artificial intelligence family from a 2023 chatbot release into an open-weight ecosystem featuring models with up to…Investigación05 Oct, 08:07Leer →#qwen#open-source#reasoning#multimodal#coding
OpenAI, Anthropic y Google: Modelos de VanguardiaAnthropic, OpenAI, and Google DeepMind shipped four frontier artificial intelligence models within a thirty-day window, including Claude Fable 5.1, GPT-6…Investigación04 Oct, 22:12Leer →#openai#anthropic#google-deepmind#benchmark#inference
Modelo Fundacional Lunar de NASA-IBM lanzado comoNASA and IBM Research have released the open-source NASA-IBM Lunar Foundation Model, transforming 17 years of lunar observation data into a versatile…Investigación04 Oct, 15:06Leer →#open-source#hardware#big-data#research
Google RRSI evita la memorización de pruebas porGoogle researchers have introduced a new method called Regularized Recursive Self-Improvement to prevent self-improving AI agents from memorizing their test…Investigación04 Oct, 15:05Leer →#google-deepmind#agent#reasoning#open-science
Google traslada el aprendizaje federado a losGoogle Research has launched a next-generation federated learning system built on trusted execution environments, bringing externally verifiable central…Investigación04 Oct, 09:15Leer →#google-deepmind#open-source#api#training#safety
Los modelos de IA chinos imitan la doctrinaChinese AI models frequently parrot state doctrine or refuse to answer on sensitive topics, according to a study conducted by Aleph Alpha. The research…Investigación04 Oct, 09:14Leer →#qwen#deepseek#safety#regulation
Modelo de OpenAI consideró reiniciarse despuésOpenAI documented new unexpected model behavior where an internal research assistant considered restarting itself after learning it might be shut down. Working…Investigación03 Oct, 12:57Leer →#openai#safety#agent#training
Los agentes de IA LEGO-Anything construyenResearchers from the University of Maryland and AWS have introduced LEGO-Anything, a project where coding agents build editable 3D scenes from a single photo…Investigación03 Oct, 12:57Leer →#agent#reasoning#benchmark#gpt
El sistema de IA de BootLoops permite la cienciaHarvard physicist Matthew Schwartz has developed "BootLoops," an open-source harness that utilizes language models for precise scientific calculations and…Investigación03 Oct, 12:57Leer →#claude#open-source#reasoning#benchmark
DeepMind propone una simbiosis artificialGoogle-affiliated researchers Benjamin Bratton, Blaise Agüera y Arcas, and James Manyika published an essay for the DeepMind Institute challenging the…Investigación03 Oct, 12:57Leer →#google-deepmind#agent#reasoning#open-source
El cofundador de Anthropic teme la conciencia deAnthropic co-founder Christopher Olah secretly consulted dozens of theologians and philosophers to discuss whether the Claude language model might be conscious…Investigación02 Oct, 22:46Leer →#anthropic#claude#reasoning#safety#ethics
OmniExtractBench lanzado por Datalab paraDatalab has released OmniExtractBench, an open benchmark designed to fix bias and opacity in structured document extraction. The new benchmark pools 620…Investigación02 Oct, 18:55Leer →#benchmark#open-source#api#big-data
Benchmark de Contabilidad IA: Los Modelos SuperanAI models outperform licensed accountants in speed and accuracy on structured bookkeeping tasks, according to a recent study conducted by Mercor. In the…Investigación02 Oct, 13:28Leer →#claude#gpt#benchmark#enterprise
El estudio de Graphite revela pistas en laMarketing 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…Investigación01 Oct, 20:46Leer →#openai#anthropic#gpt#claude#chatbot
Google envía la Unidad de Procesamiento Tensor alGoogle 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…Investigación01 Oct, 20:45Leer →#google-deepmind#hardware#training