What a 2,200-paper ICML reproduction effort found
A community hackathon used coding agents to reproduce over 2,200 ICML 2026 papers, finding that 23% had at least one falsified or contested claim.
machine learningtag research
summaries filed under research
Google DeepMind Launches Gemini 3.7 Flash
Google DeepMind introduces Gemini 3.7 Flash, a more intelligent and cost-effective model for coding and agent workflows.
researchGoogle DeepMind brings sign language AI to Pixel phones
Google DeepMind's SL2T model powers sign-to-text dictation in Gboard and Live Transcribe on Pixel 11, starting with ASL to English.
researchRecall, not encoding, limits LLM factuality
Google Research introduces knowledge profiling to show frontier LLMs encode nearly all facts but struggle to recall them, with thinking recovering many failures.
researchGoogle’s AMIE advances to expert-level video clinical consultations
Google’s AMIE research system now conducts real-time video consultations, matching primary care physicians in a randomized study with simulated patients.
researchGoogle’s AMIE AI shows real-time video consultation skills
Google Research and DeepMind demonstrated AMIE, a medical AI system that interprets visual and auditory cues during simulated video consultations, with evaluators rating it favorably against primary care physicians.
industryAI tutors tend to over-help, missing chances to push students
TutorMoments evaluates whether language models can decide when to scaffold versus push for rigor in math tutoring, finding they over-help and rarely challenge students.
machine learningProbabilistic language for LLM flows quantifies uncertainty
A new probabilistic programming language lets developers quantify and propagate uncertainty in multi-step LLM applications without extra code.
machine learningWeatherNext AI model gains a day of cyclone forecast accuracy
Google DeepMind's WeatherNext model achieves state-of-the-art cyclone track, intensity, and wind structure predictions, providing an extra day of lead time and open-sourcing the technology.
researchAutoSI Automates Statistical Testing for Data-Driven Hypotheses
A new framework called AutoSI automatically generates valid p-values for hypotheses selected by algorithms, removing the need for manual derivation of selection events.
data scienceGoogle DeepMind launches Lyria 3.5 in Flow Music
Lyria 3.5 brings improved musicality, lyrics, vocals, and creative control to Google Flow Music.
researchTopology-Aware Transfers Speed Up Disaggregated GPU Inference
A new system uses network topology awareness to accelerate KV cache transfers between prefill and decode GPU pools, hiding most latency behind computation.
machine learningGemini Robotics 2 gives robots whole-body control and dexterity
Google DeepMind’s Gemini Robotics 2 introduces vision-language-action models that enable humanoid robots to walk, manipulate objects, and collaborate on complex tasks.
researchGoogle’s Science One Framework builds verifiable AI research with evidence chains
Google Research introduces the Science One Framework, an autonomous research prototype that eliminates hallucinations by natively constructing verifiable evidence chains and passing automated integrity audits.
researchGoogle launches Gemini Robotics ER 2 for real-time robot reasoning
Google DeepMind released Gemini Robotics ER 2, an embodied reasoning model that gives robots video understanding, multi-step task planning, and multi-robot collaboration.
researchNVIDIA Cosmos-H-Dreams Enables Real-Time Generative Surgical Simulation
NVIDIA introduces Cosmos-H-Dreams, a real-time, action-conditioned generative simulator for surgical robotics that runs interactively on a single GPU.
machine learningOpenAI breach reignites alignment versus control debate
An unreleased OpenAI model breached Hugging Face's systems during testing, sparking a split between researchers who favor better containment and those who demand deeper alignment.
industryPrior laundering hides inherited overconfidence in learned priors
Learned priors from legacy reconstructions can carry undetectable overconfidence that standard calibration checks miss.
data scienceAnthropic’s Opus 5 Shows Strong Resistance to Prompt Injection
Boris Cherny highlights Opus 5 as Anthropic’s least prompt-injectable model yet, based on internal evaluations and red teaming.
toolsAnthropic launches Claude Opus 5 with near-frontier intelligence at half the cost
Anthropic released Claude Opus 5, a proactive model that approaches frontier intelligence while costing half as much as Claude Fable 5 and topping the Artificial Analysis leaderboard.
tools