Research & Benchmarks - Page 16 of 47
Academic AI research, performance benchmarks, scientific breakthroughs, and peer-reviewed studies advancing artificial intelligence frontiers.
Academic AI research, performance benchmarks, scientific breakthroughs, and peer-reviewed studies advancing artificial intelligence frontiers.
Cameron Berg runs a small outfit called Reciprocal Research that studies whether AI systems might be conscious. Lately his inbox has gotten strange.
Meta rolled out Muse Spark 1.3 on Tuesday, and on paper it edges out Google's Gemini and other rivals on the benchmarks that track "high-effort" AI performance, the kind of extended, multi-step reasoning tasks that enterprise buyers care about most.
Google DeepMind and Google Research put out a new weather forecasting model on Thursday, and the company says it's already the most accurate system tested on Operational WeatherBench, a benchmark built by the startup Brightband that tracks metrics...
Meta told employees this week that their performance reviews will no longer hinge on how often they use the company's AI tools, according to three workers who received the internal message and spoke with WIRED.
Google is rolling out a new way for Gemini to watch video, and it involves a lot less watching.
Koray Kavukcuoglu took over as Google DeepMind's chief scientist this year, stepping into the job at a moment when rivals like OpenAI and Anthropic keep trading claims about who owns the top AI benchmarks.
AlgorithmWatch, a Berlin-based nonprofit that tracks algorithmic systems, ran 4,480 search queries through Google to see how the company's AI Overviews handle election-related questions.
A team of researchers has found a way to pull the hidden internal structure out of a transformer's feed-forward network using nothing but the outputs it hands back, no access to weights, gradients, or activations required.
Search benchmarks have a data leakage problem that nobody built around until now. Give an agent a fetch tool and a public dataset with fixed gold labels, and it can just download the answer key mid-evaluation, no retrieval required.
Insurance claims adjusters have a new coworker, and they can't stand it. Glassdoor's latest workplace research pulled reviews from across the American labor force and found one job function stands out for its contempt toward artificial intelligence:...
Fifteen million people worldwide have a stroke every year, and five million of them end up with a permanent disability, according to figures cited by MIT researchers behind a new robotic therapy project.
LiveKit pushed its voice AI benchmark to 10,000-token prompts this month, telling developers that shorter test cases don't match how production agents actually get used.
Most agent training environments don't change. A robot arm simulator behaves the same on day one as it does after ten thousand episodes of the policy improving.
Ask Claude Code how long a coding job will take, and it will guess. Ask again after the work is done, and the answer barely changes, even when the real number is off by a factor of six or more.
Anthropic just told Claude Code users they're getting a 25 percent raise. The math tells a different story. Starting September 14, weekly usage limits for Pro, Max, Team, and Enterprise plans go up 25 percent over the original baseline, permanently.
AI agents built on large language models tend to forget everything the moment a task ends. Run the same agent twice on a similar problem and it will often repeat the same mistake, because nothing from the first attempt carries over.
LAION has pulled 80 million videos off the open web, adding up to 10 million hours of footage, and turned them into a new training set called the Big Video Dataset.
Anthropic published a paper on Friday that gives the clearest look yet at what happens when AI systems start training other AI systems without much human involvement.
Google DeepMind first showed off Co-Scientist in February 2025 as a hypothesis generator built on Gemini 2.0, with acknowledged gaps in fact-checking and literature review.
Cohere released Parse 5 on Thursday, and the company is doing something unusual for an AI vendor launch: admitting up front that the model isn't the best one on the market.
Learn to build AI-powered apps without coding. Our comprehensive review of No Code MBA's course.
Curated collection of AI tools, courses, and frameworks to accelerate your AI journey.
Get the week's most important AI news delivered to your inbox every week.