Research & Benchmarks - Page 16 of 28
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.
Teaching robots to navigate our world demands a torrent of visual data. Real footage is costly and scarce. So the industry relies on simulations, but their tell-tale artificiality is a problem—the AI models can spot it.
The Trump administration was done punishing Anthropic. Until it wasn't. At a federal court hearing this week, the government's lawyer was asked a simple question. Would the White House stop hitting the AI startup with new sanctions?
YouTube has never been able to decide if AI is its biggest problem or its best product. It purges junk channels filled with fake trailers one day and gives creators slick AI tools for making videos the next.
Andrej Karpathy built a robot that writes research papers. The former OpenAI and Tesla luminary quietly posted the code, called Autoresearch, on GitHub. It automates the scientific method—for code. You give it a problem.
The machines can spot the pattern. They just can’t tell you if it matters. A 12% swing in spend appears in the variance report, healthy growth or hidden collapse? The algorithm won’t know.
Bigger tiles should mean fewer memory accesses, faster attention, right? Not on NVIDIA GPUs. Across every sequence length tested, large CUDA tiles actually cratered Flash Attention throughput by 18 to 43 percent.
A research team slashed the memory load of large language models by 98 percent. Their secret? Compressing the model's key-value cache fiftyfold.
Anonymous posting just got harder, full stop. Researchers have crafted a method using large language models that can effectively unmask people.
The White House just secured a commitment from seven of the world’s largest tech companies, Amazon, Google, Microsoft, and others, to shoulder the financial burden of their own exploding energy appetite.
Bigger isn't smarter. While the AI industry obsesses over trillion-parameter behemoths, Microsoft just released a model that fits in your pocket. Phi-4-reasoning-vision-15B is a 15-billion parameter model. It is small. It is fast.
A coding agent is only as good as its environment. Give it the right tools, and it transforms. LangSmith CLI just gave agents three new portable skills, trace, dataset, and evaluator, that snap directly into any repo. These aren't abstract concepts.
Behind closed doors, 94% of attendees approved the least popular stance on AI resistance. That number alone should stop you cold. The dissenters? They didn’t matter. The partisanship that fractures public debate, Grok vs. Anthropic, “based” vs.
Forget Silicon Valley. The new front line for the AI boom is a frozen field in northern Sweden. Europe is running out of space and power for the vast server farms that make AI models work.
Your AI probably can't remember its own instructions. The prompts that tell a large language model how to behave have become massive, delicate documents. They're a tax on every query. Microsoft's answer is a training tweak called OPCD.
Wall Street is twitching. A single research note on artificial intelligence now triggers a sell-off; an offhand CEO comment about model training can spark a buying frenzy.
Riley Walz has been called the Jester of Silicon Valley, a title earned through irreverent, wildly inventive web experiments that often feel more like pranks than products. Now he’s joining OpenAI.
Biologists are drowning in cell data. They have machines that listen to a cell's RNA, others that photograph its chromatin. Each tells a different, partial story. The real work begins when you try to stitch those stories together.
Every tech company wants a smarter AI. A few are asking for a kinder one. The current pursuit of raw intelligence in machines is missing the point. A growing number of researchers argue the real goal should be building an AI that feels something.
The AI industry is eating itself, and the menu is full of irony. Researchers resign in protest, only for their creations to turn around and hire the very humans who built them.
Everyone wants faster AI. Almost nobody wants to rebuild their entire system to get it. Researchers have just shown a way to get both: three times faster inference by tweaking the model's own brain, not its house. The trick is a mask token.
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.