the-decoder2d
Frontier models fail hard at "Humanity's Last Exam" but experts question if it matters
An international research team has developed a new benchmark that reveals the current limitations of LLMs. Even the most advanced models fail at 90 percent of the tasks - for now. The test, called ...
the-decoder2d
OpenAI emerges as sole customer for proposed Stargate AI infrastructure project
Inside sources told the Financial Times that the Stargate AI infrastructure project will provide computing power exclusively to OpenAI. The project, announced earlier this week by OpenAI, SoftBank, ...
the-decoder5d
Trump's reversal of Biden's AI safety rules puts US on fast track in AI race
Donald Trump has eliminated his predecessor's AI safety regulations, creating a regulatory gap for artificial intelligence development in the United States. In one of his first moves as president, ...
the-decoder3d
"Nerd sniping" and "think less" attacks emerge as AI models get more time to reason
A new study by OpenAI shows that AI models become more robust against manipulation attempts if they are given more time to "think". The researchers also discovered new methods of attack. A recent ...
the-decoder3d
OpenAI's Operator and Computer-Using Agent bring autonomous AI agents closer to reality
OpenAI has just launched Operator, an AI assistant that can navigate the web on its own. The tool, currently only available to US ChatGPT Pro subscribers, represents a step toward AI assistants that ...
the-decoder3d
Perplexity announces new assistant for Android smartphones
Perplexity is stepping into Google's territory with a new AI assistant for Android that can control apps and handle tasks on its own. The move puts the startup in direct competition with Google's ...
the-decoder6d
DeepSeek's latest R1-Zero model matches OpenAI's o1 in reasoning benchmarks
Chinese AI startup DeepSeek has released two new AI models that they say match OpenAI's o1 in performance. Along with their main models, DeepSeek-R1 and DeepSeek-R1-Zero, they've also launched six ...
the-decoder7d
OpenAI quietly funded independent math benchmark before setting record with o3
OpenAI's involvement in funding FrontierMath, a leading AI math benchmark, only came to light when the company announced its record-breaking performance on the test. Now, the benchmark's developer ...
the-decoder3d
AI image generation gets a boost by borrowing ideas from reasoning models
A team of researchers from NYU, MIT, and Google has found a way to improve AI-generated images by borrowing ideas from recent AI reasoning models like OpenAI's o1. Their approach enhances image ...
the-decoder6d
OpenAI CEO Altman tells followers to "chill and cut expectations 100x" amid AGI hype
OpenAI's AI reasoning expert Noam Brown says there is "lots of vague AI hype" on social media. While acknowledging there are "good reasons to be optimistic" about AI progress, Brown emphasized that ...
the-decoder6d
Sakana AI's Transformer² is a new approach to help language models learn
While today's AI systems are typically trained once to handle various tasks like writing text and answering questions, they often struggle with new, unexpected challenges. Transformer² aims to solve ...
the-decoder8d
OpenAI's new GPT-4b micro model optimizes proteins for longevity research
OpenAI is stepping into life sciences with a new LLM designed to optimize proteins. Early testing suggests the system might work better than human researchers at certain tasks. Working with startup ...