Tiernan Ray

As their self-correction and self-reasoning improve, the latest LLMs find new capabilities at a rate that makes it harder to measure everything they can do.

Can generative AI successfully explain its reasoning? I put two models to the test.

We are getting smarter as we engage in prompt engineering, but we are also still defining intelligence.

Despite some uncertainty, enterprise investments in applications soared eight-fold in 2024, with spending on AI-generated code leading the way.

Project Rainier is an 'UltraCluster' that will group hundreds of thousands of Trainium chips.

The company also demonstrates initial training of a one-trillion-parameter AI model on a single machine using conventional DDR5 memory chips.

The company also demonstrates initial training of a one-trillion-parameter AI model on a single machine using conventional DDR5 memory chips.

The engineering of threat testing and red-teaming is the most interesting contribution of Amazon's first 'frontier' models.

Being able to scale physical qubits reliably is nice, of course. Now comes the hard part: actual circuits.

Areas resistant to automation - like legal services and healthcare - are attracting novel applications that could even displace human workers, according to a Bank of America report.