Language models often can’t remember the start of a conversation. A new text-compression method could change that, and make AI chat sessions much less infuriating. Conversational AI...
ChatGPT answers burn through more tokens when you’re rude to it, raising your enterprise bill; but saying ‘please’ can shrink your costs. It’s said that politeness costs...
Though rare disease diagnosis is a particularly hard challenge for AI (as it is for humans), popular language models ChatGPT and Gemini show promising performance when...
If AI-generated video and audio get good enough, deepfake detectors based on visual artifacts or other traditional signals won’t work anymore. But given how rarely people...
New research shows AI can use people’s faces to estimate the year a photo was taken, combining age guesses with known birth years to beat current...
Popular AI chat models secretly waste huge amounts of paid tokens on pointless verbiage. Affected models actually know they are doing this, but cannot stop themselves. Large...
New research suggests that ChatGPT’s fact-check labels beat likes, shares, and even trusted news brands when it comes to shaping what people believe and want to...
It’s surprisingly hard to get AI models to watch and comment on actual video content, even if they are made for the task. They’re more interested...
New research shows that open source ChatGPT-style AI setups can potentially route callers to the right person in a call center using natural language, without following...
A new method can secretly watermark ChatGPT-like models in seconds without retraining, leaving no trace in general output and surviving all feasible removal attempts. The subtle difference...
New research demonstrates how AI systems can now write fake scientific papers that other AIs accept as real, dodging detection routines that once worked, and exposing...
Generative AI models such as ChatGPT are now flooding academic publishing platforms with AI-generated survey papers at levels that are making the signal-to-noise ratio critical. A...
ChatGPT-style vision models can be manipulated into ignoring image content and producing false responses by injecting carefully placed text into the image. A new study introduces...
According to new research, ChatGPT and similar models now show a clear bias toward text they believe was written by humans, even when that belief is...
Censorship in language models may be undermining their ability to report truth at a wider level. New research finds that the same internal mechanisms used to...
