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...
New research shows that coding AIs such as ChatGPT suffer from the Dunning-Kruger Effect, often acting most confident when they are least competent. When tackling unfamiliar...
New research indicates that AI models can get smarter at seeing by solving jigsaw puzzles. Rearranging scrambled images, videos, and 3D scenes helps them sharpen their...
A new study finds that ChatGPT-5 and Google Gemini produce hallucinations in 40% of newsroom-style queries, frequently inventing confident-sounding claims unsubstantiated by verifiable facts. Google’s NotebookLM...
ChatGPT-style vision models often ‘hallucinate’ elements that do not belong in an image. A new method cuts down on these errors by showing the model exaggerated...
A new study shows that fine-tuning ChatGPT on even small amounts of bad data can make it unsafe, unreliable, and veer it wildly off-topic. Just 10%...
Beauty filters now do more than hide blemishes: they can help deepfakes and face morphs slip past detection systems. A new study shows that even subtle...
New research shows how ads could soon be embedded directly inside ChatGPT-style answers – not as banners or popups, but woven into the response itself. A...
Emojis can be used to bypass the safety mechanisms of large language models, and trigger toxic outputs that would otherwise be blocked. By this means, LLMs...
Large language models often give confident answers even when the question can’t be answered. New research shows these models often recognize the problem internally, but still...
A new study finds that many popular image datasets used to train AI models are contaminated with test images or near-duplicates, allowing models to cheat by...
A new AI system can realistically reshape people’s bodies in photos, making them fatter, thinner, or more muscular, without changing their face, clothes, or background. The...
Attempts to censor AI image generators by erasing banned content (such as porn, violence, or copyrighted styles) from the trained models are falling short: a new...
New research finds that popular image-centric AI systems don’t just look at what’s in a photo, they also pick up on how the photo was taken....
