AI research | The Decoder

Matthias Bastian

Feb 17, 2025

AI research

GenAI turns knowledge workers from problem solvers to AI output verifiers, says Microsoft study

A Microsoft study of 319 knowledge workers shows that the use of generative AI (GenAI) can influence critical thinking. Users focus more on verifying AI results instead of gathering information independently, integrate AI answers instead of solving problems independently and monitor AI instead of executing tasks directly. The researchers warn of an "irony of automation": by taking over routine tasks, AI deprives humans of the opportunity to exercise their judgment. This "cognitive relief" could weaken their own abilities in the long term.

Jonathan Kemper

Feb 17, 2025

AI research

AI agents can be easily tricked into doing stupid things, study says

Cyberpunk illustration: gift box with "SCAM" inscription and robot in neon blue and pink lighting, futuristic design.

Matthias Bastian

Feb 16, 2025

AI research

ChatGPT passes Turing test for psychotherapy, study says

A recent study shows that 830 participants were only able to recognize just above chance level whether therapeutic responses came from ChatGPT or human therapists. The AI responses were even rated as more empathetic. The researchers attribute the better rating of the ChatGPT responses to factors such as length, more positive tone and more nouns and adjectives. However, technology skepticism was also evident: supposed AI responses were rated lower than those assigned to humans.

Jonathan Kemper

Feb 16, 2025

AI research

Meta AI reconstructs typed sentences from brain activity with 80% accuracy

Artistic keyboard composition: Blue keys form "AI/TIME/META" in the center, surrounded by "CODE/DECODE" patterns in grey.

Jonathan Kemper

Feb 15, 2025

AI research

Language models tend to favor other LLMs that make mistakes similar to their own

Stylized depiction of two hands reaching out towards each other in turquoise-coral tones on a beige background, artistic interpretation.

Maximilian Schreiner

Feb 14, 2025

AI research

Study: "Reinforcement learning via self-play" is key to reasoning in language models

Maximilian Schreiner

Feb 13, 2025

AI research

AI study reveals key factors behind LLM's long-term reasoning abilities

Jonathan Kemper

Feb 12, 2025

AI research

AI language models struggle to connect the dots in long texts, study finds

Isometric illustration: Bright purple brain with turquoise network connections, embedded in a cyberpunk-inspired city backdrop.

Jonathan Kemper

Feb 11, 2025

AI research

Goku models from ByteDance can generate realistic product videos without human actors

Collage: Three people during different skin care routines - facial cleansing, facial mask and make-up application.

Jonathan Kemper

Feb 10, 2025

AI research

Hugging Face releases small language model that beats Qwen and Llama most of the time