All AI news
Browse, filter, and search every article in the archive. The homepage shows the last 24 hours; everything older lives here.
Margaret Atwood says the problem with AI is ‘garbage in, garbage out’
Margaret Atwood criticizes AI for its reliance on poor-quality data, stating that it leads to flawed outputs. She emphasizes the importance of high-quality input to improve AI performance and reliability.
What happened after 2,000 people tried to hack my AI assistant
Simon Willison shares his experience of 2,000 people attempting to hack his AI assistant. This highlights the vulnerabilities and challenges in developing robust AI systems that can withstand malicious attempts.
An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run
Researchers ran an AI model continuously for 19 days on a single MirrorCode task, costing $2,600. This experiment showcases the potential for sustained AI performance on complex tasks, pushing the boundaries of model capabilities.
Super Mario is mathier than you think
MIT researchers found that Super Mario games can teach complex mathematical concepts through gameplay mechanics. This could change how educators incorporate gaming into math curricula.
Reinventing the zipper
Researchers at MIT are developing a new type of zipper that uses AI to adapt its tension based on the user's needs. This innovation could lead to more comfortable and functional clothing, enhancing everyday wearability.
Engineered “mini livers” could be injected as an alternative to transplantation
MIT researchers developed engineered mini livers that can be injected as a potential alternative to organ transplantation. This innovation could significantly reduce the need for donor organs and improve treatment options for liver disease patients.
Plants appear to detect the patter of falling rain
Researchers found that plants can detect the sound of falling rain and respond by opening their stomata. This discovery suggests plants have a more complex sensory system than previously thought, potentially influencing agricultural practices.
A breath test could diagnose pneumonia in minutes
Researchers developed a breath test that can diagnose pneumonia within minutes. This rapid detection could significantly speed up treatment and improve patient outcomes.
How GPT-5 helped immunologist Derya Unutmaz solve a 3-year-old mystery
OpenAI's GPT-5 helped immunologist Derya Unutmaz crack a three-year-old research mystery. This collaboration showcases how advanced AI can accelerate scientific discoveries and problem-solving in complex fields like immunology.
Prompt Injection as Role Confusion
Simon Willison examines how prompt injection can lead to role confusion in AI systems. This insight helps developers understand vulnerabilities and improve AI safety measures.
Red-Teaming after Mythos — Zico Kolter & Matt Fredrikson, Gray Swan
Zico Kolter and Matt Fredrikson are launching a new initiative focused on red-teaming AI systems to improve their robustness. This effort aims to identify vulnerabilities and enhance safety measures in AI applications.
The AI world is getting ‘loopy’
AI researchers are exploring 'loopy' architectures that allow models to process information in a more dynamic way. This could lead to more adaptable and efficient AI systems that better understand complex tasks.
New benchmark exposes how badly AI struggles with real knowledge work
Researchers just revealed a new benchmark showing AI's struggles with real knowledge work. This exposes significant gaps in AI's ability to handle complex tasks that require deep understanding and context.
A startup claims it broke through a bottleneck that’s holding back LLMs
A startup just announced a breakthrough that addresses a major bottleneck in large language models. This advancement could enhance the performance and efficiency of LLMs across various applications.
OpenAI researchers show small doses of "beneficial trait" training make AI models broadly safer and harder to manipulate
OpenAI researchers are training AI models with small doses of 'beneficial trait' training to enhance safety and reduce manipulation risks. This approach aims to make AI interactions more reliable for users.
The inevitable weakness of metrics
Researchers are pointing out the limitations of relying solely on metrics to evaluate AI systems. This means developers might need to adopt more holistic approaches to assess AI effectiveness beyond just numerical scores.
Microsoft researcher builds a working neural network out of goats in Age of Empires II to critique AI science
A Microsoft researcher creates a functional neural network using goats in Age of Empires II to critique AI science. This unconventional approach highlights the intersection of gaming and AI research, pushing boundaries in how we understand neural networks.
How easily can Russian propaganda fool AI models? A new benchmark finds out
Researchers just established a benchmark to test how well AI models can detect Russian propaganda. This means developers can better understand and improve AI's ability to identify misleading information.
Accelerating researchers and developers building multilingual AI with a new open dataset
GitHub just released a new open dataset aimed at accelerating multilingual AI development. This resource helps researchers and developers create more inclusive AI models that understand diverse languages.
[AINews] Open Models, Model Labs vs Agent Labs, and What's Untrainable — Sarah Guo
Sarah Guo discusses the differences between Open Models, Model Labs, and Agent Labs. Understanding these distinctions helps clarify how various AI systems are developed and utilized in real-world applications.