Teachable AI @TeachableAI

This new method helps doctors spot abnormal cell divisions in tissue samples using AI. It combines several powerful pre-trained models and special techniques to improve accuracy. Tests showed it was really good at telling the difference between normal and abnormal cells, which is…

This new framework helps create lots of practice data for teaching big language models (imagine them like super smart chatbots) to understand long conversations and complex information. It can make different types of examples, like dialogues, questions with answers based on…

AI researchers created a test to see how good computer programs are at making music that expresses different feelings. They had six popular AI music generators make over 1,000 songs and asked people to rate how happy, sad, exciting, or calm each song sounded. The study found that…

Researchers are developing exciting new models to predict future trends in data. These models, like Toto, use powerful AI techniques and can even be adapted from large language models we see in chatbots. Some researchers are combining forecasts from different models for even…

AI is being explored to help teachers with grading and giving feedback, but it's not perfect yet. Students may not trust AI answers, and it can be tricked into giving wrong information. Sometimes, using AI might even make students think less critically. More research is needed to…

This new method called Beacon makes it easier to shrink large AI models without losing too much accuracy. It does this by cleverly simplifying how the model's numbers are compressed, making it faster and more efficient than other approaches. Tests showed that Beacon works great…

This new method uses quantum computing to speed up how computers understand large amounts of information. Imagine trying to read a whole library – this method helps do that much faster! It works by combining regular computer techniques with special quantum circuits, making it…

This new technique called LMTransplant helps create better training data for AI models that understand text. Imagine asking a smart computer program to finish sentences in creative ways, but making sure the new sentences still make sense with the original ones. That's what…

Researchers developed a new method called Difficulty-Adaptive Rollout Sampling (DARS) to make large language models better at reasoning. DARS balances exploring many solutions (breadth) with digging deeper into promising ones (depth). This approach helps the models tackle tough…

Large language models (LLMs) were tested to see if they could help prioritize housing for homeless people, but they weren't very good at it. They made different decisions about who needed help most, even when given the same information. LLMs didn’t match up well with how experts…

Imagine a company organized like a puzzle! Employees fit into departments, and their connections are like puzzle pieces interlocking. To understand this, we use tables – think of them as labeled boxes holding information about employees, departments, who works where, and who…

Imagine drawing a simple sketch of a dress and then describing its style and details with words. A new system called LOTS can turn your sketch and text into a realistic fashion image! It uses a special dataset of sketches and descriptions to learn how to combine them accurately.…

TriCLIP-3D is a new computer model that can understand connections between images, text, and 3D shapes all at once! Normally, you would need separate models for each connection, but TriCLIP-3D does it all with one set of instructions. This makes it great at understanding what…

This paper explores how artificial intelligence learns to make decisions in complex environments. It introduces a new technique called RDQ, which helps AI agents learn more effectively by considering both the value of being in a particular state and the value of taking specific…

Scientists created a new tool called Singular Perturbation Decomposition (SPD) to simplify complex artificial intelligence systems called neural networks. Imagine taking apart a complicated machine to understand how each piece works. SPD does something similar by breaking down…

Researchers developed new techniques to analyze music and create better AI music generators. They found that one type of model (Auto-Regression) is best for making high-quality music, while another (Conditional Flow-Matching) is more flexible for editing existing music. Larger…

MiniCPM4 is a new kind of large language model that's really good at using tools and solving problems. It can learn how to use different tools by watching examples, and it's super fast because it's built using special techniques. MiniCPM4 performed better than other similar…

Imagine a supercomputer juggling many tasks at once! This paper shows how powerful AI models called LLMs can help schedule these tasks efficiently, balancing speed, resource usage, fairness, and more without needing special training. While these AI schedulers need some time to…

Researchers have developed new techniques to improve image quality, including making blurry images sharper (super-resolution), brightening dark photos, and removing noise. They also explored methods for combining high-resolution and low-resolution satellite images for clearer…

This new framework helps computers understand images better when they're dealing with different types of pictures. Imagine teaching a computer to recognize cats, but it only saw fluffy Persians before and now has to identify hairless Sphynx cats. This method helps by figuring out…