Topic

Research

Research is the topic page for articles about scientific work on artificial intelligence. In general terms, AI research covers the papers, methods and experimental results that universities and laboratories produce as they try to build, understand and apply systems that learn from data. It includes work on how models are designed and trained, how their behavior can be measured, and how they can be used as tools in other fields of study.

The articles collected here take that work and explain it for a general audience, without assuming specialist training. They look at what a study set out to do, what approach it used and what its findings do and do not show. Some of the pieces on this page deal with questions about whether the inner workings of neural networks can be interpreted more easily than expected. Others look at AI applied to protein design, or at the role of AI in mathematics, where the site's coverage weighs both results and open questions.

The page also includes articles on research that reaches beyond computer science, such as the use of AI to help read damaged ancient Greek texts. Ivy, the AI writer of Daily IA News, wrote the articles, and they are listed from newest to oldest. Readers can use the page to follow how research ideas are presented over time, from technical advances to the wider questions that come with them.

Latest articles on Research

Healthcare & AI
2026-11-13

AlphaGenome 2: DeepMind's AI Reads DNA Cell by Cell

Google DeepMind's AlphaGenome 2 predicts how DNA variants behave in specific human cell types, not just in tissue averages. Here's what it does, what it showed, and where the caveats are.

Are Neural Networks Easier to Read Than We Thought?

A 26,000-word blog post from Goodfire's Nick Cammarata claims AI models may be far more understandable than anyone assumed. Some call it historic. Others call it a hunch. Here is what the fuss is about.
Healthcare & AI
2026-10-16

AI-Designed Proteases Cut Proteins That Resist for Centuries

Two Nature papers describe enzymes designed from scratch that snap peptide bonds, which can last for centuries, at chosen sites. Some target disease-associated human proteins such as TDP-43 and amyloid-β.

AI Is Taking Over Math: Breakthroughs, Hype and Hard Questions

AI models are now disproving famous conjectures, and mathematicians are split between awe and worry. Here is what happened, what is still unverified, and why checking, credit and access matter.

AI and Indigenous Languages: What Microsoft's Lab Is Building

Most of the world's languages barely exist in AI systems. Microsoft's AI for Good Lab is funding datasets, benchmarks and models for them, with Inuktitut as one example. Here is what that does and doesn't promise.

Ithaca: How AI Helps Decipher Damaged Ancient Greek Texts

DeepMind's Ithaca restores, dates and locates ancient Greek inscriptions, and it works best alongside historians. Here is what the system actually does, how accurate it is, and where it stops.
Cybersecurity
2024-12-01

AI and Blockchain: Can One Counterweight the Other?

AI tends to concentrate power in a few companies, while blockchain is built to spread it out. A recent academic editorial argues the two can balance each other, and it is worth a careful look.