🔬 AI Research

Academic papers, research breakthroughs, and scientific advances in machine learning and AI.

585 articles

LeCun: recent AI cyberattacks are just a brief transitional phase

Yann LeCun, founder and chair of Advanced Machine Intelligence, says the world will "probably" enter an era of "unprecedented safety of software systems." In his view, today's AI-driven cyberattacks belong to a short period of transition. Source: semafor.com

ML Healthcare Forecasts in India Ignore Historical Bias

New research finds that long-standing inequalities in access to healthcare strongly shape machine learning predictions of healthcare utilization in India. Models trained on past usage data risk carrying those disparities into their projections. Source: google.com

Concrete Strength ML Models Falter on Independent Data

A validation study shows that most machine learning models built to predict concrete strength could not repeat their high accuracy on independent datasets. The findings raise doubts about how well published results hold up beyond the data used to develop the models. Source: google.com

Hybrid ML Streaming Framework Detects Network Intrusions in Under 40 ms

Network security researchers built and tested an end-to-end distributed stream-processing framework that combines hybrid machine learning for intrusion detection. It reaches a median detection time below forty milliseconds. Source: google.com

Google I/O 2026 recap: Gemini 3.5 Flash, Spark agent, Omni video and XR glasses

Google used I/O 2026 to unveil Gemini 3.5 Flash, the Gemini Spark personal agent, Gemini Omni video AI, Samsung XR glasses and a new AI Ultra plan. The event covered a broad range of AI products and hardware. Source: techjournal.org

Google's August 2026 AI roundup: Gemini 3.7 Flash, Pixel 11 and more

Google recapped its biggest AI updates from August, including the launch of Gemini 3.7 Flash, Gemini 3.5 Transcribe and the new Pixel 11 series of devices. The company says it has invested in machine learning and AI research, tools and infrastructure for more than 20 years. Source: blog.google

Isomorphic Labs, founded by ex-DeepMind chief Hassabis, raises $5 billion

Isomorphic Labs, the AI drug discovery startup started by former Google DeepMind head Demis Hassabis, has secured $5 billion in new funding. The round highlights the intense investor appetite for AI-driven pharmaceutical research. Source: cnbc.com

London AI drug firm of DeepMind founder Hassabis now valued at $50bn

Isomorphic Labs, the AI drug discovery company led by Sir Demis Hassabis, has raised $5bn. The deal makes it one of the biggest start-ups in the UK. Source: thetimes.com

Hassabis' Isomorphic Labs lands $5B Series C months after Google exit

Isomorphic Labs closed a $5 billion Series C round on Tuesday. The startup was founded by Demis Hassabis, who previously led Google DeepMind. Source: newsfilter.io

NTT Tests Machine Learning Code with SDK for Optical Quantum Computers

Nippon Telegraph and Telephone (NTT) has tested new machine learning code using an SDK it developed for optical quantum computers. The company confirmed that the code runs on real quantum hardware. Source: google.com

Gemini Apps release notes: latest updates and improvements

Google's release notes for Gemini Apps list the newest changes, including upgrades to generative AI capabilities and wider access to features. Source: gemini.google

Split-and-Scale Approach Advances Latent 3D Shape Representations

Latent 3D representations have greatly broadened what modern shape modeling can do, allowing 3D scenes and objects to be encoded compactly. The new Split-and-Scale work builds on this direction to improve how such representations are structured and scaled. Source: machinelearning.apple.com

MIT Tech Review: Why LLMs Don't Actually Reason

MIT Technology Review's research reports section highlights a popular deep dive in its artificial intelligence coverage, arguing that people shouldn't be fooled into thinking LLMs reason. The piece features the quote "We were looking at it wrong." The snippet is brief and offers no further detail. Source: technologyreview.com

Reinforcement Learning Used to Design Quantum Repeater Protocols

Researchers at Universidad Politécnica de Madrid and Universidad Autónoma de Madrid are applying reinforcement learning to design quantum repeater protocols. The goal is to handle the demands of complex quantum networks. Source: google.com

Decision models promise to make AI hallucinations disappear

Before decision models, teams that wanted to use a classifier had to gather large amounts of data and stand up a machine learning pipeline. The article argues that decision models remove much of that burden and, with it, the problem of hallucination. Source: spectrum.ieee.org

Yann LeCun opens up in a relaxed NTU Singapore interview

Turing Award laureate Yann LeCun sat down for an informal, behind-the-scenes conversation at NTU Singapore. The interview focuses on the personal side of one of AI's pioneers, touching on his life and thinking. Source: m.youtube.com

Hassabis warns of an AI 'oil spill' if development goes unchecked

Google DeepMind co-founder Demis Hassabis cautions that unchecked AI development could end in a major accident. He compares the potential damage to an oil spill. Source: timesnownews.com

Open-source community wants Jensen Huang to come back

The open-source movement is calling for Nvidia's Jensen Huang to return to its side. The piece also cites DePue's claim that his results were verified by senior researchers at OpenAI and Anthropic. Source: technologyreview.com

Hassabis: self-improving AI could slip out of human control

Demis Hassabis warns that unchecked AI self-improvement could produce systems that are hard to control. He fears that without human supervision, such AI could become a serious problem for the industry. Source: indiatoday.in

Trustworthy AI comes from fencing it in, not forecasting its behavior

An argument for building trustworthy AI through hard constraints rather than attempts to predict what a model will do. The piece points to OpenAI's case, where the safeguards lived inside a system the agents were able to edit themselves. Source: technologyreview.com