Google unveils new Gemini Carbon model
Google has announced its new Gemini 4 Carbon model, which it describes as its most powerful model equipped with advanced capabilities in cybersecurity and long-horizon tasks, just two weeks after releasing the Gemini 4 Argon model. The new model topped version 5.0 of the Artificial Analysis intelligence index with 59 points, and will be made available first to a select group of cyber defenders.
A revolution in medicine: DeepMind chief pledges to discover new drugs in months instead of years thanks to artificial intelligence
Demis Hassabis, the mastermind behind AlphaFold and Nobel laureate, announces that artificial intelligence will radically shorten the drug development journey. Are we on the threshold of a new era in medicine, or is the road still long?
Great Sky tests superconducting artificial intelligence chip
American hardware company Great Sky announced the success of a research trial for an artificial intelligence processor relying on superconducting electronic circuits and optical communication to process nuclear fusion data. The tests were conducted in collaboration with researchers from U.S. national laboratories at a processing speed of one million samples per second.
Goodfire researcher affirms that neural networks are easy to interpret
Nick Cammarata, a researcher at AI interpretability lab Goodfire, has put forward a new scientific thesis asserting that reading and interpreting neural networks is much easier than researchers have assumed for many years. Cammarata argues that prominent evidence for the superposition hypothesis has been misunderstood and that neurons grow through organized iterative generalization rather than random chaos.
Microsoft invents smart tool for long-range weather forecasting
Microsoft researchers have published a new machine learning methodology that increases the accuracy of weather forecasts in the two-to-six-week range and excels at flood detection. This coincided with a scientific review of Google DeepMind’s performance during Hurricane Isaias, which showed its superiority in track determination and its failure to estimate wind strength.
DeepMind opens robotics lab in Britain to discover future materials with artificial intelligence
Google DeepMind has announced plans to open the UK’s first fully “automated” research lab, integrating robotics with artificial intelligence to accelerate the discovery of new materials for energy and computer chips.
Google Gemini 4 Pro spotted under an alias on LM Arena
Independent testers have spotted a trial version believed to be Google’s flagship Gemini 4 Pro on LM Arena, disguised under the name Gemini 3.8 Flash. This intriguing appearance follows DeepMind’s confirmation that the model has entered the post-training phase and is nearing its official release to compete with top global models.
Conflict and exchange of accusations between AI lab leaders in Davos
The World Economic Forum in Davos witnessed sharp verbal confrontations between AI company leaders, as the heads of DeepMind and Anthropic criticized OpenAI’s commercial and security strategies amid a charged competitive atmosphere.
Google prepares to launch new Gemini coding model
Google is preparing to launch a new artificial intelligence model with highly advanced programming capabilities, paving the way to bridge the competitive gap in code writing against rivals Anthropic and OpenAI. The new model is named Gemini 3.8 Flash and is known internally as Schemakai, with reliable press reports indicating it could be launched as early as this week.
AI labs race to prevent dangerous bioweapons design
Major artificial intelligence companies are intensifying precautionary efforts to prevent their most advanced models from being used to engineer dangerous viruses or design destructive biological weapons, amid warnings that a single regulatory misstep could lead to a humanitarian catastrophe. Companies such as OpenAI, Anthropic, and Google DeepMind are expanding biological risk assessment testing and implementing strict access controls as generative model capabilities grow.