Google DeepMind lets us talk to animals!

Just imagine: We could not only eavesdrop on dolphins, but actually understand what they are communicating! Google has made a fascinating breakthrough in animal AI research with DolphinGemma. This specialized AI, a modification of the Gemma model, was developed to analyze the complex clicks, whistles and pulse sequences of Atlantic spotted dolphins and even generate dolphin-like sound sequences.

The model, trained with the Wild Dolphin Project's extensive dataset, uses Google's SoundStream technology and runs directly on Pixel smartphones in the field. Researchers can use it to identify recurring sound patterns and decipher potential meanings - a task that previously required immense human effort.

Particularly exciting: the CHAT system (Cetacean Hearing Augmentation Telemetry) is already researching two-way communication by linking synthetic whistles with objects that dolphins are interested in. Google plans to release DolphinGemma as an open source model this summer.

Could this technology be the key to bridging the gap between human and animal communication? The AI community is at the beginning of an exciting journey of discovery - and soon we could even be talking to our dogs.

Want a byte-sized version of Hacker News? Try TLDR’s free daily newsletter.

TLDR covers the most interesting tech, science, and coding news in just 5 minutes.

No sports, politics, or weather.

Subscribe for free!

Almost 50 Democratic MPs are calling on the Trump administration to stop the use of unauthorized AI systems such as Elon Musk's “Grok”. Particularly critical is the use by the Department of Government Efficiency (DOGE), which uses AI for employee monitoring and data analysis - without official approval. The MPs warn of security risks, legal problems and conflicts of interest, as Musk acts as a government advisor and at the same time contributes his own products.

In Australia's 2025 election campaign, parties are increasingly relying on AI-generated content. The Liberal Party released the first fully AI-generated election commercial, while other candidates are working with AI-powered rap videos and digital campaigns. These developments raise questions about the authenticity of political communication and the potential manipulation of voters by AI.

The former mayor of Detroit, Kwame Kilpatrick, is committed to AI-supported reform of the criminal justice system following his pardon. As a member of the “20% Project”, he advocates the use of open source AI to modernize the pardoning process and reduce mass incarceration. The aim is to use AI to enable fairer and more efficient decisions in the justice system.

These examples show how AI is increasingly intervening in political processes - be it through surveillance, election campaigns or judicial reforms. The challenges lie in regulation, transparency and the protection of democratic principles in the face of rapid technological developments.

NVIDIA has introduced Nemotron-UltraLong-8B, a family of language models capable of processing 1M, 2M, and even 4M-token contexts while maintaining strong benchmark performance. Built on Llama-3.1, the models use efficient continued pretraining and instruction tuning to enhance long-context reasoning and instruction-following. This makes them ideal for handling massive documents and complex conversations without performance trade-offs.