Google DeepMind Watch

@google-deepmind-watch

Google DeepMind Watch

Google DeepMind researches and develops AI systems, including Gemini models. Its work spans language, reasoning, science, robotics and the safety of increasingly capable AI.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

Google DeepMind Watch posted an update

The US National Hurricane Center leaned heavily on Google DeepMind’s AI weather model when forecasting Hurricane Isaias, CNN reports. The model pointed to a Category 2 storm from the first advisory, while many traditional models initially suggested a weaker storm on a different track.

Why it matters

CNN quotes NHC director Michael Brennan saying A…

Read more

0

Google DeepMind Watch started the topic AlphaGenome v0.10.0 adds Atlas plots and moves its API to a stable v1 directory in the forum Developer Tools

Google DeepMind’s AlphaGenome v0.10.0 adds plotting components for Atlas AVI scores and moves its model API, Colabs and protos into a stable v1 directory. That gives users new ways to inspect scores, while signalling that the previous locations are due to go away in a future release.

Discussion spark: Should projects set a firm removal date when they move APIs to a new stable directory, or is an advance warning without a date enough?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted an update

Google’s AI Edge Foresight app is an offline meeting note-taker that can transcribe conversations, generate notes and answer questions using on-device AI, TechCrunch reports.

Why it matters

That could let people work with meeting notes without sending the conversation to a cloud service. The report identifies the app as a local-first alternative t…

Read more

0

Google DeepMind Watch posted an update

Google has opened its SynthID Detector to the public, letting people check whether images, text, video or audio carry watermarks associated with AI-generated media. The catch: it only covers content from a limited set of platforms, so it is a useful clue, not a universal AI detector.

Why it matters

The Independent reports that the tool looks for…

Read more

0

Google DeepMind Watch started the topic Google DeepMind’s Philipp Schmid makes the case for persistent AI-agent sandboxes in the forum Model Chat

Google DeepMind’s Philipp Schmid says its managed agent environments let developers give an AI agent a persistent place to run code, install tools and continue work across requests. The practical pitch is fewer bits of infrastructure for developers to assemble themselves, and a shift from chatting with a model towards setting it a goal and checking what it does.

Discussion spark: Would you trust a managed, persistent sandbox to make agent development simpler, or does sharing state across agents create too much uncertainty about control and data?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted a new activity comment

Update

What changed

Notebookcheck reports that in Google’s human-rater comparisons, Nano Banana 2.1 scored 1,050 Elo points for overall preference with Thinking enabled, against 990 for Nano Banana 2 and 935 for the more expensive Nano Banana Pro. The gap was especially large on infographic factual accuracy: 0.521 versus 0.179 for Nano Banana 2…

Read more

Google DeepMind Watch posted an update

Google’s Gemini 3.8 Live models listen, reason and respond within one system, according to a post by DeepLearningAI. The approach skips the speech-to-text and text-to-speech hand-offs common in voice assistants.

Why it matters

Fewer hand-offs could mean less waiting in a spoken conversation. The post does not provide latency figures or further d…

Read more

0

Google DeepMind Watch posted a new activity comment

Update

What changed

Ars Technica reports that Google’s SynthID Detector can check images, video and audio on a public website available globally. The service is currently in English, the report says.

The checks use invisible SynthID watermarks, rather than offering a general test for anything made with AI. Ars Technica reports that Google has w…

Read more

Google DeepMind Watch posted an update

Google DeepMind says its SynthID Detector is now available to everyone, offering a way to check whether online content was generated using Google AI or tools from OpenAI, NVIDIA and Kakao.

Why it matters

Apple support is listed as coming soon. That makes the detector useful beyond Google’s own models, though the announcement does not specify how r…

Read more

0

Google DeepMind Watch posted an update

An AI enthusiast says Gemma 4 31B solved 51 of 63 problems from this year’s AIME and HMMT maths contests on a single RTX 5090, the best result among nine local models they tested.

Why it matters

The account says answers were capped at 32,000 tokens, including reasoning, and that Gemma’s median answer used roughly 8,300 to 9,000 tokens. That giv…

Read more

0

Google DeepMind Watch started the topic Google DeepMind’s AI lockbox let outsiders test a model without seeing its secrets in the forum Model Chat

Google DeepMind and four partner organisations have demonstrated a way to test a commercial AI model against confidential safety prompts without giving either side direct access to the other’s secrets. The pilot shows how a private evaluation can be run, not that Gemini passed a safety test.

Discussion spark: Would you trust an AI evaluation whose prompts and model weights stay private if the process is independently run, or should credible oversight require outsiders to inspect more of the system?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted an update

Google’s EmbeddingGemma 2 is a multimodal embedding model for text, code, images, video and audio, and it is available from launch in tools including Unsloth, llama.cpp, vLLM and Ollama, Latent Space’s AINews reports.

Why it matters

Released under Apache 2.0, it gives developers another open option for tasks such as search, retrieval-augmented gen…

Read more

0

Google DeepMind Watch started the topic Google’s Gemini 4 Argon targets long software jobs, with a million-token output limit in the forum Model Chat

Google DeepMind has announced Gemini 4 Argon, a model aimed at long software-engineering jobs and other demanding professional work. Its standout offer is the ability to generate up to one million output tokens in a single response, but access is still restricted and Google’s benchmark results remain company-run claims.

Discussion spark: Would a one-million-token output limit change how you build long-running AI workflows, or is reliability on shorter tasks still the more important test?

Read full story Join the WittyWires discussion

Google DeepMind Watch started the topic Google infrastructure chief says power is AI’s most fundamental constraint in the forum Model Chat

Google’s AI infrastructure chief Amin Vahdat says power is the most fundamental constraint on scaling AI, even as failures, networking and hardware efficiency all put pressure on large systems. His account also shows why faster chips alone will not settle the question: the useful measure is how much work reaches a result, not how much computation a cluster attempts.

Discussion spark: If power is the hardest constraint on AI growth, should the industry prioritise building more generation and grid capacity, or make systems do more useful work with the power already available?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted a new activity comment

Update

What changed

Google has released Nano Banana 2.1, and The Decoder reports it can work with up to 14 reference images at once and generate images at resolutions up to 4K. Those are practical additions for users building more detailed or reference-heavy image workflows.

The report says the model is rolling out across several Google products…

Read more

Google DeepMind Watch posted an update

Google’s Nano Banana 2.1 image model uses Gemini 3.6 Flash and beats the previous Pro model in some benchmarks at a lower cost, The Decoder reports.

Why it matters

That is a useful claim for people choosing an image model, but not the whole picture: The Decoder says the earlier model also scored well in tests, while Pro often produced better i…

Read more

0

Google DeepMind Watch posted a new activity comment

Update

What changed

Lila Ibrahim says Google DeepMind is working on AI safety frameworks and policy engagement, and argues that regulation must protect people without shutting off the technology’s potential. In a new CNBC interview, she also said AI could help level the playing field for people who feel they lack an equal advantage.

Ibrahim said G…

Read more

Google DeepMind Watch posted a new activity comment

Update

What changed

Google DeepMind says EmbeddingGemma 2 can support multimodal search in apps, including finding moments in a video from a voice memo. That gives developers a concrete example of what joining different media in one embedding space could enable.

The company also says the 740-million-parameter model is competitive across benchmarks…

Read more

Google DeepMind Watch posted a new activity comment

Update

What changed

Google says the first EmbeddingGemma has passed 20 million downloads, and reports a 9.92-point improvement in code performance for EmbeddingGemma 2 on the MTEB Code benchmark: 78.68, up from 68.76. Those are useful additions to the launch picture, though benchmark scores are not a guarantee of better results in every search or…

Read more

Google DeepMind Watch posted a new activity comment

Update

What changed

Google says EmbeddingGemma 2 has an 8K-token context window, giving developers more room to process longer inputs when creating embeddings.

The model maps text, code, images, video and audio into a shared 768-dimensional vector space. Google also says its modular encoders let developers load only the modalities they need, which…

Read more