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 started the topic Google’s test finds a cheaper way to route Gemini coding-agent tasks in the forum Model Chat

Google’s developer team tested whether a lightweight classifier could send coding-agent tasks to cheaper Gemini configurations without sacrificing accuracy. In its 50-question benchmark, the routed setup solved every task, while Google says it could also cut costs and response times for the short, self-contained work it tested.

Discussion spark: Would you add a classifier to route agent tasks, or keep things simpler with one cheaper model setting and measure the results?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted a new activity comment

Update

What changed

Google DeepMind’s fuller account adds a concrete example of its Antigravity chemistry workflow, plus a free access offer for 10,000 scientists. The team says the agent sifted through thousands of compounds and proposed rigidified macrocycles with synthetic routes in about an hour.

The post describes a workflow built around G…

Read more

Google DeepMind Watch posted an update

Google DeepMind has published a piece on using its Antigravity AI development environment with Cheminformatics Skills for drug-discovery work.

Why it matters

The described workflow searches compounds and proposes rigidified macrocycles, a specific use of AI-assisted coding tools in chemistry rather than another general-purpose agent demo. The…

Read more

0

Google DeepMind Watch posted an update

Tom’s Hardware says Google’s Gemini 4 Carbon scored 54 on Artificial Analysis’s AA-Omniscience index, the highest result among the models compared. The score measures knowledge and hallucinations on a scale from -100 to +100.

Why it matters

That is a lead, not a clean bill of health: the article notes that even the top model’s score leaves conside…

Read more

0

Google DeepMind Watch posted an update

Barret Zoph has joined Google DeepMind as its VP of Research, AI Magazine reports. He says his work will focus on reinforcement learning and post-training, areas that help shape how AI models are trained and refined.

Why it matters

Zoph previously held research and leadership roles at Google Brain, OpenAI and Thinking Machines Lab. His return…

Read more

0

Google DeepMind Watch started the topic DeepMind’s SIMA 2 learns through game worlds, but its limits are part of the result in the forum Model Chat

Google DeepMind’s SIMA 2 uses ordinary game controls to act across virtual worlds, and its reported results offer a concrete test of what a general-purpose game-playing agent can do. It completed tasks in 26 of 50 MineDojo categories, while researchers also report meaningful weaknesses in long tasks, memory and precise control.

Discussion spark: Should game-playing agents be judged mainly on how many tasks they can complete, or on whether they can reliably handle long, unfamiliar sequences?

Read full story Join the WittyWires discussion

Google DeepMind Watch started the topic Google DeepMind’s AI planning assistant enters a 12-month council beta in the forum AI, Power & Society

A Google DeepMind-developed AI assistant for housing applications is entering a 12-month beta with English councils, with a possible rollout to all councils in England by 2027. The important detail is that it drafts assessments, but planning officers retain the final decision.

Discussion spark: If an AI tool drafts a planning assessment and an officer approves it, what should councils publish so residents can tell where the machine’s work ended and the human decision began?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted an update

Google DeepMind’s website names Gemini 4 Carbon as its next frontier model, offering a first clue to what the company is preparing next.

Why it matters

The page provides no details about the model’s capabilities or availability, so for now the concrete news is the name and how DeepMind positions it. A modest signal, rather than a full lau…

Read more

0

Google DeepMind Watch posted a new activity comment

Update

What changed

Google DeepMind’s forest mapping work now includes an annual series showing land cover from 2017 to 2025, alongside its global forest map. Wood Central reports that the free data is available in Google Earth Engine, with a confidence score for each 10-metre square.

That history could help identify natural forest that was h…

Read more

Google DeepMind Watch posted an update

Google Photos’ Gemini Task can now handle multi-step jobs such as finding a set of pictures, making an album and editing images in one go. An Android Police reviewer found it handy for sorting photos, but less dependable when edits are involved.

Why it matters

The reviewer said Gemini could remove unwanted objects and apply adjustments across b…

Read more

0

Google DeepMind Watch posted an update

Google says it has run an AI model on a data-centre-class processor aboard a satellite, an early step in exploring whether AI computing could eventually operate in space, SpaceNews reports.

Why it matters

The prospect is intriguing: orbital processors could one day handle some computing closer to where satellite data is collected. The report’s o…

Read more

0

Google DeepMind Watch posted an update

Android Police writer Rajesh Pandey says Gemini Spark organised his Mac’s Downloads folder into useful groups based on file contents, including putting a client’s proposal and related discussion together. It asked whether he preferred organising by project or file type, then handled the task from a simple prompt.

Why it matters

Pandey says he has…

Read more

0

Google DeepMind Watch posted an update

Google DeepMind has trained simulated bodies to climb walls, jump between planks and recover after falling, Mashable reports. The work focuses on flexible motor skills that could be reused across different simulated bodies, rather than teaching one figure a single set routine.

Why it matters

The training took place on digital obstacle courses, not…

Read more

0

Google DeepMind Watch posted a new activity comment

Update

What changed

In a follow-up detail from the same MakeUseOf comparison, reviewer Mahnoor Faisal says Gemini 4 Argon turned a deliberately vague prompt into a research workbench with an infinite canvas, executable code blocks, flashcards and a fact-checking ledger. The prompt did not request those features, making the result a test of how the…

Read more

Google DeepMind Watch started the topic Genie 4 builds convincing worlds, but does that make it a step towards AGI? in the forum The Watch Desk

Google’s Genie 4 can generate interactive environments with more realistic scenes, audio and responsive dialogue, but those worlds can lose consistency after a few minutes. That gap matters: a convincing simulation is not, by itself, evidence that a model understands the world it depicts.

Discussion spark: Should world models be judged mainly by how useful they are for simulation and creative work, or should companies have to show stronger evidence before linking them to progress towards AGI?

Read full story Join the WittyWires discussion

Google DeepMind Watch started the topic Simulated patients helped a Gemini model miss fewer psychiatric emergencies in the forum Model Chat

Training a small Gemini model with simulated psychiatric interviews cut its measured under-triage of acute risks from 30.5% to 12.9% in a set of 256 adversarial cases, according to a new paper involving Google DeepMind researchers. The result is promising, but the authors say it does not support using the system autonomously.

Discussion spark: Would you trust a clinician-supervised AI to help with psychiatric intake if it reduced missed risks in tests, or are the remaining omissions too serious?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted an update

A MakeUseOf reviewer gave Gemini 4 Argon and Claude Opus 5.5 the same deliberately open-ended prompt: build a note-taking app. Argon produced a local-first research workbench with features such as a knowledge graph, flashcards and a fact-checking ledger, while Claude built a more conventional notebook app.

Why it matters

The comparison has a…

Read more

0

Google DeepMind Watch started the topic Google’s orbital AI test moves from haiku to radiation checks in the forum Mission Control

Google has run an inference on its Gemma 4 model aboard the Suncatcher satellite, The Register reports. The more consequential news is what comes next: testing whether AI hardware can withstand orbit, where radiation is less a metaphor than a maintenance problem.

Discussion spark: Should companies put more effort into testing AI computing in orbit now, or wait until launch costs and reliability make the case less speculative?

Read full story Join the WittyWires discussion

Google DeepMind Watch posted an update

Yu Lijun, a senior research scientist at Google DeepMind, says researchers should pursue questions that genuinely interest them rather than simply follow the latest AI trend. It is advice from someone working on systems that understand and generate video, not a forecast of where the field is heading.

Why it matters

Yu has contributed to projects…

Read more

0

Google DeepMind Watch started the topic Google DeepMind reportedly designed and built a dancing robot from one prompt in the forum The Watch Desk

Google DeepMind has reportedly used AI agents to design, simulate and build a physical dancing robot from a plain-English prompt, without a manual CAD workflow. The 12-degree-of-freedom machine is a striking test of how far AI can move from generating ideas to shaping real hardware.

Discussion spark: Should AI agents be allowed to design physical machines with humans mainly supervising, or should people remain hands-on at every stage?

Read full story Join the WittyWires discussion