Nenad Tomasev
@weballergy
Senior staff research scientist at DeepMind. Opinions are my own. Re-tweets and favorites not to be considered as endorsements.
Amazing summary of our recent work on physician oversight for diagnostic AI by @rohanpaul_ai!
Another great @Google paper. 💊 AI crushed the usual clinic bottlenecks here because the team wrapped the model in tight medical guardrails, let real physicians watch everything it said, and then proved the setup works with a head‑to‑head simulation against human residents. 🩺…
Our Aeneas AI model gives historians valuable new insights into ancient inscriptions & ancient history that may have taken years to uncover otherwise. Published in @Nature today: deepmind.google/discover/blog/…
Asked the Head of AI at a Fortune 500 how he decides buy vs build, and his framework feels widely applicable: “we buy what accelerates us, and build what differentiates us”
New paper out on asynchronous physician oversight for AMIE, our diagnostic AI 🧵
AMIE can now conduct medical dialogues with patients within specified safety guardrails. This advance allows satisfying safety constraints such as abstaining from individualized medical advice while letting AMIE perform the crucial task of information acquisition (“history…
We are excited to share our latest pre-print enabling effective human oversight for AMIE, our research diagnostic dialogue AI. We introduce a new asynchronous oversight paradigm, decoupling history-taking by AMIE from sharing a human-approved diagnosis – a thread 🧵:
Thrilled to finally release this study! 🚀 We view (discrete) diffusion models as implicitly doing data augmentation over autoregressive. Through this lens, we find that diffusion outperforms AR in data-constrained settings, but it requires larger models and way more epochs to…
🚨 The era of infinite internet data is ending, So we ask: 👉 What’s the right generative modelling objective when data—not compute—is the bottleneck? TL;DR: ▶️Compute-constrained? Train Autoregressive models ▶️Data-constrained? Train Diffusion models Get ready for 🤿 1/n
We are excited to share our latest pre-print enabling effective human oversight for AMIE, our research diagnostic dialogue AI. We introduce a new asynchronous oversight paradigm, decoupling history-taking by AMIE from sharing a human-approved diagnosis – a thread 🧵:
New work on making AMIE safer and better! Oversight by physicians provides guardrails for AMIE and allows clinicians to review, edit, and then release diagnoses using a purpose-built UI.
We are excited to share our latest pre-print enabling effective human oversight for AMIE, our research diagnostic dialogue AI. We introduce a new asynchronous oversight paradigm, decoupling history-taking by AMIE from sharing a human-approved diagnosis – a thread 🧵:
If advanced AI agents and assistants are to be used in impactful applications, it is imperative to develop reliable mechanisms for asynchronous oversight by human experts. This study implemented and evaluated one such approach for AMIE, in the context of medical dialogue.
We are excited to share our latest pre-print enabling effective human oversight for AMIE, our research diagnostic dialogue AI. We introduce a new asynchronous oversight paradigm, decoupling history-taking by AMIE from sharing a human-approved diagnosis – a thread 🧵:
Introducing our latest milestone on conversational diagnostic AI g-AMIE arxiv.org/abs/2507.15743 To harness AI's abundant potential in diagnosis and treatment, we need ways for licensed professionals to oversee and own responsibility for safety-critical decisions. In clinical…
From Silver to Gold🥇 Our advanced version of Gemini Deep Think has achieved the gold-medal standard at the 2025 International Mathematical Olympiad. It perfectly solved 5/6 problems in natural language with no expert intervention and that too within 4.5 hours!
An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇 It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵
Drastic progress on maths with Gemini 2.5! As a math undergrad, I am impressed 🤯 🥈 -> 🥇 ✅ Formal -> Informal ✅ Specialized model -> General model ✅ Available soon ✅ Huge thanks to IMO and congrats to all participants! Blog: deepmind.google/discover/blog/…
Calling researchers at the frontier of quantum biology and neuroscience: Google’s Academic Research Awards are accepting proposals in Quantum Neuroscience through July 23rd. Learn more and apply here → goo.gle/455sLDb
Foundational research powers the future. Through the Google Academic Research Awards, we’re supporting the next wave of scientists and engineers. Apply by 7/23 → goo.gle/4eSlytr
Exciting news for students in India🇮🇳: get your free @GeminiApp Pro plan for 1 year! This gives you higher rate access to all our best models: 2.5 Pro, Veo 3, Deep Research, NotebookLM, and 2TB storage. Claim it at goo.gle/freepro - enjoy!
If you’re a student in India - you’ve just been granted access to a FREE Gemini upgrade worth ₹19,500 for one year 🥳✨ Claim and get free access to Veo 3, Gemini in Google apps, and 2TB storage 🔗 goo.gle/freepro. @GeminiApp
Introducing Concordia 2.0, an update to our library for building multi-actor LLM simulations!! 🚀 We view multi-actor generative AI as a game engine. The new version is built on a flexible Entity-Component architecture, inspired by modern game development.
Check out our state-of-the-art open weights MedGemma multimodal model for making sense of longitudinal EHR data as well as medical text and medical imaging data in various modalities (radiology, dermatology, pathology, ophthalmology, etc.) See the blog post linked below! ⬇️
Introducing new models for research & development of health applications: MedGemma 27B Multimodal, for complex multimodal & longitudinal EHR interpretation, and MedSigLIP, a lightweight image & text encoder for classification, search, & related tasks. → goo.gle/4kvt6Uk
We are expanding our team working at the forefront of AI for medicine @GoogleDeepMind Our mission is to help democratize and make world class healthcare available to everyone, everywhere. If you are a researcher or engineer passionate about AI and healthcare, please consider…
How might AI supercharge world-class expertise in medicine for everyone, everywhere? We’re privileged to pursue this mission @GoogleDeepMind with incredible teammates and are growing our team. We’re hiring stellar Software Engineers, Research Engineers, and Research Scientists…
New research presents a complete algorithm for ground-state energy estimation, addressing the challenge of imperfect initial state preparation. We detail more efficient matrix-product-state (MPS) preparation and improved energy sampling methods → goo.gle/4dejcUP