Want to join Google DeepMind as a research scientist working on AI for health and medicine?

Wanna check how well a model can share knowledge between languages? Of course you do! 🤩 But can you do it without access to the model’s weights? Now you can with ECLeKTic 🤯

Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/

.@NitCal will be presenting "Empty Shelves or Lost Keys? Recall Is the Bottleneck for Parametric Factuality" at ICML 2026 next month. (tl;dr: we show encoding is near-saturated on frontier LLMs, but models still struggle to recall encoded facts.) One recurring piece of feedback we've gotten since posting the paper: "you show LLMs struggle with factual recall, but does that even matter when today's agents can use external retrieval?" Here's how I currently think about this, and more broadly about the role of parametric knowledge in today's systems: The theoretical argument for why knowledge matters (true in principle, but I don't know of work that measures this in practice): parametric knowledge is important for making efficient use of search and for knowing how to properly integrate retrieved information. Imagine finding some weird pizza recipe online — can you trust it without knowing a lot about cooking, chemistry, etc.? I think this is going to become a bigger issue moving forward, the more "sloppier" the internet becomes. The realistic case for why knowledge matters: today's agents are far from producing responses that are fully grounded in external evidence. Even when search triggers properly — which it often doesn't — only the "big" claims tend to be grounded, while models still volunteer a lot of extra information from their parametric knowledge. Since models are still poor at "knowing what they know" (more on that in my next post, about our other ICML paper...), our best bet is making models actually more knowledgeable — and our paper reveals where the headroom for that actually lies.

[1/7] Why do frontier LLMs make factual errors? Is it because they never learned the fact… or because they can’t access knowledge they already encoded? In our new paper, we show: The bottleneck is not encoding; it is recall. 🧵👇 Paper: arxiv.org/abs/2602.14080 Many thanks to @_galyo @bd_eyal @zorikgekhman @eran_ofek59358 @GoogleResearch

Hello. How are you? Thank you. I love you. Please. Some of the most frequently translated phrases of the past 20 years! Google Translate began twenty years ago with a mission to help people understand one another, regardless of the language they speak. What started as a small experiment has become a global tool that helps over 1 billion users every month. In that time Translate has evolved from simple pattern matching to true understanding. In 2006, it relied on statistical machine learning to look for patterns in small word clusters. By 2016, we pioneered a shift to neural networks to move beyond literal word-for-word translations, and today we’re using our powerful Gemini models to make Translate even more helpful. We are moving from text to fluid, real-time conversations. With our latest models, you can even use your headphones as a personal interpreter that preserves your original tone and cadence - it’s an amazing experience! One of the interesting things about AI is that as we make progress, we begin to take it for granted. If you met a person who could translate across a hundred languages faster than any human can, you would be so impressed. Today, one product does that for nearly 250 languages, and we kind of just shrug. Being able to say thank you in 250 languages is not something I take for granted. So to the 1 billion who use Google Translate - merci, dhanyavaad, arigatō, gracias, and thank you! Let’s see what the next 20 years will bring.

Donald Knuth is vibemathing now. real tough day for the stochastic-parrot crew.

[1/7] Why do frontier LLMs make factual errors? Is it because they never learned the fact… or because they can’t access knowledge they already encoded? In our new paper, we show: The bottleneck is not encoding; it is recall. 🧵👇 Paper: arxiv.org/abs/2602.14080 Many thanks to @_galyo @bd_eyal @zorikgekhman @eran_ofek59358 @GoogleResearch