Rahma

Writing a STRONGLY WORDED LETTER with Gemini Diffusion and its Instant Edit tab

The Future of Development: Gemini Diffusion

Today we introduced Gemini Diffusion⚡️ (& DeepThink, Veo3, Imagen4, 2.5 updates...). It's been a dream of mine to remove the need for "left to right" text generation. It's so fast, that we had to *slow down* the video during the presentation. deepmind.google/models/gemini-…

Such a privilege to work on Gemini Diffusion with an amazing team! From a small research project to launching at I/O - we've got unstoppable aura 🚀 Welcome to the era of live vibe coding ⚡️

🚀Meet Gemini Diffusion, our first diffusion-based and super fast language model, just announced at Google I/O!🚀 Very excited to be able to share what I've been working on for the past little while with our amazing small team @GoogleDeepMind.

Well now, the prospect of using this 'Gemini Diffusion' contraption... frankly, it's a rather delightful thought altogether!

It's been an incredible experience working on Gemini Diffusion. So much pride in what we've accomplished bringing this from a small research project to an I/O launch

We’ve developed Gemini Diffusion: our state-of-the-art text diffusion model. Instead of predicting text directly, it learns to generate outputs by refining noise, step-by-step. This helps it excel at coding and math, where it can iterate over solutions quickly. #GoogleIO

1000+ words per second! ⚡ We just unleashed Gemini Diffusion at #GoogleIO! 🚀 Awesome being part of the team that took this from a small research project all the way to I/O @GoogleDeepMind 🪐

In December we began the Gemini Era, and we’ve continued to make relentless progress since. Today we’re thrilled to introduce the next generation: Gemini 1.5 - hugely enhanced performance, highly efficient architecture & long-context length breakthrough blog.google/technology/ai/…

@rljfutrell @tallinzen Did you see this blog.google/technology/ai/… ?

@tallinzen My experience is that ChatGPT is very bad at this kind of thing, especially for languages with lots of morphology

Very nice idea for a language model benchmark: can the model learn to translate a language (Kalamang, spoken by a small number of people in New Guinea) into English just from a grammar written by a field linguist? arxiv.org/abs/2309.16575

Another tour de force by @ibalazevic, and a thoroughly enjoyable new collaboration with @YugeTen, @pinelopip3, @Rahma_Chaa, and @skandakoppula You can find the paper here arxiv.org/abs/2402.05861

Humans and animals reason about events spanning days, weeks, and years, yet current CV systems live largely in the present. Introducing Memory-Consolidated ViT, whose context extends far into the past and sets a new SOTA in long-video understanding with a 10x smaller model

Read more in a BlackBox EMNLP 2021 paper by @Rahma_Chaa , @robdessi, and myself arxiv.org/abs/2107.01366

Transformer seq2seq models enjoy epic practical success, however, they miserably fail on SCAN data that probes for compositional reasoning.. - wait! compositionality is a cornerstone of the language. Is there a contradiction? Is work on SCAN compositionality even useful then? 🧶\

Wanted: Phd candidate Université de Paris 2021 Meaning-based approach to locality constraints Advisors: Anne Abeillé, Barbara Hemforth, Edward Gibson Please apply and retweet! llf.cnrs.fr/en/node/6836

The way people use words to describe colors is remarkably consistent across thousands of languages. New Facebook AI research published in the “Proceedings of the National Academy of Sciences” shows that cutting-edge AI systems name colors in a similar way: ai.facebook.com/blog/ai-names-…

@TonyBelpaeme @LanguageMIT @SimonKirby Apologies for missing your work. It's most definitely not a deliberate snub. We were not aware of it.

@LanguageMIT @SimonKirby I can't believe they missed our earlier work on colour categorisation under the influence of communication. It almost seems a deliberate snub. Here's two: pubmed.ncbi.nlm.nih.gov/16209771/ and journals.sagepub.com/doi/10.1177/10…

Communicating artificial neural networks develop efficient color-naming systems pnas.org/content/118/12…

We demonstrate next that the nature of the emergent systems depends on communication being discrete. Our study suggests that efficient semantic categorization is a general property of discrete communication systems, not limited to human language.

We show that artificial neural networks trained to play a discrimination game develop emergent communication systems whose distribution on the accuracy/complexity plane closely matches that of human languages.