🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

🚀 We fixed a major LLM post-training bottleneck! Our new method (TBA) combines trajectory balance with asynchronous training to speed up LLM RL 5-50x while improving results+scalability. For example, using VinePPO's GSM8K setup, we obtain +1.2% accuracy and 50x faster RL.

🚀 really excited by the multimodal reasoning capabilities of this model!

This is really cool!

Google DeepMind India is hiring for research scientist role in multicultural & multimodal modeling. Strong candidates with proven research experience are encouraged to apply I shall be at #icassp2025 Hyderabad on Apr 8, happy to meet and chat, pls DM job-boards.greenhouse.io/deepmind/jobs/…

Great opportunity for MS/PhD students working on speech/language. Sumanth and team work on cool stuff and are really fun!!

You write the script, Veo 2 brings it to life. 🎥 Starting today, Google Gemini App Advanced users can create stunning 8-second videos, in 720p cinematic quality, with just one text prompt. ✨

Touching down in Singapore attending #ICLR2025 this week. If you're into Multimodal modelling, visual reasoning, RL for diffusion, multiculturality or anything remotely close - let's link up!! Always up for good discussions, food recs and better coffee 🌊🏖️🍛☕️

Super intriguing panel discussion at the SSI workshop Yoshua Bengio: The naming of the workshop — is genuinely worrisome. Appreciate the spice and insights, Noah! #ICLR2025

Watch Gemini 2.5 Pro implement a landmark Google DeepMind research paper. 🕹️ It codes the reinforcement learning algorithm, visualizes the training live and even debugs errors. ↓

Proactive T2I Agents is accepted to #ICML2025 🚀 We show it is possible to significantly improve multimodal generation (atleast 2x gains) with multi-turn agents that act based on uncertainty principles.

Our work on proactive text-to-image agents – enabling them to ask for clarification and share their understanding – has been accepted at #ICML2025! So proud of the team's efforts. See you in Vancouver! 🙌

Ahead of I/O, we’re releasing an updated Gemini 2.5 Pro! It’s now #1 on WebDevArena leaderboard, breaking the 1400 ELO barrier! 🥇 Our most advanced coding model yet, with stronger performance on code transformation & editing. Excited to build drastic agents on top of this!

Gemini-2.5-Pro-preview-05-06 isn't only the best model for coding. its the best model for everything! across all tasks!

Physics physics physics! #veo3

Oh oh, they found out it can rap :)

The Veo3 effect on traffic to Google DeepMind. Have you tried it yet?

Abhipsa Basu Venkatesh Babu Danish Pruthi Simran Khanuja Nithish Kannen 🚨 I'll be giving a short contributed talk about CuRe at CVPR 's DemoDiv workshop today (06/11) at 10 AM! Please drop by, I'd love to chat 😁 x.com/polkirichenko/…

🚨 New Google DeepMind paper 𝐑𝐨𝐛𝐮𝐬𝐭 𝐑𝐞𝐰𝐚𝐫𝐝 𝐌𝐨𝐝𝐞𝐥𝐢𝐧𝐠 𝐯𝐢𝐚 𝐂𝐚𝐮𝐬𝐚𝐥 𝐑𝐮𝐛𝐫𝐢𝐜𝐬 📑 👉 arxiv.org/abs/2506.16507 We tackle reward hacking—when RMs latch onto spurious cues (e.g. length, style) instead of true quality. #RLAIF #CausalInference 🧵⬇️