@Google researcher, working on Gemini Safety.
Joined November 2019
- Tweets15
- Following163
- Followers123
- Likes54
Flavien Prost retweeted
Today we are releasing the best open-weights model you can run on a single device reaching 1339 Elo on LMsys for Gemma 3 27B (aka zizou-10)!
Very strong capabilities on math, multilingual, coding, instruction following, function calling !
Gemma 3 is out!
We are focused on bringing you open models with best capabilities while being fast and easy to deploy:
- 27B lands an ELO of 1338, all the while still fitting on 1 single H100!
- vision support to process mixed image/video/text content
- extended context window of 128k
- broad language support
- function call / tool use for agentic workflows
Try it out at aistudio.google.com/prompts/…
Flavien Prost retweeted
Very excited about this tutorial at #AAAI2025 on inducing privacy, fairness or robustness to distribution shifts when data is imperfect (e.g. unlabeled, noisy)! You can check out the slides at amartya18x.github.io/files/T…
Very shortly at @RealAAAI , @alexandrutifrea and I will be giving a Tutorial on the impact of Quality and availability of labels and data for Privacy, Fairness, and Robustness of ML algorithms
See here amartya18x.github.io/files/T…
@MLSectionUCPH @DIKU_Institut @GoogleDeepMind
Flavien Prost retweeted
Inference-time procedures (e.g. Best-of-N, CoT) have been instrumental to recent development of LLMs. The standard RLHF framework focuses only on improving the trained model. This creates a train/inference mismatch.
Can we align our model to better suit a given inference-time procedure?
We answer this affirmatively, check out the thread below.
Flavien Prost retweeted
Excited to share 𝐈𝐧𝐟𝐀𝐥𝐢𝐠𝐧!
Alignment optimization objective implicitly assumes 𝘴𝘢𝘮𝘱𝘭𝘪𝘯𝘨 from the resulting aligned model. But we are increasingly using different and sometimes sophisticated inference-time compute algorithms.
How to resolve this discrepancy?🧵
ALT InfAlign: Inference-aware language model alignment Ananth Balashankar, Ziteng Sun, Jonathan Berant, Jacob Eisenstein, Michael Collins, Adrian Hutter, Jong Lee, Chirag Nagpal, Flavien Prost, Aradhana Sinha, Ananda Theertha Suresh, Ahmad Beirami
Come and join us! We are looking for great and passionate people who want to improve Gemini Safety!
My team at @GoogleDeepMind in Zurich is hiring. If you are passionate about safety, and the opportunity to actually do pre-training on Gemini scale models (!!) excites you then this is for you!
1
112
Flavien Prost retweeted
Come to the poster and check out the paper if you'd like to learn more! arxiv.org/abs/2312.02592
This is joint work with amazing collaborators at @GoogleDeepMind: @PreethiLahoti, @packer_ben, Yoni Halpern, @abeirami, and @FlavienProst.
Flavien Prost retweeted
I am very happy to announce that Gemma 1.1 Instruct 2B and “7B” are out! Here are a few details about the new models:
1/11
Flavien Prost retweeted
Finally, we've just published on arXiv my work during an internship in the Responsible AI team@Google Research (joint work with @PreethiLahoti, @packer_ben, Yoni Halpern, @abeirami, @FlavienProst). Let's chat, if you're interested in (post-proc) mitigations for group fairness!
Flavien Prost retweeted
⚠️#Internship2024 Alert⚠️
Are you a 𝐏𝐡𝐃 𝐬𝐭𝐮𝐝𝐞𝐧𝐭 excited about #ResponsibleAI in foundation models? Do you have experience training/evaluating them?
Team members: @PreethiLahoti @infoxiao @ananthbshankar @sroy_subhrajit @aradsinha @YaoQinUCSD (RADML at @GoogleResearch)
Flavien Prost retweeted
[Call for papers]
#NeurIPS2023 R0-FoMo Workshop
Robustness of Zero/Few-shot Learning in Foundation Models We solicit novel contributions that relate broadly to zero/few-shot learning in foundation models, with both empirical and theoretical nature.
Please R/T
Flavien Prost retweeted
Announcing the release of MinDiff, a new regularization technique available in the @TensorFlow Model Remediation library for effectively and efficiently mitigating unfair biases when training #MachineLearning models. Learn more below: goo.gle/3f3Qacu
Flavien Prost retweeted
Super happy to share that our paper “Fairness without Demographics through Adversarially Reweighted Learning" has been accepted at #NeurIPS2020.
A big thanks to all my co-authors, reviewers, and colleagues at @GoogleAI, #MPI-INF and #MPI-SWS for their valuable feedback!
How can we train ML models to achieve fairness when we do not know protected group memberships? Excited to share a new paper on this (arxiv.org/abs/2006.13114) from my internship @GoogleAI. Joint work with @alexbeutel @edchi @chenjilin @Nithum @FlavienProst and others. 1/n
Flavien Prost retweeted
Excited to announce TensorFlow Recommenders, just in time for #RecSys2020!
☝ We have a recommendation for that
Introducing TensorFlow Recommenders, an open-source package that makes building, evaluating, and serving recommender models easy. Find recommendations for movies, restaurants, and much more!
Get started → goo.gle/2HrCwDq
This work focuses on improving the model performance on worst case groups when no demographic information are available! It is achieved via an adversarial re-weighting learner which identifies computationally identifiable subgroups that underperform.
How can we train ML models to achieve fairness when we do not know protected group memberships? Excited to share a new paper on this (arxiv.org/abs/2006.13114) from my internship @GoogleAI. Joint work with @alexbeutel @edchi @chenjilin @Nithum @FlavienProst and others. 1/n
3
Check out our poster at #neurIPS19 in the workshop "ML with Guarantees".
We will describe how Maximum Mean Discrepancy can improve the trade-off between performance and fairness in production systems (arxiv.org/abs/1910.11779).