Google Research AI blog

APRIL 30, 2023

If you’re registered for ICLR 2023, we hope you’ll visit the Google booth to learn more about the exciting work we’re doing across topics spanning representation and reinforcement learning, theory and optimization, social impact, safety and privacy, and applications from generative AI to speech and robotics. Mukund Varma T.

Google 105

Google Language Model Jing 105

The AI Alignment Forum

FEBRUARY 7, 2025

We think this adversarial style of evaluation and iteration is necessary to ensure an AI system has a low probability of catastrophic failure. Wed like to support more such evaluations, especially on scalable oversight protocols like AI debate. Were interested in further work on what factors reinforce or prevent alignment faking.

Research 52

Research Fund Open Technique 52

Stanford AI Lab Blog

DECEMBER 6, 2021

Kochenderfer Contact : philhc@stanford.edu Links: Paper Keywords : deep learning or neural networks, sparsity and feature selection, variational inference, (application) natural language and text processing Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss Authors : Jeff Z.

Contact 40

Contact Learning Theory Authoring Offline 40

The AI Alignment Forum

FEBRUARY 28, 2025

Some relevant criteria for evaluating a specification language include: How expressive is the language? If we had satisfactory answers to these questions, then we would have a much better understanding of which reward learning methods are reliable (if any). What kinds of training data do we need, and how much data do we need?

Research 40

Research Learning Method Policy 40

Stanford AI Lab Blog

APRIL 25, 2022

Larsen, Stanislav Fort, Nic Becker, Surya Ganguli Contact : bwlarsen@stanford.edu Links: Paper Keywords : loss landscape, high-dimensional geometry, random hyperplanes, optimization How did the Model Change?

Contact 40

Contact Award Authoring Language 40

The AI Alignment Forum

FEBRUARY 28, 2025

The Perils of Optimizing Learned Reward Functions: Low Training Error Does Not Guarantee Low Regret In this paper , we look at what happens when a learnt reward function is optimised. However, does this guarantee that we get a low regret relative to the underlying true reward function when that reward model is optimised?

Learning 40

Learning Discussion Classes Policy 40

NonProfit Hub

OCTOBER 16, 2015

Beth has over 30 years working in the nonprofit sector in technology, training, capacity building, evaluation, fundraising and marketing. Beth is an expert in facilitating online and offline peer learning, curriculum development based on traditional adult learning theory and other instructional approaches.

Social 28

Social Social Media Student Cancer 28

Google Research AI blog

NOVEMBER 28, 2022

Platt , Fernando Pereira , Dale Schuurmans Keynote Speakers The Data-Centric Era: How ML is Becoming an Experimental Science Isabelle Guyon The Forward-Forward Algorithm for Training Deep Neural Networks Geoffrey Hinton Outstanding Paper Award Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding Chitwan Saharia , William Chan (..)

Google 52

Google Language Tutorial Offline 52

The AI Alignment Forum

MARCH 28, 2025

And its a small transformer and theres not that much going on, and you train a transformer that in fact works at this, and I guess your job is: how can you prove that it actually works at this without just running it on all of the examples? This is to train the crosscoder? Jason Gross (00:10:33): So it works fine for max of four.

Model 52

Model Network Train Training 52