Google Research AI blog

JANUARY 18, 2023

I will begin with a discussion of language, computer vision, multi-modal models, and generative machine learning models. Language Models The progress on larger and more powerful language models has been one of the most exciting areas of machine learning (ML) research over the last decade. Let’s get started!

Language 132

Language Model Generation Research 132

Google Research AI blog

JUNE 2, 2023

Building audiovisual datasets for training AV-ASR models, however, is challenging. In contrast, the models themselves are typically large and consist of both visual and audio encoders, and so they tend to overfit on these small datasets. LibriSpeech ). LibriSpeech ). Unconstrained audiovisual speech recognition.

Model 103

Model Avatar Audio Phase 103

Google Research AI blog

FEBRUARY 23, 2023

Pre-training on diverse datasets has proven to enable data-efficient fine-tuning for individual downstream tasks in natural language processing (NLP) and vision problems. So, we ask the question: Can we enable similar pre-training to accelerate RL methods and create a general-purpose “backbone” for efficient RL across various tasks?

Offline 97

Offline Train Training Learning 97

Forum One

NOVEMBER 26, 2024

In fact, training a single advanced AI model can generate carbon emissions comparable to the lifetime emissions of a car. And with the rapid advancement of generative AI models potentially slowing down , this provides a unique opportunity to take a breath and reimagine and mature our approach.

Digital 104

Digital Impact United States Integration 104

TechCrunch

AUGUST 15, 2023

OpenAI claims that it’s developed a way to use GPT-4 , its flagship generative AI model, for content moderation — lightening the burden on human teams. “We can repeat [these steps] until we’re satisfied with the policy quality.” ” But color me skeptical. Has OpenAI solved this problem?

Proposal 98

Proposal Content Policy Model 98

Beth's Blog: How Nonprofits Can Use Social Media

MARCH 25, 2014

Earlier this month, I presented on a panel called “ Learn, You Will ” with Cindy Leonard, John Kenyon, and Andrea Berry on the topic of designing effective nonprofit technology training at the Nonprofit Technology Conference hosted by NTEN. 6 Tips for Evaluating Your Training Session.

Train 101

Train Training Reflection Resource 101

sgEngage

DECEMBER 5, 2024

It may feel intimidating at first, but here’s the exciting part: today, more than ever, nonprofits have the tools and resources to make a smooth shift to the grants-plus-fundraising model. Adding fundraising to your funding model gives you the agility to stay mission-focused no matter what comes your way.

Grant 98

Grant Fundraising Fund Nonprofit 98

Nonprofit Tech for Good

JULY 15, 2024

They forget that training, equipment, and hiring resources also contribute to the cost. While this is understandable, a void of guidance and official policy at the top of the organization leads to employees taking things into their own hands and using AI tools without proper transparency and training.

Nonprofit 291

Nonprofit Language Donor Tools 291

Google Research AI blog

NOVEMBER 8, 2022

Posted by Shunyu Yao, Student Researcher, and Yuan Cao, Research Scientist, Google Research, Brain Team Recent advances have expanded the applicability of language models (LM) to downstream tasks. On the other hand, recent work uses pre-trained language models for planning and acting in various interactive environments (e.g.,

Language 90

Language Model Sample Wikipedia 90

TechCrunch

JULY 13, 2022

Companies face several hurdles in creating text-, audio- and image-analyzing AI models for deployment across their apps and services. Cost is an outsize one — training a single model on commercial hardware can cost tens of thousands of dollars, if not more. Geifman proposes neural architecture search (NAS) as a solution.

Model 99

Model Tech Images Poll 99

VentureBeat

MARCH 13, 2025

Carnegie Mellon University researchers propose a new LLM training technique that gives developers more control over chain-of-thought length. Read More

Technique 128

Technique Proposal Help University 128

Google Research AI blog

MAY 23, 2023

As part of this process, the reviewer inspects the proposed code and asks the author for code changes through comments written in natural language. However, with machine learning (ML), we have an opportunity to automate and streamline the code review process, e.g., by proposing code changes based on a comment’s text.

Comment 140

Comment Review Model Authoring 140

Google Research AI blog

MAY 12, 2023

Recent vision and language models (VLMs), such as CLIP , have demonstrated improved open-vocabulary visual recognition capabilities through learning from Internet-scale image-text pairs. The category text embeddings are obtained by feeding the category names through the text model of pretrained VLM (which has both image and text models)r.

Language 96

Language Model Open Train 96

Google Research AI blog

FEBRUARY 7, 2023

In the last 10 years, AI and ML models have become bigger and more sophisticated — they’re deeper, more complex, with more parameters, and trained on much more data, resulting in some of the most transformative outcomes in the history of machine learning.

Research 118

Research Learning Google Sample 118

Get Fully Funded

FEBRUARY 10, 2025

The Letter of Inquiry is part of the grantmaking process because it would be overwhelming and time-consuming for funders to review 15-page proposals from everyone seeking money. Due to space limitations, it’s better to wait to include photos for the invitation to submit a full proposal. That would simple take too long.

Grant 98

Grant Template Children Foundation 98

The AI Alignment Forum

APRIL 14, 2025

The trajectory is based on a sequence of five fictional models (M1-M5) with progressively advanced capabilities. For each model, we define an AI Control Level (ACL) based primarily on its threat model-specific capabilities.

Evaluation 52

Evaluation Measure Research Model 52

The Verge

APRIL 21, 2021

A bipartisan group of lawmakers has proposed banning police from buying access to user data from data brokers, including ones that “illegitimately obtained” their records — like, its sponsors say, the facial recognition service Clearview AI. Photo by Caroline Brehman/CQ-Roll Call, Inc via Getty Images.

Proposal 85

Proposal Accessibility Data Sponsor 85

Google Research AI blog

AUGUST 28, 2023

However, modern object detectors are limited by the manual annotations of their training data, resulting in a vocabulary size significantly smaller than the vast array of objects encountered in reality. Thus, it could be beneficial for open-vocabulary detection if we build locality information into the image-text pre-training.

Train 71

Train Training Awareness Open 71

Google Research AI blog

APRIL 6, 2023

BayesOpt works by repeatedly constructing a surrogate model of the black-box function and strategically evaluating the function at the most promising or informative input location, given the information observed so far. The Gaussian process model constructs a probability distribution over possible functions.

Train 72

Train Training Process Evaluation 72

TechCrunch

APRIL 21, 2023

The growing compute power necessary to train sophisticated AI models such as OpenAI’s ChatGPT might eventually run up against a wall with mainstream chip technologies. CNBC, speaking to analysts and technologists, estimates the current cost of training a ChatGPT-like model from scratch to be over $4 million.

Train 110

Train Training Proposal Analysis 110

Candid

NOVEMBER 14, 2024

By examining details about these projects, such as the project title and abstract (if available), the award amount, the location of the funded organization, and other relevant metrics, nonprofits can tailor their grant proposals to that foundation. Those texts can be mined and used to improve the model for everyone.

Grant 59

Grant Search Proposal Award 59

Association Analytics

JUNE 21, 2023

Older models tended to be AMS-centric, leading to siloed data, static reports, and that trapped feeling. Provide training opportunities for employees and review your taxonomy from time to time to update any new processes. You may also consider training for all employees to become more familiar with how your association uses data.

Associations 319

Associations Challenge Data Taxonomy 319

Candid

MAY 7, 2024

A narrative that illustrates your nonprofit’s program impact The logic model is a classic framework for illustrating a program’s impact. A proposal focused on program impact includes: Need: What need does your program serve? These inputs are discussed in the project description and budget sections of your proposal.

Proposal 64

Proposal Grant Story Impact 64

DataRobot

APRIL 21, 2022

This increase in accuracy is important to make AI applications good enough for production , but there has been an explosion in the size of these models. It is safe to say that the accuracy hasn’t been linearly increasing with the size of the model. They define it as “buying” stronger results by just throwing more compute at the model.

Green 145

Green Environment Metrics Measure 145

Google Research AI blog

APRIL 26, 2023

Posted by Shekoofeh Azizi, Senior Research Scientist, and Laura Culp, Senior Research Engineer, Google Research Despite recent progress in the field of medical artificial intelligence (AI), most existing models are narrow , single-task systems that require large quantities of labeled data to train.

Images 130

Images Foundation Train Training 130

TechCrunch

APRIL 12, 2021

Contracts and briefs in legal work, leases and agreements in real estate, proposals and releases in marketing, medical charts, etc, etc. At the core of XML was the idea that a document should be structured almost like a webpage: boxes within boxes, each clearly defined by metadata — a hierarchical model more easily understood by computers.

Model 108

Model NSF Grant Doc 108

Amy Sample Ward

FEBRUARY 28, 2011

There are a few core elements of the way communities work that we can learn from as a model for innovation as well. Community-Driven Model. But in a community-driven model, you can’t just listen for the sake of learning. What are the technologies that support community-driven engagement and models of working? Principles.

Library 241

Library Community Collaboration Model 241

Google Research AI blog

FEBRUARY 8, 2023

In most scenarios users have a limited labeling budget, and sometimes there aren’t even any labeled samples during training. In “ SPADE: Semi-supervised Anomaly Detection under Distribution Mismatch ”, we propose a novel semi-supervised AD framework that yields robust performance even under distribution mismatch with limited labeled samples.

Data 112

Data Ratio Sample Train 112

Google Research AI blog

JULY 6, 2023

Collecting such labeled data is costly, and models trained on this data are often tailored to a specific use case, limiting their ability to generalize to different datasets. To address these challenges, in “ Pic2Word: Mapping Pictures to Words for Zero-shot Composed Image Retrieval ”, we propose a task called zero-shot CIR (ZS-CIR).

Map 85

Map Picture Images Proposal 85

Google Research AI blog

MAY 16, 2023

While large language models (LLMs) are now beating state-of-the-art approaches in many natural language processing benchmarks, they are typically trained to output the next best response, rather than planning ahead, which is required for multi-turn interactions. We address these challenges using a novel RL construction.

Conversation 127

Conversation Learning Open Sound 127

Google Research AI blog

JUNE 13, 2023

Modeling human attention (the result of which is often called a saliency model) has therefore been of interest across the fields of neuroscience, psychology, human-computer interaction (HCI) and computer vision. Attention-guided image editing Human attention models usually take an image as input (e.g.,

Attention 78

Attention Model Experiment Images 78

Candid

JUNE 10, 2024

Beyond providing instructions for completing and submitting the form, consider offering additional guidance such as proposal examples, file-naming conventions, FAQs, and how-to videos. Diversify and train the reviewer pool.

Application 105

Application Grant Practice Review 105

The Verge

DECEMBER 16, 2020

Massachusetts legislators passed the first major state-level facial recognition ban, following a model set by individual cities like Boston and San Francisco. Facial recognition is just one facet of the sweeping bill, which aims to reform police tactics, training, and accountability.

Massachusetts 116

Massachusetts San Francisco Boston Registration 116

Google Research AI blog

JUNE 6, 2023

We fine-tuned a large language model to proactively suggest relevant visuals in open-vocabulary conversations using a dataset we curated for this purpose. For example, out of context, the transcription model misunderstood the word "pier" as "pair", but Visual Captions still recommends images of the Santa Monica Pier.

Language 69

Language Model Conference Video 69

The Verge

JUNE 24, 2020

A coalition of AI researchers, data scientists, and sociologists has called on the academic world to stop publishing studies that claim to predict an individual’s criminality using algorithms trained on data like facial scans and criminal statistics. Algorithms trained on racist data produce racist results.

Research 140

Research Studies Authoring Law 140

The Fundraising Authority

OCTOBER 28, 2021

Maybe you need to create an appeal for an email campaign, a proposal, or a new set of thank you letters and you’re at your desk just wishing that coherent sentences would magically start appearing? I have used it to help me write appeals, emails, thank you letters, and draft some 1 and 2-page proposals. Fundwriter.ai Fundwriter.ai

Cancer 130

Cancer Los Angeles Proposal Donor 130

Google Research AI blog

JULY 13, 2023

Scaling up language models has unlocked a range of new applications and paradigms in machine learning, including the ability to perform challenging reasoning tasks via in-context learning. Language models, however, are still sensitive to the way that prompts are given, indicating that they are not reasoning in a robust manner.

Language 75

Language Model Learning Instruction 75