Blog | Surge AI

Introduction to Reinforcement Learning with Human Feedback

Featured

Introduction to Reinforcement Learning with Human Feedback

Edwin Chen

Large Language Models

Learn about reinforcement learning with human feedback (RLHF) — a new technique for training large language models that has been behind many of the major advances in OpenAI's ChatGPT and InstructGPT LLMs, DeepMind's Sparrow, Anthropic's Claude, and more.

Recent
articles

The latest in AI, language, and data labeling. Subscribe below!

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form.

How Anthropic uses Surge AI’s RLHF platform to train their LLM Assistant on Human Feedback

How Anthropic uses Surge AI’s RLHF platform to train their LLM Assistant on Human Feedback

Large Language Models

Learn RLHF from Hermione Granger. RLHF Models Explain RLHF!

Learn RLHF from Hermione Granger. RLHF Models Explain RLHF!

Large Language Models

How RLHF Shifts LLMs from Autocompletion to Conversational Understanding

How RLHF Shifts LLMs from Autocompletion to Conversational Understanding

An Illustrated Guide to Reinforcement Learning with Human Feedback

An Illustrated Guide to Reinforcement Learning with Human Feedback

Large Language Models

Introduction to Reinforcement Learning with Human Feedback

Introduction to Reinforcement Learning with Human Feedback

Large Language Models

2022 Blog Recap: Trends in AI, Language, & Data

2022 Blog Recap: Trends in AI, Language, & Data

We Evaluated ChatGPT vs. Google on 500 Search Queries

We Evaluated ChatGPT vs. Google on 500 Search Queries

Large Language Models

AI Red Teams for Adversarial Training: How to Make ChatGPT and LLMs Adversarially Robust

AI Red Teams for Adversarial Training: How to Make ChatGPT and LLMs Adversarially Robust

Large Language Models

HellaSwag or HellaBad? 36% of this popular LLM benchmark contains errors

HellaSwag or HellaBad? 36% of this popular LLM benchmark contains errors

Large Language Models

Real-World ML Failures: The Violence, Racism, and Sexism Uncaught by Twitter's Content Moderation Systems

Real-World ML Failures: The Violence, Racism, and Sexism Uncaught by Twitter's Content Moderation Systems

How TikTok is Evolving the Next Generation of Search

How TikTok is Evolving the Next Generation of Search

Evaluating Generative AI: Did Astral Codex Ten Win His Bet on AI Progress?

Evaluating Generative AI: Did Astral Codex Ten Win His Bet on AI Progress?

Hate Speech Datasets in English, Spanish, Japanese, Arabic, and More

Hate Speech Datasets in English, Spanish, Japanese, Arabic, and More

Free Toxicity Dataset (September 2022)

Free Toxicity Dataset (September 2022)

Why Instagram is Losing Gen Z: We Asked 100 Users to Compare TikTok vs. Reels

Why Instagram is Losing Gen Z: We Asked 100 Users to Compare TikTok vs. Reels

Andrew Mauboussin

Andrew Mauboussin

The $250K Inverse Scaling Prize and Human-AI Alignment

The $250K Inverse Scaling Prize and Human-AI Alignment

Large Language Models

Labeled Dataset of Misinformation on Facebook

Labeled Dataset of Misinformation on Facebook

Dataset of Financial Transactions, Labeled with Intent and Expense Category

Dataset of Financial Transactions, Labeled with Intent and Expense Category

Andrew Mauboussin

Andrew Mauboussin

Human Evaluation of Large Language Models: How Good is Hugging Face's BLOOM?

Human Evaluation of Large Language Models: How Good is Hugging Face's BLOOM?

Large Language Models

30% of Google's Emotions Dataset is Mislabeled

30% of Google's Emotions Dataset is Mislabeled

Search Behind-the-Scenes: How Neeva Uses Human Evaluation to Measure Search Quality

Search Behind-the-Scenes: How Neeva Uses Human Evaluation to Measure Search Quality

Generating Children's Stories Using GPT-3 and DALL·E

Generating Children's Stories Using GPT-3 and DALL·E

AI Red Teams and Adversarial Data Labeling with Redwood Research

AI Red Teams and Adversarial Data Labeling with Redwood Research

Andrew Mauboussin

Andrew Mauboussin

Dataset of Roe v. Wade Tweets Labeled by Abortion Stance

Dataset of Roe v. Wade Tweets Labeled by Abortion Stance

Case Study: Content Moderation for a Leading Social Media Company

Case Study: Content Moderation for a Leading Social Media Company

Humans vs. Gary Marcus vs. Slate Star Codex: When is an AI failure actually a failure?

Humans vs. Gary Marcus vs. Slate Star Codex: When is an AI failure actually a failure?

Dataset: Reddit Crypto Sentiment

Dataset: Reddit Crypto Sentiment

I wanted burritos. Facebook's search engine sent me to a dead restaurant 45 minutes away.

I wanted burritos. Facebook's search engine sent me to a dead restaurant 45 minutes away.

Human Evaluation

How Surge AI Built OpenAI's GSM8K Dataset of 8,500 Math Problems

How Surge AI Built OpenAI's GSM8K Dataset of 8,500 Math Problems

New Data Labeling Tools for Training Chatbots and Conversational Assistants

New Data Labeling Tools for Training Chatbots and Conversational Assistants

Andrew Mauboussin

Andrew Mauboussin

Dataset: Stock Market Sentiment on Social Media

Dataset: Stock Market Sentiment on Social Media

10 Egregious Failures in Gmail Spam Detection

10 Egregious Failures in Gmail Spam Detection

Content Moderation

How to Pick the Best Data Labeling Platform

How to Pick the Best Data Labeling Platform

Is Elon right? We labeled 500 Twitter users to measure the amount of Spam

Is Elon right? We labeled 500 Twitter users to measure the amount of Spam

Content Moderation

We asked 100 humans to draw the DALL·E prompts

We asked 100 humans to draw the DALL·E prompts

The average number of ads on a Google Search recipe? 8.7

The average number of ads on a Google Search recipe? 8.7

Google Search is Falling Behind

Google Search is Falling Behind

Building a No-Code Machine Learning Model by Chatting with GitHub Copilot

Building a No-Code Machine Learning Model by Chatting with GitHub Copilot

Andrew Mauboussin

Andrew Mauboussin

Surge AI Achieves SOC 2 Compliance

Surge AI Achieves SOC 2 Compliance

Andrew Mauboussin

Andrew Mauboussin

Writing a Super Bowl Worthy Commercial with GPT-3

Writing a Super Bowl Worthy Commercial with GPT-3

Moving Beyond Engagement: Optimizing Facebook's Algorithms for Human Values

Moving Beyond Engagement: Optimizing Facebook's Algorithms for Human Values

Andrew Mauboussin

Andrew Mauboussin

Using Surge AI with a Hybrid Cloud Model

Using Surge AI with a Hybrid Cloud Model

Andrew Mauboussin

Andrew Mauboussin

Holy $#!t: Are popular toxicity models simply profanity detectors?

Holy $#!t: Are popular toxicity models simply profanity detectors?

Content Moderation

Surge AI's Free Toxicity Dataset

Surge AI's Free Toxicity Dataset

Content Moderation

An Analysis of Omicron Tweets: 30% are Skeptical of the Medical Establishment

An Analysis of Omicron Tweets: 30% are Skeptical of the Medical Establishment

Is Google Search Deteriorating? Measuring Google's Search Quality in 2022

Is Google Search Deteriorating? Measuring Google's Search Quality in 2022

Human Evaluation

The World's Largest Profanity List

The World's Largest Profanity List

Inter-Annotator Agreement: An Introduction to Krippendorff’s Alpha

Inter-Annotator Agreement: An Introduction to Krippendorff’s Alpha

Andrew Mauboussin

Andrew Mauboussin

Why I Joined Surge AI

Why I Joined Surge AI

How Good is Your Chatbot? An Introduction to Perplexity in NLP

How Good is Your Chatbot? An Introduction to Perplexity in NLP

ML School: Neural Networks

ML School: Neural Networks

Inter-Annotator Agreement: An Introduction to Cohen's Kappa Statistic

Inter-Annotator Agreement: An Introduction to Cohen's Kappa Statistic

5 Examples of the Importance of Context-Sensitivity in Data-Centric AI

5 Examples of the Importance of Context-Sensitivity in Data-Centric AI

The Pitfalls of Inter-Rater Reliability in Data Labeling and Machine Learning

The Pitfalls of Inter-Rater Reliability in Data Labeling and Machine Learning

A Visual Introduction to Language Models in NLP (Part 1: Intuition)

A Visual Introduction to Language Models in NLP (Part 1: Intuition)

Andrew Mauboussin

Andrew Mauboussin

How Surge AI Helps NYU Study the Impact of Social Media

How Surge AI Helps NYU Study the Impact of Social Media

Manifold: Using Neural Embeddings to Explore the Shape of Your Data

Manifold: Using Neural Embeddings to Explore the Shape of Your Data

Getting started with Surge AI's Python SDK

Getting started with Surge AI's Python SDK

Andrew Mauboussin

Andrew Mauboussin

Supercharge your NLP with Surge AI’s NER labeling tool

Supercharge your NLP with Surge AI’s NER labeling tool

The AI Bottleneck: High-Quality, Human-Powered Data

The AI Bottleneck: High-Quality, Human-Powered Data