OpenAI Releases ChatGPT - This Week in AI

Peter Foy -

Welcome to our This Week in AI roundup - this week we have stories about OpenAI's new ChatGPT model for dialogue and a new startup working on "copilot for lawyers".

  1. OpenAI Releases ChatGPT: Optimizing Language Models for Dialogue
  2. Copilot for Lawyers raises $5M from OpenAI Fund

💎 Top Stories

OpenAI Releases ChatGPT: Optimizing Language Models for Dialogue

OpenAI unveiled a prototype general-purpose chatbot and it electrified the Twittersphere. It is capable of debugging and coding, writing long-form content, scripts, and essays.

The model they trained is called ChatGPT, which interacts in a conversational way and has the ability to answer follow-up questions, admit mistakes, challenge incorrect premises, and reject inappropriate requests.

Key points:

  • ChatGPT is fine-tuned from a model in the GPT-3.5 series.
  • OpenAI used Reinforcement Learning from Human Feedback (RLHF) methods for training the model.
  • The company used supervised fine-tuning for initial model training, with human AI trainers providing conversations.
  • They collected comparison data for reward model training, by sampling several alternative completions from conversations with AI trainers.
  • They are also using the Moderation API to warn or block certain types of unsafe content.
source

Limitations of the model include:

  • The model can be verbose, overuse certain phrases, and have inaccurate or nonsensical answers.
  • It is sometimes sensitive to tweaks to the input phrasing or attempting the same prompt multiple times.
  • Users can provide feedback on problematic model output through the UI

Read the full story

ChatGPT: Optimizing Language Models for Dialogue
We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. ChatGPT is a sibling model to InstructGPT, which is train…
OpenAIOpenAI

Copilot for Lawyers raises $5M from OpenAI Fund

Harvey, a startup building a "copilot for lawyers," has emerged from stealth with $5 million in funding from OpenAI Startup Fund, Jeff Dean, and Elad Gil.

Key points:

  • Harvey was founded by Winston Weinberg, a former securities and antitrust litigator, and Gabriel Pereyra, a former research scientist at DeepMind, Google Brain, and Meta AI.
  • Harvey provides lawyers with a natural language interface for their existing workflows.
  • It can answer questions in natural language and generate legal arguments and drafts.
  • Harvey's purpose is to serve as an intermediary between tech and lawyers, making them more efficient and allowing them to focus on higher value tasks.
  • Harvey is currently in beta and is being used by lawyers and legal aid organizations.
  • It faces competition from companies like Casetext, Klarity, and Augrented.
  • The company was funded by OpenAI Startup Fund, which in addition to capital, provides companies with access to new OpenAI systems and Azure resources from Microsoft.

Read the full story

Harvey, which uses AI to answer legal questions, lands cash from OpenAI
Harvey, a startup using AI to answer legal questions, has secured an investment from OpenAI as well as Google AI head Jeff Dean.
TechCrunchKyle Wiggers

That's it for this edition of This Week in AI, if you were forwarded this newsletter and would like to receive it you can sign up here.