11
10 Comments

This OpenAI update makes it easier to build tons of cool new products

People working on laptops with an "OpenAI" logo above their heads

Two days ago, OpenAI launched an experimental GPT-4 "long output" model.

What's new: This "long output" model can produce up to 64,000 tokens, a significant increase from GPT-4's initial 4,000-token limit. That’s a 16-fold increase.

Background: Most OpenAI models have a token limit of 128,000 tokens. Think of tokens as similar to words (OpenAI offers a calculator to see how many tokens your text contains).

Here’s the deal: There are two types of tokens:

Input tokens: When people refer to token limits, they usually mean the maximum number of tokens you can use as input. Longer input tokens are useful if you want to provide extensive context to the AI model. For example, you can input an entire book and then ask questions about it.

Output tokens: This refers to the amount of text the AI generates in response. Previously, if you input a mini report and wanted OpenAI to rewrite it, you’d be limited to 4,000 tokens. With this new model, that limitation is significantly increased.

What can you build

A disclaimer: While building these tools was possible before, it was much more challenging. You had to prompt the AI to "continue" its generation, which could cause loss of context and introduce errors.

With the new context models, you can get a comprehensive response in one go:

  • Comprehensive Book Summaries: Input entire books and generate detailed chapter-by-chapter summaries.

  • In-Depth Research Reports: Create extensive research reports or whitepapers by feeding in large amounts of source material.

  • Advanced Content Rewriter: Rewrite lengthy articles, reports, or documents in a single pass.

  • Detailed Code Reviews: Input a large codebase and get reviews for the entire code.

  • Long-Form Content Creation Assistants: AI-powered tools that can generate entire ebooks, whitepapers, or comprehensive research reports in one go.

  • Business Plan Generators: 64,000 tokens is a substantial limit for creating detailed business plans.

  • Creative Writing Assistants: Generate novel outlines, plot developments, or even entire drafts of novels or screenplays.

Hope you found this useful!

  1. 2

    I wonder if there's a tendency for the tokens towards the end of the generated stream to become more and more "incorrect", simply due to stochastic drift.

  2. 1

    "Exciting times ahead with OpenAI's new long output model! 🎉 The ability to generate up to 64,000 tokens opens up endless possibilities for creating innovative products—from comprehensive book summaries to detailed business plans. Can't wait to see what cool tools the community will build with this powerful update!"

  3. 1

    Interesting. What about Gemini 1.5 Pro or other experimental versions?

  4. 1

    What do indiehackers use when they build products around AI, OpenAI? Is there anyone using an AI service other than Open AI? is anyone hosting their own AI model?

    1. 1

      OpenAI, Google Gemini is amazing, Claude, Perplexity are all great for diverse things

    2. 1

      Not at the moment, but planning to. AWS Bedrock would be my go to.

  5. 1

    Nice, keep up the good work man :)

  6. 1

    This is exciting news and hopefully it will lead to more reliable content.