Disappointment with 8192 Output Length Limit for Powerful AI Models

I am writing to express my disappointment with the current output length limit of 8192 tokens. While I understand there may be technical limitations, it feels severely restrictive, especially considering the immense capabilities of the AI models we are working with.

For many complex tasks and applications, 8192 tokens simply aren’t sufficient. This limitation hinders the potential of the models for generating detailed reports, creative writing pieces, comprehensive code, and many other applications that require longer outputs.

I believe that increasing the output length limit would significantly enhance the usability and power of Google AI. It would unlock new possibilities and allow users to fully leverage the models’ strengths.

I urge the Google AI team to consider increasing the output length limit as a priority. This improvement would be greatly appreciated by the community and would further solidify Google AI’s position as a leader in the field.

5 Likes

Is that a UI bound limit in the AI Studio UI for Gemini 1.5? Gemini models API-wise lead with context length (1M and lately 2M for some closed circles).

Not an UI limit. The 1 or 2 million is the limit for input tokens. The limit for output tokens is 8192 for the 1.5 models (older models had half of that).

2 Likes

Increasing the output limit will significantly make the Google AI popular. People want to create books with ai. They want to talk with books etc.

4 Likes

I fully agree. While it´s double the size of ChatGPT and all the other models on the market (limited to 4096 output tokens), it would be great to increase this beyond 8192, especially given the huge input tokens limitation of up to 2M.

2 Likes

Hi,
I cant even get pro to output above 1800 tokens :frowning: I’m trying to get gemini-1.5-pro-002 to write a report and have given it large reference inputs and a well designed detailed prompt, including concepts like variables for target word count, but i can seem to get this to occur. Other than something like batch processing image descriptions, i cant seem to get large output. Have you had any success wiht getting that and any suggestions, thx?

10 Easy Steps for those who can’t count, or use AI tools.

  1. Create an Index (Prompt: Make me an index of 10 Chapters of “The Wild Life of Albert” The story of one of the world’s best known but unknown geniuses.)
  2. Check the index to make sure it’s what you want (Remove chapter of 3rd wife. Unproven conspiracy. Change Title of Chapter 5 to “Poincare, I hardly knew ya.”)
  3. Have Gemini create Chapter one (Write Chapter 1)
  4. Have Gemini Create Chapter two (Chapter 2)
  5. Have Gemini create chapter 3 ( 3)
  6. Figure out if you really want chapters 4, 5, 6, 7, 8, 9 and 10 because it’s SOOOO much hard work.
  7. Complain to your friends that it took you two whole hours to write a 200 page book and your fingers are tired.
1 Like

Useful tips. And then, we can adopt this to create our own AI crew too. :star_struck:

I’m not just saying it, I’m trying it. It can be time consuming, and painful and really embarrassing if you don’t read EVERY word it spits out, you are risking your job. It routinely throws in tangents and made up stuff assuming everything from the internet is equally likely to be true no matter how wrong that is. It is much better than a year ago, but still hallucinates more than most grad students. The amazing thing is that you can write a book in a few weeks if you happen to be expert enough to already have been able to write that book in a few years. That’s a force multiplier.