I tried using 'OpenPlayground', a tool that can collectively compare responses from multiple chat AIs such as ChatGPT



The speed of evolution of chat AI is tremendous, and new language models are appearing one after another. There are multiple versions of ChatGPT alone, and language models specialized for specific uses have also appeared, so it is easy to get lost as to which model to choose. `` OpenPlayground '' developed by Nat Friedman, former CEO of Github, is a tool that provides a useful UI when comparing multiple language models. I immediately tried using it to see what it was like.

GitHub - nat/openplayground: An LLM playground you can run on your laptop

https://github.com/nat/openplayground

Looking at the GitHub page, it seems that there are two installation methods. This time, I will use the installation method using Docker.



How to install Docker is described in detail in the first half of the following page.

``Dream Booth Gui'' review that allows you to easily use ``Dream Booth'' that allows you to additionally learn patterns and styles from just a few illustrations to the image generation AI ``Stable Diffusion'' - GIGAZINE



After Docker can be started, just enter the command described in the installation method at the command prompt. Required data will be downloaded automatically.



Once installed, open your browser and open 'localhost:5432' to open the OpenPlayground page.



First, you need to connect to each chat AI on the 'Settings' page. Looking at the left column, you can connect with 'openai', 'cohere', 'huggingface', 'huggingface-local', 'anthropic', 'aleph-alpha' and 'openplayground'.



This time, I will connect to OpenAI. Enter

the OpenAI API key and check the comparison candidate model. This time, I selected 4 'text-ada-001', 'text-babbage-001', 'text-curie-001' and 'text-davinci-003' from the top.



First, check the operation by itself. Select a model, enter characters and click 'Submit' ......



AI-generated text has been added with highlighting.



After confirming the operation, move to the 'Compare' tab. The chat AIs that were included as candidates earlier are lined up in the lower right.



If you check the chat AI in the lower right, a frame with a name will be displayed in the left column.



If you put text in the upper input field and click 'Submit', all AI will answer at the same time. You can compare what kind of reply is returned for each AI.



When I asked if he could use Japanese, it seems that only davinci can be used among the AIs installed this time.



I want to play in Japanese a little more, so I went back to the 'Settings' tab and put 'gpt-3.5-turbo' and 'gpt-4' in the comparison.



When asked what the AI thinks about being compared to other chat AIs, davinci straight out said, 'I don't like it.' I feel a strong will to decide my own worth. On the other hand, gpt-3.5 seems to be a slightly calmer response, with glimpses of pride in some parts.



I was hoping that gpt-4 would return a more mature reply from here ... but unfortunately gpt-4 could not be used due to an error at the time of writing the article.

in Posted by log1d_ts