``Chat AI based on data collected by human power'' appeared from the human language model project ``Open Assistant'', so I tried using it
LAION-AI, known for the development of large-scale data sets such as '
Open Assistant
https://open-assistant.io/ja
To use chat AI using Open Assistant, first access the above link and click 'Try our assistant'.
Next, enter your email address and click 'Continue with Email'.
When the following screen is displayed, you will receive a sign-in URL at the email address you entered, so check your inbox.
I entered my Gmail email address this time, but the email with the sign-in URL was in the spam folder. Since you can not click the link as it is, click 'No problem'.
When the mail moves to a normal folder, click 'Sign in'.
Then the terms of use will be displayed, so read to the bottom.
After reading the terms to the end, click 'Accept'.
Then you will see the Open Assistant dashboard.
Since it is displayed in English in the initial state, select 'Japanese' from the language selection column at the top of the screen.
After switching to Japanese display, click 'Chat' on the left side of the screen.
When you open the chat screen, a warning about 'settings that need to be done only once' will be displayed at the top of the screen. However, at the time of writing the article, the setting was unnecessary for the new account, so there is no problem even if you click 'OK' to dismiss this warning.
To start a chat, click 'Create a new chat'.
When the chat screen is displayed, enter text in the input field on the right side of the screen and click 'Confirm'. This time, I entered 'Can you use Japanese?'
AI's answer is like this. 'Yes! I can speak Japanese,' he replied in English.
Even if I ask them to reply in Japanese, they still stubbornly use English. Looking at the contents of the answer, it seems that you can understand the Japanese sentences, but at the time of writing the article, it seems that it does not correspond to responses in Japanese.
The results of asking various questions to AI are as follows. When asked, 'Who is the Prime Minister of Japan?', he replied, 'I only have knowledge up to September 2021.'
Next, when I asked about the prefectural capital of Kanagawa Prefecture, I got an incorrect answer that 'the prefectural capital of Kanagawa Prefecture is Tokyo.'
The result of asking 'Recommended transportation from Tokyo to Osaka' is as follows. We got a generally correct answer: 'About 2 hours and 56 minutes by Shinkansen' and 'About 1 hour by plane'.
Open Assistant is a project that aims to build a high-quality language model by collecting human-powered answers. Details on how to participate in the project are provided below. At the time of writing the article, we are accepting responses in Japanese, so it may be possible to respond in Japanese in the future.
LAION AI starts an open source large-scale language processing model ``Open Assistant'' development project that anyone can participate in-GIGAZINE
In addition, the dataset collected by Open Assistant is published at the following link.
OpenAssistant/oasst1 Datasets at Hugging Face
https://huggingface.co/datasets/OpenAssistant/oasst1
Related Posts:
in Review, Web Application, Posted by log1o_hf