# Discourse AI + Data Explorer?
**URL:** https://meta.discourse.org/t/discourse-ai-data-explorer/274011
**Category:** Feature
**Tags:** data-explorer, completed, ai, ai-bot
**Created:** [August 4, 2023, 6:54pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011 "2023-08-04T18:54:32Z")
**Posts on this page:** 20
**Page:** 1
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 4, 2023, 6:54pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/1 "2023-08-04T18:54:33Z")
I decided I wanted a particularly query today. Not wanting to bug my engineers (too much) I decided to ask ChatGPT to write a [data explorer](https://meta.discourse.org/t/32566?silent=true) query for me based on what I knew, and what I knew I wanted. I was pretty blown away that it got everything correct (aside from adding a semicolon at the end that it shouldn’t have).
Here’s the query it built for me when I gave it some parameters and asked it to build me a query that gave me a running total number of all users by month:
```plaintext
SELECT
timeframe "Month, Year",
user_count as "Users Joined This Month",
SUM(user_count) OVER (ORDER BY TO_DATE(timeframe, 'Month, YYYY'))::integer as "Users (Running Total)"
FROM (
SELECT
TO_CHAR(u.created_at, 'Month, YYYY') as "timeframe",
count(u.id) as "user_count"
FROM
users u
GROUP BY
timeframe
ORDER BY
timeframe
) subquery
ORDER BY
TO_DATE(timeframe, 'Month, YYYY')
```
Has there been consideration for integrating AI into the [data explorer](https://meta.discourse.org/t/32566?silent=true) area? 🙂
---
### Author:  [@JammyDodger](https://meta.discourse.org/u/JammyDodger)
#### Post date: [August 8, 2023, 8:47am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/2 "2023-08-08T08:47:52Z")
A post was split to a new topic: [Could we re-open the topic to share Explorer queries?](https://meta.discourse.org/t/could-we-re-open-the-topic-to-share-explorer-queries/274314)
---
### Author:  [@EricGT](https://meta.discourse.org/u/EricGT)
#### Post date: [August 5, 2023, 7:56am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/3 "2023-08-05T07:56:08Z")
> [@jordan-violet](#):
>
> ask ChatGPT to write a [data explorer](https://meta.discourse.org/t/32566) query
Care to share the prompt, the version of ChatGPT and the model used with ChatGPT. 🙂
For example
* * *
[ChatGPT August 3 Version](https://help.openai.com/en/articles/6825453-chatgpt-release-notes)
Model: GPT-3.5
**Prompt**
What is Discourse [Data Explorer](https://meta.discourse.org/t/32566?silent=true)?
**Completion**
As of my last update in September 2021, there is no specific tool or product called “Discourse [Data Explorer](https://meta.discourse.org/t/32566?silent=true)” that I am aware of. However, I can provide a general explanation based on the information available up to that point.
Discourse is an open-source, modern forum software that allows online communities to have discussions and interactions. It’s widely used for hosting discussions, Q&A forums, and support communities. Discourse provides a rich set of features for managing discussions, user engagement, and moderation.
A “[Data Explorer](https://meta.discourse.org/t/32566?silent=true)” typically refers to a tool or feature that allows users to query and explore data within a system or platform. It’s often used in the context of databases or data analytics platforms.
Given the two components, “Discourse [Data Explorer](https://meta.discourse.org/t/32566?silent=true)” could potentially be a hypothetical tool or plugin that provides users with the ability to explore and query data within a Discourse forum. This could be helpful for community administrators or data analysts who want to gain insights into user behavior, content trends, or other metrics related to the discussions taking place on the forum.
It’s important to note that my information might be outdated, and new developments might have occurred after my last update. If “Discourse [Data Explorer](https://meta.discourse.org/t/32566?silent=true)” is indeed a real product or feature, I recommend searching for the latest information to get accurate details about its capabilities and usage.
* * *
The reason for asking is that you have an excellent idea and it was nice to see it shared but in the future others may try to ask another prompt and sharing the prompt, completion and details of ChatGPT are needed if help is needed to understand any problems.
Note: I am a category moderator on the OpenAI Discourse forum so have lots of experience with those seeking help with ChatGPT prompts. ([ref](https://community.openai.com/u/ericgt))
* * *
The version of ChatGPT can be found in the lower right below where a prompt is entered
The model can be found near the top of the page in the middle
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 6, 2023, 4:48am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/4 "2023-08-06T04:48:41Z")
I…don’t like this bot.
---
### Author:  [@simon](https://meta.discourse.org/u/simon)
#### Post date: [August 6, 2023, 6:30am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/5 "2023-08-06T06:30:22Z")
> [@jordan-violet](#):
>
> Has there been consideration for integrating AI into the [data explorer](https://meta.discourse.org/t/32566) area?
Extending this idea a bit, allowing users to interact with applications using natural language is pretty much my dream as far as AI goes. With Discourse and the [Data Explorer](https://meta.discourse.org/t/32566?silent=true) plugin, it could be possible to get results from prompts like “return a list of users who have expressed interest in moderation.” Or even better, “send an invitation to the community\_moderation group to all users who have expressed an interest in moderation.”
Getting back to what’s possible now:
> [@EricGT](#):
>
> Care to share the prompt, the version of ChatGPT and the model used with ChatGPT.
Testing with ChatGPT 3.0 and 3.5, I’ve gotten good results when asking it to return a PostgreSQL query and copying the structure of the tables I’m interested in into the prompt. For example:
> **Summary**
>
> **Me** : Return a postgresql query that lists all usernames for users who are in the “community\_moderators” group for the following database structure:
>
> ```plaintext
> # == Schema Information
> #
> # Table name: groups
> #
> # id :integer not null, primary key
> # name :string not null
>
> # == Schema Information
> #
> # Table name: group_users
> #
> # id :integer not null, primary key
> # group_id :integer not null
>
> # == Schema Information
> #
> # Table name: users
> #
> # id :integer not null, primary key
> # username :string(60) not null
>
> ```
>
> **ChatGPT 3.5** : Sure, here’s a PostgreSQL query that lists all usernames for users who are in the “community\_moderators” group based on the given database structure:
>
> ```sql
> SELECT u.username
> FROM users u
> JOIN group_users gu ON u.id = gu.user_id
> JOIN groups g ON gu.group_id = g.id
> WHERE g.name = 'community_moderators';
>
> ```
>
> This query retrieves the usernames from the `users` table by joining it with the `group_users` table on the user ID and then joining with the `groups` table on the group ID. It filters the results to only include usernames for users who are in the “community\_moderators” group.
>
> **Me** : Here is the structure of the application’s topics and posts tables:
>
> ```plaintext
> # == Schema Information
> #
> # Table name: topics
> #
> # id :integer not null, primary key
> # title :string not null
>
> # == Schema Information
> #
> # Table name: posts
> #
> # id :integer not null, primary key
> # user_id :integer
> # topic_id :integer not null
> # post_number :integer not null
> # raw :text not null
>
> ```
>
> Return a query that lists all topic ids, post ids, and usernames for all posts created by members of the “community\_moderators” group that give some indication that the poster is interested in how to moderate polarizing discussions.
>
> **ChatGPT 3.5** : To achieve this, you would need to join the `posts` , `topics` , `users` , and `group_users` tables based on the provided schema. Here’s a PostgreSQL query that retrieves the requested information:
>
> ```sql
> SELECT t.id AS topic_id, p.id AS post_id, u.username
> FROM posts p
> JOIN topics t ON p.topic_id = t.id
> JOIN users u ON p.user_id = u.id
> JOIN group_users gu ON u.id = gu.user_id
> JOIN groups g ON gu.group_id = g.id
> WHERE g.name = 'community_moderators'
> AND (p.raw ILIKE '%moderate%' OR p.raw ILIKE '%polarizing%' OR p.raw ILIKE '%discussion%');
>
> ```
>
> This query retrieves the `topic_id`, `post_id`, and `username` for all posts created by members of the “community\_moderators” group that contain certain keywords indicating an interest in moderating polarizing discussions. The `ILIKE` operator is used for case-insensitive pattern matching in the `raw` column of the `posts` table.
>
> Please note that this query assumes that the keywords you provided are indicative of the desired posts. You might need to adjust or expand the keywords based on your specific criteria.
>
> **Me** : Write a query that returns a count of posts created in each weekly period since September 2022, created by members of the community\_moderators group that indicate some interest in how to moderate polarizing discussions.
>
> **ChatGPT 3.5** : To achieve this, you can use the following PostgreSQL query to return a count of posts created in each weekly period since September 2022, created by members of the “community\_moderators” group that indicate an interest in how to moderate polarizing discussions:
>
> ```sql
> SELECT
> DATE_TRUNC('week', p.created_at) AS week_start,
> COUNT(p.id) AS post_count
> FROM
> posts p
> JOIN
> users u ON p.user_id = u.id
> JOIN
> group_users gu ON u.id = gu.user_id
> JOIN
> groups g ON gu.group_id = g.id
> WHERE
> g.name = 'community_moderators'
> AND (p.raw ILIKE '%moderate%'
> OR p.raw ILIKE '%polarizing%'
> OR p.raw ILIKE '%political%'
> OR p.raw ILIKE '%contentious%'
> OR p.raw ILIKE '%debate%'
> OR p.raw ILIKE '%controversial%'
> OR p.raw ILIKE '%discussion%')
> AND p.created_at >= '2022-09-01' -- Start date: September 2022
> GROUP BY
> week_start
> ORDER BY
> week_start;
>
> ```
>
> This query calculates the count of posts created by members of the “community\_moderators” group that contain the specified keywords in each weekly period since September 2022. It groups the results by week and orders them chronologically. Please make sure to adjust the column names and table relationships based on your actual database schema.
* * *
The results seem good to me and represent the types of queries that took me a fair amount of time to write in the past. I’m assuming it would be possible to train a model on the Discourse database structure so that details about the structure could be left out of the prompts.
---
### Author:  [@Lilly](https://meta.discourse.org/u/Lilly)
#### Post date: [August 6, 2023, 3:01pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/6 "2023-08-06T15:01:31Z")
> [@simon](#):
>
> I’ve gotten good results when asking it to return a PostgreSQL query and copying the structure of the tables I’m interested in into the prompt.
oh this is great, i’ve been having mixed results with chatGPT [data explorer](https://meta.discourse.org/t/32566?silent=true) sql queries but doing it this way makes so much sense. i didn’t even think of providing table structure and asking for postgreSQL queries. 🤦🏻♀️ thank you for the tip @simon 🙂
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 6, 2023, 4:45pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/7 "2023-08-06T16:45:49Z")
Is there anywhere I can see a complete schema of the discourse database(s)?
---
### Author:  [@Lilly](https://meta.discourse.org/u/Lilly)
#### Post date: [August 6, 2023, 5:12pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/8 "2023-08-06T17:12:36Z")
hi @so-jordan-violet, i just use the collapsible table tool thing (i don’t know what it’s called! lol) on the right side in the [data explorer](https://meta.discourse.org/t/32566?silent=true) query edit interface:
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 6, 2023, 5:33pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/9 "2023-08-06T17:33:01Z")
If we can get a text-based document (especially if it’s in the sql create table format) we could supply that as instructions and probably get pretty reliable results.
---
### Author:  [@Falco](https://meta.discourse.org/u/Falco)
#### Post date: [August 6, 2023, 5:33pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/10 "2023-08-06T17:33:30Z")
> [@jordan-violet](#):
>
> Has there been consideration for integrating AI into the [data explorer](https://meta.discourse.org/t/32566) area? 🙂
Yes, we have this item on the AI team backlog since early this year. The idea is to ingest the whole (or the important part) of the Discourse schema in the prompt to allow it to create any queries.
We’ve been building [GitHub - discourse/data-explorer-examples: Examples for Discourse Data Explorer · GitHub](https://github.com/discourse/data-explorer-examples) to help provide examples to the model, so it can be better grounded.
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 6, 2023, 5:35pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/11 "2023-08-06T17:35:01Z")
I wish I had more time to peruse the code base, but do you have the SQL stored somewhere (publicly) for the crest table schemas?
I’m doing some LLM work with my own product/company right now and would love to play with this.
---
### Author:  [@Lilly](https://meta.discourse.org/u/Lilly)
#### Post date: [August 6, 2023, 5:49pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/12 "2023-08-06T17:49:59Z")
> [@Falco](#):
>
> We’ve been building [GitHub - discourse/data-explorer-examples: Examples for Discourse Data Explorer](https://github.com/discourse/data-explorer-examples) to help provide examples to the model, so it can be better grounded.
oh this is great. thanks for the link Falco! 🙂
---
### Author:  [@sam](https://meta.discourse.org/u/sam)
#### Post date: [August 10, 2023, 5:45am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/13 "2023-08-10T05:45:54Z")
> [@jordan-violet](#):
>
> I wish I had more time to peruse the code base, but do you have the SQL stored somewhere (publicly) for the crest table schemas?
I just generate this direct from the schema
>
Something like this should get you started:
>
* * *
I have been exploring all sorts of approaches for getting this going. One problem though is that GPT 3.5 is just so steerable it gets confused.
Giving it schema certainly helps, but I am finding that I need to be ultra strategic with examples.
Just telling it a story about “how to count the archived topics for a user” ends up steering a query such as “how many posts did a user have” towards archived topics.
I was thinking:
```mermaid
graph TD
A[Add generic helpful information] --> B[Generate embeddings for question]
B --> C[Do a similarity search across examples]
C --> D[Add 3 examples to the prompt]
D --> E[Add schema to the prompt]
E --> F[Ask question]
F --> G[Get SQL]
```
But it turns out that unless the examples are spectacular and you do not include any counter examples … stuff gets bad.
I am starting to think that the best we have for now is just giving it schema and a preamble. Maybe adding another step at the end for query validation and retry.
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 10, 2023, 5:29pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/14 "2023-08-10T17:29:20Z")
> [@sam](#):
>
> Something like this should get you started:
>
> [schema · GitHub](https://gist.github.com/SamSaffron/bed2d9f942c4b55d9ec24a5927867457)
Would you have a complete schema that shows data types?
---
### Author:  [@sam](https://meta.discourse.org/u/sam)
#### Post date: [August 10, 2023, 9:00pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/15 "2023-08-10T21:00:46Z")
Not sure if it fits into token budget, will try
---
### Author:  [@simon](https://meta.discourse.org/u/simon)
#### Post date: [August 10, 2023, 9:01pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/16 "2023-08-10T21:01:53Z")
> [@sam](#):
>
> Giving it schema certainly helps
With ChatCPT 3.5 I’m getting unexpected results when I provide the full schema. For example, it keeps trying to find the topic author in the `topic_users` table.
I get successful results by limiting the information in the prompt to the tables and columns that are required to write the query. Also by adding some additional information about how the columns are used. For example:
```plaintext
# Table name: user_actions
#
# id :integer not null, primary key
# action_type :integer not null (:like=>1,:was_liked=>2,:new_topic=>4,:reply=>5,:response=>6,:mention=>7,:quote=>9,:edit=>11,:new_private_message=>12,:got_private_message=>13,:solved=>15,:assigned=>16)
# user_id :integer not null (the user who is to be credited with the action)
# target_topic_id :integer
# acting_user_id :integer (the user who performed the action, for example, a staff user can perform an action on behalf of a regular user)
# created_at :datetime not null
```
Another way to look at getting this to work with the [Data Explorer](https://meta.discourse.org/t/32566?silent=true) plugin would be to have users fill out a dynamically generate form that lists the data they are looking for, and what conditions they want to apply to the data. The prompt could be generated programmatically by Discourse, then sent to the LLM to have the query written.
Edit: @jordan-violet, this might be a useful starting point. Here’s an annotated partial schema that’s working for me. It works well as long as I limit my queries so that they can be answered by the provided schema. I’ve added additional details to the schema to clarify details that seem to confuse ChatGPT. The obvious downside of this approach is that a fully annotated schema will exceed ChatGPT 3.5’s token limit. I’ve included some information in the schema that could be omitted to reduce the number of tokens that are used.
> **Summary**
>
> * * *
>
> ```plaintext
> # == Schema Information
> #
> # Table name: application_requests
> #
> # id :integer not null, primary key
> # date :date not null
> # req_type :integer not null ("http_total"=>0,"http_2xx"=>1,"http_background"=>2,"http_3xx"=>3,"http_4xx"=>4,"http_5xx"=>5,"page_view_crawler"=>6,"page_view_logged_in"=>7,"page_view_anon"=>8,"page_view_logged_in_mobile"=>9,"page_view_anon_mobile"=>10,"api"=>11,"user_api"=>12)
> # count :integer default(0), not null
> #
> # Table name: users
> #
> # id :integer not null, primary key
> # username :string(60) not null
> # created_at :datetime not null
> # updated_at :datetime not null
> # name :string (the user's real name)
> # last_posted_at :datetime
> # active :boolean default(FALSE), not null
> # username_lower :string(60) not null
> # last_seen_at :datetime
> # admin :boolean default(FALSE), not null
> # trust_level :integer not null
> # approved :boolean default(FALSE), not null
> # approved_by_id :integer
> # approved_at :datetime
> # previous_visit_at :datetime
> # suspended_at :datetime
> # suspended_till :datetime
> # date_of_birth :date
> # ip_address :inet
> # moderator :boolean default(FALSE)
> # title :string
> # locale :string(10)
> # primary_group_id :integer
> # registration_ip_address :inet
> # staged :boolean default(FALSE), not null
> # first_seen_at :datetime
> # silenced_till :datetime
> #
> # Table name: topics
> #
> # id :integer not null, primary key
> # title :string not null
> # last_posted_at :datetime
> # created_at :datetime not null
> # updated_at :datetime not null
> # views :integer default(0), not null
> # posts_count :integer default(0), not null
> # user_id :integer (the id of the user who created the topic)
> # last_post_user_id :integer not null (the id of the user who created the last post in the topic)
> # reply_count :integer default(0), not null
> # deleted_at :datetime
> # highest_post_number :integer default(0), not null
> # like_count :integer default(0), not null
> # category_id :integer
> # visible :boolean default(TRUE), not null
> # moderator_posts_count :integer default(0), not null
> # closed :boolean default(FALSE), not null
> # archived :boolean default(FALSE), not null
> # bumped_at :datetime not null
> # archetype :string default("regular"), not null (can be set to either "regular" or "private_message")
> # slug :string
> # deleted_by_id :integer (the id of the user who deleted the topic)
> # participant_count :integer default(1)
> # word_count :integer
> # excerpt :string
> # highest_staff_post_number :integer default(0), not null
> #
> # Table name: posts
> #
> # id :integer not null, primary key
> # user_id :integer (the id of the user who created the post)
> # topic_id :integer not null
> # post_number :integer not null (indicates the post's order in its topic)
> # raw :text not null (the post's content)
> # created_at :datetime not null
> # updated_at :datetime not null
> # reply_to_post_number :integer (the post_number that the post is a reply to)
> # reply_count :integer default(0), not null
> # deleted_at :datetime
> # like_count :integer default(0), not null
> # bookmark_count :integer default(0), not null
> # reads :integer default(0), not null (the number of times the post has been read)
> # post_type :integer default(1), not null (:regular=>1, :moderator_action=>2, :small_action=>3, :whisper=>4)
> # last_editor_id :integer (the id of the user who last edited the post)
> # hidden :boolean default(FALSE), not null
> # hidden_reason_id :integer (:flag_threshold_reached=>1,:flag_threshold_reached_again=>2,:new_user_spam_threshold_reached=>3,:flagged_by_tl3_user=>4,:email_spam_header_found=>5,:flagged_by_tl4_user=>6,:email_authentication_result_header=>7,:imported_as_unlisted=>8)
> # edit_reason :string
> # word_count :integer
> # wiki :boolean default(FALSE), not null
> #
> # Table name: categories
> #
> # id :integer not null, primary key
> # name :string(50) not null
> # topic_id :integer (the id of the topic that is used for the category's description)
> # topic_count :integer default(0), not null
> # created_at :datetime not null
> # updated_at :datetime not null
> # user_id :integer not null (the id of the user who created the topic)
> # topics_year :integer default(0)
> # topics_month :integer default(0)
> # topics_week :integer default(0)
> # slug :string not null
> # description :text
> # text_color :string(6) default("FFFFFF"), not null
> # read_restricted :boolean default(FALSE), not null
> # auto_close_hours :float
> # post_count :integer default(0), not null
> # latest_post_id :integer
> # latest_topic_id :integer
> # position :integer
> # parent_category_id :integer
> # posts_year :integer default(0)
> # posts_month :integer default(0)
> # posts_week :integer default(0)
> # topics_day :integer default(0)
> # posts_day :integer default(0)
> #
> # Table name: groups
> #
> # id :integer not null, primary key
> # name :string not null
> # created_at :datetime not null
> # automatic :boolean default(FALSE), not null
> # user_count :integer default(0), not null
> # title :string
> # bio_raw :text (the group's description)
> # allow_membership_requests :boolean default(FALSE), not null
> # full_name :string
> # visibility_level :integer default(0), not null (who can see the group :public=>0, :logged_on_users=>1, :members=>2, :staff=>3, :owners=>4)
> # messageable_level :integer default(0) (who can message the group :public=>0, :logged_on_users=>1, :members=>2, :staff=>3, :owners=>4)
> # mentionable_level :integer default(0) (who can mention the group :public=>0, :logged_on_users=>1, :members=>2, :staff=>3, :owners=>4)
> # members_visibility_level :integer default(0), not null (who see the group's members :public=>0, :logged_on_users=>1, :members=>2, :staff=>3, :owners=>4)
> #
> # Table name: group_users
> #
> # id :integer not null, primary key
> # group_id :integer not null
> # user_id :integer not null
> # created_at :datetime not null
> #
> # Table name: user_actions
> #
> # id :integer not null, primary key
> # action_type :integer not null (:like=>1,:was_liked=>2,:new_topic=>4,:reply=>5,:response=>6,:mention=>7,:quote=>9,:edit=>11,:new_private_message=>12,:got_private_message=>13,:solved=>15,:assigned=>16)
> # user_id :integer not null (the user who is to be credited with the action)
> # target_topic_id :integer
> # target_post_id :integer
> # target_user_id :integer (for example, the id of the user whose post was liked)
> # acting_user_id :integer (the user who performed the action, for example, a staff user can perform an action on behalf of a regular user)
> # created_at :datetime not null
> # updated_at :datetime not null
> #
> # Table name: topic_views
> #
> # topic_id :integer not null
> # viewed_at :date not null
> # user_id :integer (will be set if the topic was viewed by a logged in user)
> # ip_address :inet (will be set if the topic was viewed by an anonymous user)
> #
> # Table name: user_visits
> #
> # id :integer not null, primary key
> # user_id :integer not null
> # visited_at :date not null
> # posts_read :integer default(0)
> # mobile :boolean default(FALSE) (will be set to TRUE if the user visited on a mobile device)
> # time_read :integer default(0), not null (the value returned is the number of seconds of the visit)
>
> ```
>
> * * *
Here’s a link to a chat session where I tried it out: [ChatGPT](https://chat.openai.com/share/d108c104-3aa3-45d9-9161-6da21d5b3a77)
If you want to try expanding on this list, the easiest way to do it is to copy the schema from the bottom of the model you’re interested in, directly from the Discourse code: [discourse/app/models at main · discourse/discourse · GitHub](https://github.com/discourse/discourse/tree/main/app/models).
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 11, 2023, 9:59pm UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/17 "2023-08-11T21:59:11Z")
It would be great to get exactly this, but a comprehensive list of every table. Where did you get this from?
EDIT: Ah, I see your link at the bottom, sorry. I probably won’t have the time to put them all together. Was hoping there was a master SQL schema somewhere 😕
---
### Author:  [@simon](https://meta.discourse.org/u/simon)
#### Post date: [August 12, 2023, 12:05am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/18 "2023-08-12T00:05:31Z")
> [@jordan-violet](#):
>
> Was hoping there was a master SQL schema
The problem is that just providing the schema isn’t enough information for ChatGPT. You need to provide it with details about things like:
- what the `application_requests` `req_type` integer codes stand for
- what the `topics` `user_id` column is used for
- what the `user_actions` table’s `action_type` codes stand for a what’s the difference between that table’s `user_id`, `target_user_id`, and `acting_user_id` columns
With those types of details, GPT 3.5 seems to do a good job without any additional training. The problem then becomes that to provide this level of detail about the entire database will result in the prompt exceeding the ChatGPT token limit (4096 tokens, including both the prompt text and the generated output.) If this type of approach was used, there’s need to be a way to limit what gets set in the prompt based on what information the user wanted to get from the [Data Explorer](https://meta.discourse.org/t/32566?silent=true) query.
---
### Author:  [@jordan-violet](https://meta.discourse.org/u/jordan-violet)
#### Post date: [August 12, 2023, 12:07am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/19 "2023-08-12T00:07:33Z")
That’s not been my experience at all. I’ve had a pretty enjoyable experience with it this evening. I’ll share some results later.
It gives me the feeling of a competent, but very entry level/juniornrole. It gets things close, I tell it what it did wrong, and it fixes it. Repeat.
I’m actually really enjoying what it’s giving me, and the time that it saves me based on my usage tonight.
---
### Author:  [@sam](https://meta.discourse.org/u/sam)
#### Post date: [August 12, 2023, 12:16am UTC](https://meta.discourse.org/t/discourse-ai-data-explorer/274011/20 "2023-08-12T00:16:18Z")
GPT 3.5 does 16k just fine today
The trouble starts happening when the requests are complicated, accounting for edge cases and so on
[Next page](https://meta.discourse.org/t/discourse-ai-data-explorer/274011.md?page=2)