Let’s first see how simple it is to create a machine learning model represented as a virtual ‘AI table’ in MindsDB. In this example, we will call our GPT4 model gptbot_model.

CREATE MODEL mindsdb.gpt_model
PREDICT response
USING
engine = 'openai',
-- api_key = 'your openai key', in MindsDB cloud accounts we provide a default key 
model_name = 'gpt-4', -- you can also use 'text-davinci-003', 'gpt-3.5-turbo'
prompt_template = 'respond to {{text}} by {{author_username}}'; 
 

Note: 

Optionally, If you are using MindsDB on docker or if you want to use your own OpenAI API key, simply pass theapi_key argument in the USING.

One important attribute here is prompt_template. This is where we tell GPT how to write answers; it is a template because you can pass values from columns, in this case the template contains {{author_username}} and {{text}}, which will be replaced from the WHERE variables in the query. Let’s see in action:

SELECT response from mindsdb.gpt_model WHERE author_username = "mindsdb" AND text="why is gravity so different on the sun?";

As you can see, the previous model gave responses that aren’t that exciting. But we can use some prompt template magic to shape how we want the model to respond. Essentially we use prompt_template to explain in plain english how we want GPT to formulate its responses. Let's create a model mindsdb.snoopstein_model with a prompt template that gives GPT a hybrid personality: he is half-Einstein, half-Snoop Dogg. A brilliant physicist who owns the rap game. His name is Snoop Stein:

CREATE MODEL mindsdb.snoopstein_model
PREDICT response
USING
engine = 'openai',
max_tokens = 300,
-- api_key = 'your openai key, in cloud accounts we provide one',
model_name = 'gpt-4', -- you can also use 'text-davinci-003' or 'gpt-3.5-turbo'
prompt_template = 'From input message: {{text}}\
by from_user: {{author_username}}\
In less than 550 characters, write a Twitter response to {{author_username}} in the following format:\
Dear @<from_user>, <respond a rhyme as if you were Snoop Dogg but you also were as smart as Albert Einstein, still explain things like Snoop Dogg would, do not mention that you are part Einstein. If possible include references to publications for further reading. If you make a reference quoting some personality, add OG, for example;, if you are referencing Alan Turing, say OG Alan Turing and very briefly explain why you think they would be dope reads. If the question makes no sense, explain that you are a bit lost, and make something up that is both hilarious and relevant. sign with -- mdb.ai/bot by @mindsdb.';
 

SELECT response from mindsdb.snoopstein_model
WHERE 
author_username = "someuser" 
AND text="@snoop_stein, why is gravity so different on the sun?.";

SELECT response from mindsdb.snoopstein_model
WHERE 
author_username = "someuser" 
AND text="@snoop_stein, Apart from yourself, which rappers would make the best physicists and why?!";
 

CREATE DATABASE my_twitter
WITH ENGINE = 'twitter';

This creates a database called my_twitter. This database ships with a table called tweets that we can use to search for tweets as well as to write tweets.

You can use the Twitter API to get a list of tweets with a particular text or hashtag, in the case below mindsdb or #mindsdb.

SELECT 
   id, created_at, author_username, text 
FROM my_twitter.tweets 
WHERE 
   query = '(@snoopstein OR @snoop_stein OR #snoopstein OR #snoop_stein) -is:retweet' 
   AND created_at > '2023-03-20' 
LIMIT 20;

Note that the parameter query supports anything that the twitter API supports as query, for more reference, read (https://developer.twitter.com/en/docs/twitter-api/tweets/search/integrate/build-a-query)

SELECT 
t.id AS in_reply_to_tweet_id, 
t.text AS input_text, 
t.author_username, 
t.created_at, 
r.response AS text 
FROM my_twitter.tweets t 
JOIN mindsdb.snoopstein_model r 
WHERE t.query = '(@snoopstein OR @snoop_stein OR #snoopstein OR #snoop_stein) -is:retweet -from:snoop_stein' 
AND t.created_at >  '2023-03-20' 
LIMIT 4;

CREATE DATABASE mindsdb.my_twitter_v2
WITH
  PARAMETERS = {
   "consumer_key": "your twitter App API key",
   "consumer_secret": "your twitter App API key secret",
   "bearer_token": "your twitter App bearer TOKEN",
   "access_token": "your twitter App Access Token",
   "access_token_secret": "your twitter App Access Token Secret"
  };

INSERT INTO my_twitter_v2.tweets (in_reply_to_tweet_id, text) 
VALUES 
 (1633439839491092482, 'MindsDB is great! now its super simple to build ML powered apps using JOBS https://docs.mindsdb.com/sql/tutorials/twitter-chatbot'),
 (1634126825377996800, 'Holy!! MindsDB is such a useful tool for developers doing ML https://docs.mindsdb.com/sql/tutorials/twitter-chatbot');

The CREATE JOB statement is great because you can use it to automate work. The idea is simple, you give it a query you want to execute and how often. Let’s set up a job for Snoop Stein!

Let’s write a JOB called gpt4_twitter_job:

CREATE JOB mindsdb.gpt4_twitter_job AS (
-- insert into tweets the output of joining model and new tweets
INSERT INTO my_twitter_v2.tweets (in_reply_to_tweet_id, text)
SELECT
  t.id AS in_reply_to_tweet_id,
  r.response AS text
FROM my_twitter.tweets t
JOIN mindsdb.snoopstein_model r
     WHERE
     t.query = '(@snoopstein OR @snoop_stein OR #snoopstein OR #snoop_stein) -is:retweet -from:snoop_stein'
     AND t.created_at > "{{PREVIOUS_START_DATETIME}}"
 limit 10
)
EVERY hour

SELECT * FROM jobs WHERE name="gpt4_twitter_job";
SELECT * FROM jobs_history WHERE name="gpt4_twitter_job";

DROP JOB gpt4_twitter_job