Internal development test of the upcoming OpenAI GPT-3 robot skill (Currently for private beta only)
+ How To Add This Control To Your Project (Click to Expand)
- Make sure you have the latest version of ARC installed.
- Select the Get button in this page to download the archive file.
- Double click the downloaded archive file to execute installer.
- The installer will add this control to ARC.
- Load ARC and press the Project -> Add Control button from the menu.
- Choose the Artificial Intelligence category tab.
- Press the Openai Test icon to add the control to your project.
Generative Pre-trained Transformer 3, commonly known by its abbreviated form GPT-3 robot skill, is an unsupervised Transformer language model. This GPT-3 robot skill can be connected to speech recognition and chatbot responses (i.e. pandoraBot, BotLibre, etc.) to assign scripts that will be executed based on the related topic. Phrases can be added for various moods, actions, and responses with respective scripts. This GPT-3 robot skill is not a chatbot (see bottom of this manual for what this isn’t).
*Note: This is an internal development test of the upcoming OpenAI GPT-3 robot skill (Currently for private beta only). This skill is under development and only API keys are not available to the public, yet. Stay tuned!
1) For example, a humanoid may have the phrase "stand up" with an associated script to execute an Auto Position motion of standing up. When a conversation detects a topic related to requiring the robot to stand up, the associated script is executed. An example conversation would be "robot, you have fallen over." or "robot, you are on the ground".
2) Various moods can be described in phrases with associated scripts. When a mood/sentiment is detected in the conversation, the associated script is executed. The script may trigger an Auto Position action to move the robot into a position or change the stance, appearance, eye color, etc.. This is useful for InMoov style robots with chatbot response and behavior.
3) An example of having the robot execute an action or behavior based on a subtle topic hint within a conversation. In this case, the user asked if the robot could hear the music and the robot chooses to dance.
4) Another example of having the robot move into various positions based on mood/sentiment, for example with InMoov robots, based on the mood of the conversation. This is similar to using one of the sentiment analysis robot skills (such as Vader).
The configuration allows the user to add phrases with associated scripts. The phrases can be longer sentences for increased detail or words for generic detection.
Open AI's GPT-3 Overview
Generative Pre-trained Transformer 3, commonly known by its abbreviated form GPT-3, is an unsupervised Transformer language model and the successor to GPT-2. It was first described in May 2020. OpenAI stated that the full version of GPT-3 contains 175 billion parameters, two orders of magnitude larger than the 1.5 billion parameters in the full version of GPT-2 (although GPT-3 models with as few as 125 million parameters were also trained).
OpenAI stated that GPT-3 succeeds at certain "meta-learning" tasks. It can generalize the purpose of a single input-output pair. The paper gives an example of translation and cross-linguistic transfer learning between English and Romanian, and between English and German.
GPT-3 dramatically improved benchmark results over GPT-2. OpenAI cautioned that such scaling up of language models could be approaching or encountering the fundamental capability limitations of predictive language models. Pre-training GPT-3 required several thousand petaflop/s-days of computing, compared to tens of petaflop/s-days for the full GPT-2 model.
Like that of its predecessor, GPT-3's fully trained model was not immediately released to the public on the grounds of possible abuse until now for select partner companies.
What This Is NOT
The videos and excitement around Open AI's GPT-3 demonstrate many uses. The greatness of GPT-3 is that the database has been trained using a controlled process. This means that every bit of data was thoughtfully planned and categorized during the learning process. Now that Open AI is allowing access to GPT-3 by select partners (Synthiam being one of them), there are strict usage rules that prevent their data from being "poisoned". This is because the dataset is still being trained by usage from partners. Because of this, Open AI's strict rules prevent their GPT-3 service from being used in the following manner...
- NO chatbots
- NO automatic social media posting emulating humans
- NO generation of essays, news, etc.
Because we're all focused on robotics, and chatbots are a common focus, this first rule does limit our usage of the GPT-3. As an Open AI partner, we respect that rule and therefore will not be enabling chatbot functionality with GPT-3. Please look for alternative robot skills for a conversational chatbot, such as BotLibre, PandoraBot, PandoraBot AIML2, and AIML BOT