mrdio / applied-ai-technologies Goto Github PK

Context
Employees of a large enterprise company wish for a more powerful search function in the social intranet of the company. The problem is, the content is spread around multiple ressources, such as intranet articles, GitHub, Mattermost, etc. The Bing Chat, which runs on the GPT-4 model by OpenAI, could represent a possible solution to that problem. In addition to generating responses, Bing Chat also searches for information in the internet, includes links to that information and integrates these links directly into the generated response.

Task
Create a proof-of-concept implementation of an ai assisted ChatBot similar to Bing Chat.

Details

• Since it is a POC, it is not required to access Company resources. Search in openly available sources, such as (at least) Google, public GitHub repositories and StackOverflow is sufficient.
• OpenAI (e.g. the gpt-3.5-turbo model) or a similar library should be used.
• GUI is not required, CLI interaction is sufficient.

An AI that signalizes the user about his posture. Users can get a notification if they have a bad posture for too long and should sit accordingly. In contrary users can get a notification complimenting the users posture. In addition, a weekly report can be send to the user.

More information: https://github.com/nfiniity/universe/issues/2

More information: https://github.com/nfiniity/universe/issues/1

More: https://github.com/nfiniity/universe/issues/4

Login and deep dive in Kaggle.com. Show how this plattform works, what challanges and dataseDeep Dive into the COVID-19 related datasets. Show some notebooks and examples.

Context
Current AI assisted ChatBots, such as ChatGPT or Bing Chat, are extremely powerful and can be used as personal assistants on daily basis. However, they are limited by the data they have been trained on. In the context of an enterprise company, it is interesting how the underlying GPT model can be trained on company specific data.

Task
Train and verify an openly available GPT model, such as OpenAI gpt-3.5-turbo, on a set of specific data.

Details

• The data could come from a openly available specialized domain, such as StackOverflow.
• It is important to use data, which is not already included in the trained model, e.g. more recent information than September 2021 in case of gpt-3.5-turbo.
• Prior to training, the model should not be able to generate satisfactory responses. After the training, it should.

This presentation introduces to definitions and topics of AI and gives some examples. The math behind the magic rounds this introduction to an easy to understand overwiew.

Build a filter that can be used with any videoconferencing tool and not only changes you background but also makes you professionally dressed.

This task includes a short presentation and intro to colab with a simple use case e.g. image classifier or MNIST. Timeframe: 1 Week. Duration 10 mins.

More: https://github.com/nfiniity/universe/issues/5

Check out platforms can be used to give others your Model to try out e.g. as online API. Maybe this will be the start of your AI-Company! Check out fast.ai deployment and get inspired by clarifai.com. Tell us your best practices