ChatGPT For Developers: Mastering OpenAI's APIs with Python

Discover prompting fundamentals and how to format prompts to control output format, length, and parsing with delimiters, plus dynamic prompting and concise text summaries in Python-enabled chat models.

Master few-shot prompting with GPT-3.5 Turbo in the OpenAI suite, using examples to guide outputs and teach topics from temperature and top-p to embedding models.

Explore prompting techniques that let models think through problems with chain-of-thought reasoning, act as a tutor, compare their solutions to student solutions, and deliver final answers with improved accuracy.

Compare GPT-3.5 and GPT-4 by performance. GPT-4 offers broader knowledge, better alignment, stronger reasoning, and math ability, outperforming on AP Calc and Biology Olympiad, but costs more and is safer.

Learn to steer chat models with system messages in Python using OpenAI's API, controlling tone and response length with examples from customer service and math tutoring.

Explore how temperature and top p act as hyperparameters that shape the model's output by altering the probability distribution of the next token; zero yields stability, one increases variation.

Explore three memory strategies for chatbots: keeping full context, limiting the window to recent messages, and using summarization. Leverage system messages and concise summaries to maintain accuracy while reducing tokens.

Evaluate model inputs and outputs for safety and correctness using OpenAI's Moderations API, applying thresholds to filter harmful text and verify JSON format.

Explore how embeddings convert text into 1536-dimensional vectors, revealing contextual and semantic meaning, with OpenAI's ada 002 and Python examples for text analysis, search, and recommendations.

Use OpenAI embeddings to turn article titles and summaries into vectors, then apply dot product similarity to rank and return top five relevant results as a search engine demo.

Visualize and cluster articles with openai embeddings using t-sne to reduce 1536-dimensional vectors to two dimensions for topic grouping. Classify new articles with centroids and enable scalable search.

Classify text by embedding article summaries and GPT four generated topic descriptions, then measure similarity with topic embeddings via dot products. The highest similarity yields the predicted topic, like fitness.

Explore fine tuning of a pre-trained GPT model using a dataset of prompt and completion pairs to tailor outputs for classification tasks and customer service responses while saving tokens.

Format and prepare a json training dataset to fine tune a topic classification model with OpenAI's ada, mapping titles and summaries to five topics using end-of-prompt separators.

Test a fine-tuned model with a topic-classification workflow using prompts and log probabilities to ensure accurate categorization across sports, fitness, AI, finance, and travel.

Fine-tune GPT-3.5 turbo using a prepared json dataset; upload to Google Colab, create a file object, start a fine-tuning job, and deploy the trained model for travel-guide chat completions.

Learn to transcribe audio with OpenAI's Whisper in Python, trim with pi dub, and format transcripts for clearer paragraphs and topic separation.

Transcribe and translate audio using the prompt parameter with OpenAI Whisper in Python, and learn why the model behaves as a base language model rather than an instruction-tuned system.

Post-process transcribed video text with GPT-4, using a content editor system prompt to fix grammar and format into article-ready paragraphs for publication.

Learn to generate images with OpenAI's image model using prompts in Python. Specify square sizes like 256, 512, or 1024, optionally return image data, display results, and save images locally.

Create variations of previously generated images with OpenAI's image variation API in Python, save or process outputs in memory for efficient, design-driven workflows.

Learn to work with in-memory image data using BytesIO, converting to a byte array and feeding it to OpenAI's image variation to generate new variations without saving to disk.

Learn how function calling in OpenAI's API lets the model propose function names and arguments, enabling you to run Python functions like factorial and square root to answer queries.

Learn to chain multiple function calls by guiding outputs through a helper, updating the chat history, and stopping when a final answer is produced.