2.4 Working with GPTs

 2.4 Working with GPTs

Sat, 18 Apr 26

Multilingual Translation Demo

• ChatGPT can translate book content into multiple languages

• Works with free version

• Demonstrated with book excerpt “Journeys Through the Echoes”

• Generated translations: Spanish, French, German, Chinese, Arabic, Hindi, Swahili

• Verified Swahili translation accuracy for Amazon jungle content

LinkedIn Profile Optimization

• ChatGPT creates professional LinkedIn profiles from resume input

• AI optimization critical for job applications

1. Most companies use AI to scan resumes/profiles

2. Non-optimized profiles get missed even when qualified

• Optimization strategies:

1. Upload resume to ChatGPT

2. Request LinkedIn profile creation

3. Ask for domain-specific optimization (finance, healthcare, tech)

4. Target specific company types (FAANG, healthcare, etc.)

• Can analyze existing profile URLs and suggest improvements

Sentiment Analysis Techniques

• Few-shot prompting demonstrated for customer review analysis

• Categories: Positive, Negative, Neutral, Mixed

• Example results:

• “Tunnel Visions” → Mixed (good acting vs predictable plot)

• “The Nightingale” → Positive (blown away, highly recommend)

• Restaurant review → Negative (poor service, high prices)

• Token consumption warning: avoid repetitive prompts in same chat session

Multimodal AI Capabilities

• Definition: Single LLM handling multiple data types (text, image, audio, video)

• ChatGPT/Gemini can process different modalities simultaneously

• Architecture uses sub-LLMs or agents for different capabilities

• Applications:

• Image captioning and analysis

• Video analysis

• Virtual assistants

• Travel recommendations from map analysis

• Model cards on Hugging Face show specific capabilities and limitations

Token Economics & Cost Management

• Tokens = unit of measurement for LLM processing capacity

• Billing based on token consumption (input + output)

• Cost optimization strategies:

1. Clear, concise prompts reduce token usage

2. Clean data requires less processing

3. Vague instructions cause expensive token waste

• Context window limits require new chat sessions when full

• Both prompt length and response length affect total cost