LLMs are machine learning models trained on a huge amount of data that understand and can generate text in a human-like way. An LLM is the engine behind a chatbot — so GPT-o3 is the LLM behind ChatGPT.

LLMs use complex math to predict the next word to write. Think of them like a calculator for words.

Key limitations are that they’re not great at getting our writing style just right, they lack nuanced context (like we have as humans), they may hallucinate, their context window (aka short term memory) has limits, and their training data will always be out of date

Prompts are the instructions you give a chatbot, and there’s an art to doing them well to optimize outputs.

Carve’s prompt crafting framework is COIF: include Context, Output, Intent, Format in every prompt for best results.

10 “buckets” of tasks you can use a chatbot for in your workflow: writing and editing, research, meetings, minutes & events, project planning, travel, expenses, brainstorming as a thought partner, learning and business intelligence, translation and cultural nuances, data cleaning & analysis (see below for detailed appendix).

AI risk overview:

• Job displacement

  • First major automation impact on white-collar jobs

  • World Economic Forum predicts more jobs lost than gained to AI

• Data privacy concerns

  • Most AI tools train on user data by default

  • Companies typically store conversations for 30 days to monitor for unsafe use, even if not training on your prompts/uploads

  • ChatGPT experienced a minor data breach back in 2023, highlighting security risks

• Other key kisks

  • Inaccuracy and hallucination in AI outputs

  • Bias in models, particularly visible in image generation

  • Environmental impact due to high compute requirements

  • Copyright and IP uncertainties around AI-generated content

  • Risk of mediocrity through over-reliance on AI