### primer on llm’s

<img src="https://hackmd.io/_uploads/B1_UAC9x0.jpg" height="500"/>

---

### agenda

- useful models for thinking about ai
- components of a good prompt
- things to be cautious of when working with ai
- putting it into action

---

### **intro**

<div style="display: flex;">
<div>
<img src="https://hackmd.io/_uploads/Sk7iC05eR.png" width="500"/>

</div>

<div>
<img src="https://hackmd.io/_uploads/rJ0hACcg0.png" height="150"/>

<img src="https://hackmd.io/_uploads/HJUaC05eC.png" height="150"/>
</div>

</div>

Note:
- coding bootcamp based in london
- we've been experimenting with using ai in an educational setting
- amazed how much it's improved my day-to-day workflows and i wanted to share that with others

---

### a few definitions

---

**large language models** (llms) are ai technologies designed to process, understand, and generate text in a way that resembles human language. these models leverage massive datasets to learn language patterns, enabling sophisticated tasks like conversation, text completion, and question answering

---

### autocomplete on steroids

<img src="https://hackmd.io/_uploads/ryOe1-je0.jpg" height="500"/>

Note:
- They've read a lot of text and learned patterns in language.
- When you ask them something, they use statistics to guess the most likely response based on what they've seen before.
- They're not thinking or understanding like humans, but rather making educated guesses based on language patterns.

### useful models for thinking about ai

---

### no instruction manual

<img src="https://hackmd.io/_uploads/S1uD1Wil0.png" height="350"/>
<img src="https://hackmd.io/_uploads/ByHD1bogR.png" height="350"/>

**cost of entry 🎟️: 10 hours of experimentation and play**

Note:
- you'll also develop intuition on the boundaries of its capabilities
- not all tasks can be improved with ai. in fact, trying to use ai for tasks outside of its boundaries can results in a significant decrease in productivity

---

### llm's different personalities


---

### people-pleasing intern

<img src="https://hackmd.io/_uploads/Hy71xZseR.png" height="375"/>

**infinitely helpful intern who is a little naive and wants to make you so happy that they will make up facts rather than disappoint**

Note:
- they will lie and assert their lies with full confidence

---

### use ai to reason, not to look things up

- unlike google, which is designed for searching and retrieving information, **llm’s are not designed for looking things up***

- large language models are **not repositories of knowledge**, rather they are **reasoning engines**

- think: *summarise, synthesise, rewrite, explain, analyse, provide examples etc*

Note:
- show perplexity

---

### two models for effective human-ai collaboration

<img src="https://hackmd.io/_uploads/B1KMgZsl0.jpg" height="350"/>
<img src="https://hackmd.io/_uploads/B1RMgbol0.jpg" height="350"/>

**centaur & cyborg**

Note:
- **Centaur**: Maintains a clear separation between tasks done by the human and tasks delegated to the AI.
- **Cyborg**: Deeply integrates the efforts of the human and AI in a more fluid, granular way.
- Both models involve **the human and AI working together** on tasks, rather than the AI simply replacing the human or the human working entirely independently.
- In both models, the **human remains in the loop**, exercising judgment, oversight, and decision-making rather than blindly deferring to the AI.

---

### breakout room

- what has been your experience of using llm's so far?

- in what areas of your work do you think it could be helpful?

---

### any questions so far?

---

### primer on effective prompting

**the more you put in, the more you'll get out** 💪🤖

---

### prompt ~~engineering~~ crafting

**definition**: *the process of iteratively refining the ai input to improve responses*

prompt → **trial & error** → effective output

Note:
- "Prompt crafting" emphasizes the **creative and intuitive** approach to guiding AI, rather than the technical rigidity of "prompt engineering"
- The term acknowledges **the blend of art and science** in AI interaction, accurately capturing the essence of working with generative AI

---

### add context

- give the ai a persona (you’re marketing pro)

- specify an audience (you’re writing for secondary school students)

- define an output format (give me a table in a word document)

Note:
- show explanation about why the sky is blue as physics professor versus primary school teacher
- write an email telling a client i'm going on holiday
- i'll be gone from 05/05 - 15/05
- i won't be contactable
- i'll tie up the proposal before i go away
- to get going again after i'm back i'll need the deliverables (copy and photos) to be ready for 16/05

- show conversation with the raycast docs
- **automation scripts**: ask ai for script to output md content to one file
- i have a directory with markdown files. the markdown files are in the main directory but also in sub-directories. i would like to run a script in the terminal that outputs the content of all the markdown files into one `.txt` file. how do i do this?
- https://github.com/raycast/extensions/tree/main/docs
- https://download-directory.github.io/

- please use the docs i've pasted as your source of truth for any code you write.

i want to create a raycast extension that will run a shell command from a specific directory. how would i do this?

---

### add examples (few-shot prompting)

Note:
- show email prompt

---

### go step by step (chain of thought prompting)

- ask the ai to go step-by-step through instructions

- e.g. first, outline the results; then produce a draft; then revise the draft; finally, produce a polished output

Note:
- go through image creation with chatty g, doing the brainstorming, critique and final image generation things separately
- I'd like to come up with an image to advertise a workshop i'm running on using large language models as a personal tutor to enhance my learning
- Please brainstorm multiple ideas
- critique those ideas given the context
- make revised versions of
- the ideas, and then finally generate the image.

---

### **these techniques require experimentation to get right**

Note:
- no need to include all components in all prompts
- show prompt library

---

### demos

---

### summarising meeting notes

Note:
- show curriculum concoctions

---

### research areas relevant to your work

Note:

- https://www.youtube.com/watch?v=eeOsD553gnA
- https://chatgpt.com/share/fc91d3dc-448d-40b4-bdac-2fa53ee8bf32

---

### get answers quickly from overly long and boring documents

Note:
- this is my travel insurance policy. i need the answer to the following questions
am i covered for travelling to the us?
am i covered if i lose my laptop?
am i covered if my flights get cancelled?
- also good for filling out taxes at the end of the year

---

### Thank you!