The TLDR

Highlight: Can you believe it's almost December? As 2023 draws to a close, it's the perfect time to reflect on your career moves from this year. We made the 2023 Career Retrospect GPT, an interactive journal designed to help you reflect, move toward goals, and appreciate what you’ve accomplished. Visit the link here to try it yourself ⭐️ (ChatGPT Plus is required to create/use agents.)

Musing of the Week: What are Large Language Models (LLM)? We’re only going to be hearing more about them in 2024, and it’s not too late to understand how they work. More below ⬇️

📝 Pause & Reflect (Grab the GPT)

We’re loving the GPTs, and so is the Internet. We’ve had fun testing valuable GPTs that others have generously shared (check out the Dr. Huberman GPT (trained on all podcast episodes from The Huberman Lab) and the AI Executive Order and Policy Analyst). Take advantage of all of these awesome tools that give you access to free, structured expertise.

In that spirit, we wanted to share another one of our own: the 2023 Career Reflection GPT. (Click the link to use it via your own ChatGPT Plus account.)

Since we covered the GPT tutorial 2 weeks ago (see here), we’ve continued to mess with GPTs, configurations, training, and best practices.

🤯 New thoughts on GPT configuration:

• Think like a manager. You’re basically creating a new team member, so you need to define a clear purpose, goals, deliverable expectations, and parameters (e.g. temperature, response length).

  • To provide further expectations, include the “anti” cases - such as what is NOT the purpose, what is NOT the goal.

• Think about all of your edge cases (especially if you’re sharing the GPT), and test them in the playground multiple times to ensure the bot responds appropriately.

• Add as much context as possible by attaching a file in the “Configuration” stage. For example:

  • For a GPT writer, upload brand voice, tone, personality, and 10 examples

  • For a GPT strategist, upload business plan, monetization streams, and marketing channels

• If you’re using the GPT with a team at work, run it in beta with a select few users who are committed to providing detailed feedback for a revision.

💭 Tips for using the Career Reflection GPT:

• Treat each session as an independent instance . The GPT will not remember anecdotes or information from a previous conversation.

• The Career Reflection GPT will not end the conversation unless prompted by you, but remember that the actual ChatGPT usage plan limits you to 50 messages every 3 hours.

• Be specific with your anecdotes, and share your emotions in addition to plain facts. Be detailed with timelines and specifying past vs. present.

• The GPT has been trained to help you draw patterns and ask you thoughtful questions. Feel free to ask questions back to it or ask for more details.

• Know that the GPT has been trained to be encouraging, but also reflect hard truths when necessary.

• Know that we can’t see any of your conversations - it’s entirely between you and the bot. You get what you put in, so share as much as you are comfortable with.

🧠 Musing of the Week

Seriously, what’s an LLM?

We’ve spent a lot of time talking about Large Language Models (LLMs) (such as ChatGPT, Bard, and Jasper) and how to use them. Let’s refresh ourselves on what these models actually are and how they work.

LLMs (Large Language Models) are a type of artificial intelligence (AI). The term AI encompasses a vast array of technologies and systems, extending from the automated spam filters in your email inbox to the Amazon Alexa on your kitchen counter and even to the advanced self-driving features found in cars like the Tesla.

LLMs belong to a sub-branch of AI called Machine Learning (ML), which describes a set of techniques that allow computers to learn behaviors without specific instructions or rules. ML involves giving a model a large amount of information and allowing it to infer patterns and make predictions or decisions autonomously based on what in that data. So, how does this work for LLMs?

LLMs are trained on immense volumes of written text, allowing them to learn and predict language patterns. Consequently, these models become very good at guessing the next word in a sentence based on the context provided and the text they have been exposed to. Here’s a simplified example of how this works:

Consider the sentence 'I am a [blank].' Not every word makes sense to fill this gap; you wouldn’t say 'I am a lettuce' or 'I am a loudly' as these words don't logically complete the sentence. From its training data, the model learns to distinguish between grammatically and contextually appropriate words and those that don't fit.

However, a simple statement like this still has many potential predictions for the blank, such as 'teacher', 'musician', 'gardener', or 'enthusiast'. Each of these options is grammatically correct and could make sense depending on the additional context or information provided. But, if we provide more context like 'I love baking bread every morning. I am a [blank].', the prediction becomes more straightforward. In this case, the LLM (or a person) might reasonably predict the sentence to end with the word 'baker'.

LLMs function by scaling up this process, utilizing billions of lines of text as their training data, and leveraging any provided input as context for their responses. This allows LLMs to be highly effective tools for handling common or repetitive tasks. However, it also explains why LLMs and ML models in general struggle with generating truly novel outputs.

Axios has written some great articles if you’re interested in expanding your AI vocabulary or digging deeper into how these systems are built.

🙌 If you’re hyped about the generative AI industry specifically, here are some of the coolest roles we’ve seen this week:

• Strategic Account Director @ Movable Ink

• Senior Sales Development Representative @ Gradient AI

• AI Product Manager @ Alibaba Group

• Manager, AI Products @ Johnson & Johnson

• Senior Customer Success Manager @ Anyword

🔨 Check out these other AI tools we’ve been looking at this week (creator edition):

• Adsby: Run high-ROI Google search ads with AI-powered copy

• Logo Diffusion: AI logo generator (that can generate from your sketch!)

• Jupitrr: Generate relevant b-roll content, straight from your script

That’s all for this week. See you next Tuesday!

Lorel & Reily