How the sausage gets made
hey skimmers!
here’s the newsletter structure
🤘 Unsupervised Learning is free but you can choose a paid subscription
announcement
poll
links
term of the week
recent learning(s)
build in public update
share <3
I’ve come to a sorta, kinda, sad fact of matter…
Currently,USL isn’t providing much value to either demographic of listener (non-technical nor very technical) given that it doesn’t go high level enough or deep enough for either.
The solution? A little structure✨to the chaos.
After meeting Victor Geislinger recently, and seeing his approach to transferring knowledge- we’ll be testing out accompaniment videos/content to podcasts.
anyone can follow a cookbook
the real value is the context and understanding- Victor G
We’ll be starting at 0 and answering all those “beginner” questions.
(First episode out soon)
Vote for what you’d like to watch first?
Links of interest
Listen
- Most recent podcasts on future of Revops and AI YouTube, Spotify, Apple Podcasts,
- 🎙️a very cool podcast on decentralisation and NEOPETS
- A conversation with James Hawkins of Posthog 🦔
Read
- An upcoming podcast ep sees me speaking to Principal Research Scientist at Microsoft Ai for Good Labs - Lucas Meyer. Here’s a post he wrote I enjoyed
- Curiousity Conflict
Test
- Brainstory - talk out loud and solve your problem with socratic method of questioning
- Vapi- a software developer kit to build voice enabled assistants on top of. See me fawning over “her” here
Term of the week
Classification
What does this look like “in the wild” (aka, using Chatgpt, Gemini etc)
Let's say you ask ChatGPT "Is this a good time to invest in tech stocks?"(don’t do this irl)
The sorter gets to work The LLM analyses your question, trying to figure out what you're actually asking for.
It has been trained to recognise different intent categories like
- Seeking factual information
- Asking for an opinion
- Requesting an action
The LLM likely classifies your question as "seeking factual information" but maybe with a hint of "asking for opinion". This classification helps it decide how to formulate the best possible answer.
What I’ve tried
Last week, I made my first solo explainer content on Youtube around crewai + the difference between AI agents and chatbots.
Crew AI is a Python framework (based on lagchain) designed to make building AI agents easier. Key concepts being agents, tasks, tools + process.
Tldr- I *wanted* to test crewai with Anthropics’ Claude 3 api (to have “nicer” prompt responses) and I ran into some issues with implementation. Approximately ~10 hrs of coding flailing later
a long conversation with crewai’s customgpt and watching a few videos I reached an unlikely conclusion.
See more here for the video I made Agents vs chatbots crewai video
The unlikely conclusion?
I’ll be exploring DSPy (programming language)- an alternative to prompting.
I’ve also started a series on DSPy - by anthropomorphising the modules 🥸
@unsupervisedlearningpodMeet Denise, Leo, Sam, and the rest of the crew at the DSPy office, where AI meets creativity. 🔗 Links DSPy Documentation: dspy-docs.vercel.app GitHub github.com/stanfordnlp/dspy large language model explained who has the best large language model large language model large scale model trains#dspy #prompting #ai #innovation #technology #learnAI #weviate #skit #programminghumor
Tiktok failed to load.
Enable 3rd party cookies or use another browser
What’s next?
Here’s a list of ongoing tasks/projects at USL if you’re up for contributing!
Build in public month #2 update here
Ways to help out
- Share 🤘 (un)supervised learning
- Recommend someone for the pod
- Follow, rate, and subscribe on Spotify, Youtube and Apple podcasts
Cheers,
Renee
🤘 Unsupervised Learning is free but you can choose a paid subscription
Member discussion