Okay, if I was to start trying to help educate people on AI stuff where should I start. What questions are likely to be asked by someone really interested in learning rather than someone who's just trying to start a fight? - ThreadSky

deepfates.com.deepfates.com.deepfates.com.deepfates.com.deepfates.com • 8 days ago

Okay, if I was to start trying to help educate people on AI stuff where should I start. What questions are likely to be asked by someone really interested in learning rather than someone who's just trying to start a fight?

Reposted from ☀️✨🌿cyber plantae🌿✨☀️

There are very few authorities currently doing exactly what I’m describing, which means there’s a very open niche for it. Very few people are AI/ML engineers which means there’s a high need for communicators. People want a chance to participate even if they’re not directly building it themselves.

Comments

anamiak.bsky.social•8 days ago

The hard part is definitely not answering their questions, it's getting them interested enough to ask questions. People have a lot of unknown unknowns about AI and you kind of just have to find the concepts that hook them in some way. It's very rare that someone knows what they want to know.

gracekind.net•7 days ago

IMO I would just post about stuff you’re interested in, if people want to learn they’ll ask questions

cyberplantae.bsky.social•8 days ago

This book is a little outdated (mainly bc of the recent changes to Google AI Search) but it’s my content bible for thinking about what questions to answer ;)

valhallaresearch.net•8 days ago

I think explaining the data collection -> training -> improvement -> inference pipeline would be helpful for folks outside of the AI space. I see a lot of people saying some off the wall shit that seems to come from "energy costs for training ~= energy costs for inference" misunderstandings

cyberfuturist.bsky.social•7 days ago

Anyone asking for like use cases, capabilities, or personalization?

jcorvinus.bsky.social•7 days ago

I would start with vector embeddings and compare them to memory, then move onto explain how training data is essentially used to hammer shapes into a giant manifold, and that no wholesale copying is done

extentofthejam.bsky.social•7 days ago

I should be reading more about this myself, as I use LLMs and other transformers a fair bit. This article seems pretty decent https://www.datacamp.com/blog/vector-embedding

jcorvinus.bsky.social•7 days ago

By showing embeddings first, you can show how absolutely beautiful the relational nature of language and knowledge is. It illustrates that, like a candle can light many flames without diminishing any, the human spirit can be handed to an ANN without diminishing the source it comes from

deepfates.com.deepfates.com.deepfates.com.deepfates.com.deepfates.com•7 days ago

interesting. I like this take. I think the vector embeddings and the sort of magic of everything reflecting everything else is indeed beautiful but how do you start from there?

jcorvinus.bsky.social•7 days ago

I think that would come from a bit of serendipitous exploration but if you were making, say, a youtube video, it would be very easy to find a word that represents positive emotions, render it, then as you explain this concept have nearby words in the space pop in, and draw the lines between them

jcorvinus.bsky.social•7 days ago

After you have a 10 minute or so long-form video you could make shorts that cover the most impactful bits. Think something halfway between vsauce and 3blue1brown

rhizonymph.com•7 days ago

This actually came up a bit in one of 3b1b's videos where he did something like this.

Short clip of this part:
https://youtube.com/shorts/FJtFZwbvkI4?si=OpFZOyG0WG3M9wy3

Full video:
https://youtu.be/wjZofJX0v4M?si=BDYxCc3eGwsG9l7g

Comments

Posting Rules

Reply