Holy crap I haven't thought about AARDWOLF in YEARS!

Especially if it was remote and they decided to bring people back in.

Musk also overstayed his student visa, and never went to school. Where's ice knocking at his door?

Built and I use it every day. There's a lot of AI trash out there. VC funded, rush to market, product owners telling devs to stuff AI into everything. Doesn't mean that there's not also a lot of really solid working tools. It's like you're saying "look at all this useless rock!" in a gold mine.

I actually build these systems. It's not just theory or research. And the "complex" part is that the graph I added is highly simplified compared to the models I'm creating. And I'm just some dude, imagine what big companies and people with real resources are building.

I hope that helps a little. It's complex and quickly evolving. Even things like RAG where I currently have to decide how to categorize and store data are being improved to automatically do this in an even more fluid way.

- Vector databases store chunks of information that are similar mathematically closely
- Agent can have a part of it's process be "what data do I need to answer?" instead of answering from trained data
- Agents can review conversations and store/update memories for future reference when relevant.

Here's a simplified version of how agentic memory works.
excalidraw.com#json=bEDABfS...
- LLMs have context windows to store relevant info (short/working memory)
- Additional details and summaries can be stored outside of the context window for quick reference when needed

I did say it might sound crazy 😅

Too late for me to draw my own graph, but it's something like this

www.sciencefocus.com/the-human-bo...
pmc.ncbi.nlm.nih.gov/articles/PMC...
We're still learning a lot but we know a lot more now than we did a few years ago. Even then we could see the relationship between neural connections (more connections easier retrieval) a while back.

I appreciate that. But then I'm curious, how do you see human knowledge different from LLM knowledge? It's all weights and connections, whether between neurons in the brain or between layers in a model. We just seem to have a lot more control over the updates to our prompts and models.

Seriously. Full disclosure, I use AI tools to help improve my coding productivity (largely successfully), but the first time a tool does something like this to me, I'm removing it from my list. I have zero tools that I would trust to make changes to a production db, much less without approval.

I can give an LLM short and long term memory. I can give it an internal dialogue. I can give it access to research and encourage it to update its memories based on all of that.
I think we humans often think far too highly of ourselves and it cripples our perspectives.

What is thinking? Is it thoughts coming and going and us choosing which to hold on to? Some large percentage of people don't have an internal narrative, what is thinking for them? We get input from our environment and update our networks based on response.

What if, and I know this will sound crazy, but what if they actually went after actual well known child abusers? It's not like they're hard to find.

For real.

I'd much rather be disappointed to find out people I used to respect were on the list than to live in a reality where justice isn't served and the perpetrators live free to continue abusing people.

OMG I hate how accurate this is.