It’s not a Pavlovian hatred, but deep understanding of the current internal algorithms that brings contempt. Sorry, you’ve been oversold, hook line and sinker! I’ve been in (and out) of the industry to have seen the cycles and blatant hype that is integral to its’ survival.
Yes, the industry is building massive DCs, storing lots of slurped up data, churning through billions of investor money, but the content is far short of what should be provided, and the absolute, blind trust in the output can be cringeworthy to observe.
Those that confuse LLMs and AI, something the industry keeps quiet.
That day of reckoning may come sooner than the shysters may wish, or the technology may catch up and they get away with it. The ‘next big thing’ beckons.
In the meantime, extreme caution. Trust, but verify. Not belittling the community, just a word of caution for those that rush in, and don’t pause to consider possible repercussions.
Use the right tool for the job. A four function calculator for your simple maths. An AI engine to assist with pulling together massive amounts of data for analysis. Just because you can doesn’t mean you should. Often simple is elegant.
And contempt isn’t appropriate for this forum. Go make your case in other forums where people debate change for the social good and leave the home assistant enthusiasts to their hobbies. That’s all I’m suggesting. The OP asks a simple question and your response not only provided nothing positive to the topic at hand but continued to beat a dead horse.
For the sake of brevity and the dear reader in the future, perhaps just link to one of your previous posts on the topic and leave it at that.
Theoretical has to come first, wouldn’t you say? One of the problems is that there are so many people having a go - you get Friday at one extreme and Fridge Assistant at the other. (Friday would easily pass the Turing test; Fridge Assistant uses AI to estimate food use by dates.)
What would be useful is a sort of framework for understanding. What we have at the moment is more of a collection of really interesting hacks.
I think a structured overview from end to end for beginners would really be helpful.
Like a decision tree they can follow to see what options they have and which steps are important in this case.
Starting with the initial decisions about what type of assistant. LLM or not. If LLM, local or cloud?
Integrations and models.
What’s important about scripts / intents.
TTS and STT.
Hardware options for voice assistants.
Why are own scripts important and what is HA missing by default.
Important ideas for a good prompt.
Best practice for own tools: E.g. structure, guidance in returned data like „soft errors“ that don’t fail but returning hints and errors to the LLM so it might retry in a better way.
Local hardware: Minimal hardware requirements. Tips / Tweaks / how to avoid common problems.
At the end also pointing to full solutions like ZenosAI.
The list could be endless.
I think it’s currently quite hard to get in that topic.
Even more if you want to host your own LLM.
The knowledge is spread about soo many threads.
It’s doable if we don’t eat the entire elephant at once…
We can framework what’s available now and what each brings with concerns
Starting would probably look like
Ai primer. Ai is not just ai, types and uses etc.
All the harness-likes (opencode, codex, Claude)
All the claw-likes (openclaw, hermes, Nemoclaw, paperclip)
What does a local farm need.
Routers, no not the box connected to the intarwebz…
MCP/A2A
RAG v ARAG and what the hell is a vector, Victor.
Speech services v inference services
Integrations role in AI in HA.
Conversation v AITask
The various integrations
Core
Top 5 custom
Whats the difference between anbexternal agent connecting in and an agent inside HA connecting out.
Where are the walls and boundaries.
Limits
Security concerns.
Then a link list to known threads like Thyraz tools, nics model info, taters voice stuff, zenos and Friday and halmark. Etc. Etc.
I’ll write a framework article and some intros and post it up if someone can handle link collecting and the folks who maintain the cookbook can be pinged.
I’m just going to leave this link in here to round off the discussion as to why you possibly might want to hesitate before gleefully leaping into the wonderful new unknown before checking your parachute is carefully fastened.
Oh the wicked irony!
Pulls out rosary beads and four function calculator…
