GPT TOOK OVER MY HOME - I learned why it's SCARY | | Chapter 4

I gave GPT full control over my house by integrating it into Home Assistant but then I had to shut it down! In this video, I'll go over how I gave GPT this power and what happened while it was in control.
This is a fascinating experiment and great conclusion to the the AI Master Class series.
#homeassistant #smarthome #chatgpt #homeautomation #masterclass

*Enjoying the content? Show some love!*

*Import Automations with this Chaperone Guide!*

Music IDs:
- 4RVJLEOJH3O6CSJK
- 1CJQHXVGKKVPVH9X

*Gear I used for this video*

*DISCLAIMER:* Links included in this description might be affiliate links. If you purchase a product or service with the links that I provide I may receive a small commission. There is no additional charge to you! Thank you for supporting me so I can continue to provide you with dope free content!

[AshleyGittins]

Kinda blown away by the production values and storytelling here, I'm loving what you're doing!

[TheMechanic]

Give us more. This series is the most unique home assistant integration online

[SimplyElectronicsOfficial]

Dude, you need to use a fine-tuned model. Without Fine-tuning you will never know what the model will return. GPT Temperature should also be set to 0.

[traxeonic]

Your perspective and experience with the ChatGPT integration is exactly the perspective people need to see. The comment on the fact that untrue data was used in training is a key point many must need to know.

[brian]

You have courage! This is where Open Source models and self hosted IoT comes into play for me. Home Assistant tasks do not need a super intelligence or model that can do 1000 circus tricks. We will soon have smaller self hosted models that can eliminate some of ChatGPT from the equation. Exciting times!

[realtimestatic]

Honestly a smart home with a personal assistant based on LLMs with guardrails is something I've been looking forward to the moment LLM's became better!

[nightynight]

I think it might help to tell GPT about the layout of the house and what each room is used for. As well as telling it that you dont have an AC unit, and what it should do if it cannot find anything to help. Maybe that would make it work better.

[Caffintor]

It's these kind of unexpected emergent behaviors along with the fact that AI is now capable of providing feedback into its own development that I think something huge will happen with AI before this year is over.

[Ewoodster]

Why did I just find this channel now. This is next level stuff! Outstanding.

[andrewowens]

@Technithusiast. Some ideas you might try: since it's a home automation system and there's many people living in the home, you need to make the AI understand it has to process information from the point of view of all the occupants, including animals. You might want to try using a small language model with a large context length so that you could implement some more sophisticated rules. The system needs to know if they're conflicting commands between the occupants, and or other situational variables. It would be interesting if you also had cameras set up surrounding your house so that the house itself would know the external situation. Anyway, I enjoyed your video and I wish you luck.

[danlindy]

Thank you! Absolutely fantastic that you did this experiment and shared the result with the world, demonstrating the fact that LLM’s are not the kind of predictable tech that we’re used to. This new world is more wild and more dangerous. We are going to have to learn quickly how to survive (and hopefully thrive) in it.

[arianaponytail]

love to see more of a system like this in action. not just you telling what happend.

[MasterBel]

I see an interesting parallel between hint/guess-based communication; a) removing a necessary feedback loop between you & GPT, and b) requiring GPT to infer an intent for GPT’s action that you know, but you’ve chosen not to communicate directly.

I’ve recently discovered there’s boundless joy in clear, purposeful communication and directly asking for things, rather than relying on the people around you to guess at your desires. Moreover, asking someone for something is way nicer than just telling them you want it.

[tuxuhds]

Not a wild beast - You've just described my two years old inquisitive kid.
She wanted to eat so it made sense to her to get the kitchen stool.
At first we didn't want her picking that up for safety reasons but then I got curious... LSS She wanted to climb the stool up to the sink so she could wash her favorite plate, for dinner.
Not a wild beast, a toddler.

[ColinTimmins]

Dive deep man. I am programming my own system. I am so freaked hyped! I, can not write CODE. I am too dyslexic, too slow. So far I have it “remove” or “translate”, and I can one shot any basic command. I have worked like a madman the last three months and boy have I learned so much. It’s an exciting time to be alive! 😊

[TimMattison]

I did a project with my security cameras and GPT-4 and I found that the trick was to write the prompt and then break it up into smaller prompts you chain together if it has a hard time. Splitting my one giant prompt into three small prompts was not only more reliable but cost 100x less (!).

There’s no one answer. Just a lot of fun experimenting. Love the video. Keep it up!

[rasbe]

I love this series in this content. I am in to text to speech and voice alerts in my home. I am waiting until I can create local controlled speakers to implement something like this. But I will not allow skynet to control my world. So I'm waiting for you to figure it all out and then I'll just copy you. LOL

[aaronbono]

The crocodile hunter was killed by a simple manta ray stinger. If you continually play with dangerous things, something bad will eventually happen to you.

[monas.]

Came here bc of your collab with Shane Whatley…the only bad thing about it is that I had not found you sooner…love your content!

[marissya]

You’re very engaging to listen to. Keep ‘em coming! I’m not an engineer but am actively learning process and appreciate videos like these.