@ClaudeDevs That’s nice. I had to build something similar, add a kanban board and have the main session orchestrate each task, initiate tests etc, and keep an eye on dependencies coming next. So this is a great direction to see
Have had sessions where Astra wouldn’t do monitoring of the running process. It simply wasn’t aware it was able to do that.
At night I need it to keep monitoring to ensure it spins up things if any parts of my training runs into twouble. Not sure if you have fixed or have focus on such things. 😉
@JensenHuang@ChaseLochmiller@OpenAI I cant even buy a 5090 or a 32gb ram block without selling my kidneys.. Really looking forward to when things change a bit
@thorborg Synes klart der skal være discipliner i tøjvask, opvask eller madlavning, så de kan skubbe og overføre optimeringer vi andre kan bruge til noget 😁
Thats a cool case. Shower detections based on behavioral curves and open windows, etc. are super handy :-) especially knowing the wall temps vs. inside air and humidity lowers overall mold risks substantially. Thats what i use my twin for, to foresee so i can do the same as you describe, when it matters.
@MagnusBarsoe Læste den også lige.. Den hopper jeg ikke med på. De har i forvejen tillæg oveni abb. Fordi deres intelligente opladning ikke oplader helt optimalt.
Ville nok kunne lave en styring selv der kigger frem til klokken 6 om morgenen på billigste priser…
I think I found the missing link: @OpenAI GPT Realtime 2.
I built full control of my home and trained RL to manage heating, ventilation more effectively and cheaper.
Now my dashboard can become a conversation.
The only limitation is cost. @sama, it still forces me to install a local wake word to be economically viable.
But ill soon be saying:
“Can you prepare for a fantastic salsa evening with my girlfriend? 🍷”
And the home understands lighting. Heating. Music. Ambience.
This is amazing. 🤩
As a user in the coding space claude code is good and stabile. But when are you beginning to see the potential in this as a process engineering tool?
Due to the limitations of the current framework, i had to reinvent the wheel of Claude frontend at work, to accomodate the strategic needs of utilizing it as part of process optimization and sharing across projects and work packages. It is stronger than any framework with the right tweaks, and we unlcok domain experts to automate their own processes to some extend.
Have been focusing on building a digital twin of my house.
Fitted thermal model to MAE 0.2 degrees after Bias correction on a 24 hour prediction.
Trained my RL model to optimize energy costs, comfort and air quality with Optuna optimized reward finetuning.
Will keep improving as more data is collected.
@Yuchenj_UW Claude code CLI has been working fine all day for me. Claude via desktop doesn’t.
But overall the quality of the model is really bad. Like for real
Seriously.... im gettings so fgrustrated here. @AnthropicAI@claudeai i have no idea what you are doing to your models. but its so out of touch.
Its so behind now. You probably wont tell anything, but i promise you, now its so much that its basically not useable.
Hos os gør det intet.
Efter at have implementeret vores AI infrastruktur fundament, er det vigtigste:
At de er specialister i deres afdeling og forstår processerne. At de kan kommunikere og nedfælde logik i et simpelt AI framework, som jeg har introduceret. (On top på noget lign. Claude Code)
Det betyder, at vores specialister, som har 0 AI viden, nu ´har bevist, at de kan lave deres egen workflows og automatisere ret komplekse opgaver med meget få fejl.
Det var lige input fra virkeligheden.
@SteffenFrolund Eller en som har erfaring med implementering af AI i erhvervslivet. Forstår ikke hvorfor startup baggrund er relevant. Det er tydeligt at se trend i AI budget, når man har været med til at definere og implementere fundamentet for AI infrastruktur i en virksomhed.