Great demo, I started doing something similar in Slack using Block Kit, I just let the AI figure out the best way to use it based on the data it's trying to show me, works surprisingly well.
Interesting idea! This was actually our initial concept for Canvas (we referred to it as 'GiZmo' internally), but OAI shifted direction to focus on AGI instead of workspace products. Since developing AGI is our priority, UI updates can take longer. If we created all solutions in-house, it might reduce the value of our API. Your work looks great-keep innovating!
Most users dont know what they want nor cant clearly describe it. But they know what they want when they see what they dont want. So starting with a blank slate could be intimidating to normal users. They nornally need direction. But the future looks blissfully 😊
To improve your website, you could have a main user request input box for the first tile, just as you did. Then, add an "Add Tile" button below it that creates a new input box underneath. This would allow users to modify the already here tiles if they want to add or remove content.
This is what alignment is for, plus system prompts & a router LLM can aid in preventing malicious activity. Additionally, at least in the beginning, it will be prudent to only allow Agents to run arbitrary dynamic code in a sandboxed environment such as with Docker, otherwise Agents should always try predefined tools as much as possible. Regarding “deleting a project” it will be important for realtime file backup & git commits in addition to safeguarding certain file systems requiring human approval before execution. Ideally, agentic systems will be much like a well trained executive assistant that understands the user intimately. Eventually we will be able to interact with Agents using brain interface systems. Apple has already filed patents for a version of AirPods Pro with built in sensory capabilities allowing for measurements of brain waves and other faint biological signals where an AI model will be able to interpret said signals in real time sending commands/prompts to the users agent(s).
I had a similar idea, I called it web infinite(play on web 1,2,3) the concept is server is just a vector db with 2 end points. 1st endpoint is to request data, the 2nd is to get data, imagine one is what the client needs and the other is what the server wants to tell the client. The client side has transformers.js with a small phi 3.5 or llama3.2 model and all it does is take the raw text from the server and generate html to display the information.
how can i build an avatar i can talk to and control my computer??i have been scouring the inter net for a year and have not found the right way to build it ???
Great demo, I started doing something similar in Slack using Block Kit, I just let the AI figure out the best way to use it based on the data it's trying to show me, works surprisingly well.
Kris, please share the video and code with members. can't wait! Thank you.
Interesting idea! This was actually our initial concept for Canvas (we referred to it as 'GiZmo' internally), but OAI shifted direction to focus on AGI instead of workspace products. Since developing AGI is our priority, UI updates can take longer. If we created all solutions in-house, it might reduce the value of our API. Your work looks great-keep innovating!
And you want us to believe that you really work for OpenAI, breaking your NDA for the whole world to see??
another very good demo backing personal software. well done again
Most users dont know what they want nor cant clearly describe it. But they know what they want when they see what they dont want. So starting with a blank slate could be intimidating to normal users. They nornally need direction. But the future looks blissfully 😊
Hence natural language interaction
To improve your website, you could have a main user request input box for the first tile, just as you did. Then, add an "Add Tile" button below it that creates a new input box underneath.
This would allow users to modify the already here tiles if they want to add or remove content.
In this setup, could the user create a malicious prompt to say delete the entire project
This is what alignment is for, plus system prompts & a router LLM can aid in preventing malicious activity. Additionally, at least in the beginning, it will be prudent to only allow Agents to run arbitrary dynamic code in a sandboxed environment such as with Docker, otherwise Agents should always try predefined tools as much as possible. Regarding “deleting a project” it will be important for realtime file backup & git commits in addition to safeguarding certain file systems requiring human approval before execution. Ideally, agentic systems will be much like a well trained executive assistant that understands the user intimately. Eventually we will be able to interact with Agents using brain interface systems. Apple has already filed patents for a version of AirPods Pro with built in sensory capabilities allowing for measurements of brain waves and other faint biological signals where an AI model will be able to interpret said signals in real time sending commands/prompts to the users agent(s).
i think apps will be soon obsolete if ai integration is done at the OS level ... but ill stick around and see it with my own eyes
Two years.
@@pandoraeeris7860 i hope less
I had a similar idea, I called it web infinite(play on web 1,2,3) the concept is server is just a vector db with 2 end points. 1st endpoint is to request data, the 2nd is to get data, imagine one is what the client needs and the other is what the server wants to tell the client. The client side has transformers.js with a small phi 3.5 or llama3.2 model and all it does is take the raw text from the server and generate html to display the information.
sounds great lets collaborate...and build it.
how can i build an avatar i can talk to and control my computer??i have been scouring the inter net for a year and have not found the right way to build it ???
someone let me know when code shared members please
this just change everything bro
any reply will be appreciated!!!
this change everything
I just want to know when agent can use google flight to book air tickets for me automatically
Are you really going to trust someone else to book a flight ticket? And how often do you even do that?
2 seconds ago? Hype
interesting - )
Every App will soon BE AGI!
AI AGENT , has a long way to go .
Waste of time. No idea what he did here
Nice, you just reinvented websim but worst looking.