Like everyone else in the comments, I'm really loving the content you put out. I gotta say though, I think your channel videos are the only ones I've ever had to lower the playback speed lollll. I'm usually watching videos at 1.5x but you speed through this stuff so efficiently! Thanks for all of the knowledge sharing and keep the great content coming :D
The self driving car analogy is huge. I've been trying to build a fully autonomous car, and it's really hard. Maybe it's not even possible with the current technology. On the other hand when I build AI tools to speed up my processes, I'm way more productive. Thanks for that refocus!
We missed you man! Awesome work! Learning about so much great practical solutions ❤ Btw ive recently managed to get yt comments api to both get all comment threads and also active post reply 😏 I think you can have the first ai yt comments answering bot easily, especially with that free api if still works. Image i ask a question here and a few seconds later i get an answer, in a public setting for others to enjoy the research, that will be awesome
Really good content! Would be great if you finetune this bot in the second episode, with a more solid personality and better responses. Maybe changing his personality for people with a specific discord tag? Keep up with the channel!
very cool and impressive. The part about giving the agent knowledge it's a bit more rough to my noobie lvl. hard to follow. either than that this is pure gold
Great content!!! Thank you very much!! Could you please make a video about llama index? How we can use it in a full project, store data, and regularly update data, and query based on new/old content.
Thanks again Jason for this amazing content! I am trying to join the Naga AI discord but getting a "unable to accept invite" message, would you be able to advise on how to go about this?
I love your content! Thanks for sharing. One thought: The memory function is very clunky. AI is built to mimic humans, but humans do not typically remember every word they hear. Would it be possible to retrieve a snapshot of the node excitation state from a prompt, perhaps even judging the degree of excitement, as a more efficient memory? Perhaps save the instantaneous impact of the query on the A.I., perhaps coupled with some Meta brief that triggered it?
careful with the api keys, they probably get information request as well as its a rats nest of malware to get it working. only try and retrieve it if you know what you're doing
I do not. Just follow the rules of thumb Ive picked up along the way. Its not information id apply to my ai agents so I'm not sure I'd put it in a vectorestore@@ericcaptures
What is the approximate cost to run this agent with GPT-3.5? I get that it varies based on the amount of questions, but it seems to use tokens quite intensively: for memory, search, summarization, and generating the response. Is it worth it? Is there a way to prevent the agent from running if users start asking irrelevant questions and abusing the bot?
I second this question. What would be really even better, as far as I'm concerned, is using local models. But that too may not be a practical solution as local models such as Llama, Vicuna, etc. may not be adequate replacements for GPT4. Any thoughts on how to mitigate costs? The answer to that question would open up a host of possible applications of AI for those of us who simply can not afford OpenAI's costs.
I cannot get my open ai key to work. i tried ones from that discord and one of my own. The error is a key error on the json result item. error shows as credits_limit_reached even tho i just made the keys
Hey any one has a suggestion to fill out forms on the internet automatically? I thought about using zappier but some forms are rendered server side so the text box ID is not available in the source code inspection?
@@AIJasonZYes! A Groceries Store Worker, which can take a full shopping list, compare it with against a dataframe of items that are in the store, and address real world cases such "out of storage, offer relevant alternative", "solving ambiguities" (we have 3 brands for x, which would you like?), Etc etc while having SalesGPT esk stages like welcome, take_order, generating_bill etc etc Hope you like it, all the best!