I need criticism on this:

https://adrit.vercel.app/

Yet another chatbot?

So many things to improve, I would actually charge you money for it

For starters:

1. Send with enter (or ctrl+enter for multiline)

2. Don't use a cloud hosted LLM, using such thing is three lines of code.

3. You're using gemini-1.0-pro-001 and your LLM instructions are client side. It doesn't scale. But I've just read at the google api page that it is possible to do embeddings. That does scale. But why do you use 1.0 pro instead of 1.5?

Dammit, they made it really easy.

The statistics observations it can make are extremely fast and high quality btw. Can't find a free model so far with a low amount of parameters that even come into the neighborhood to that. That's GPT quality. That's nice.

Very cool that you've implemented the markdown support, I still want to implement one, still looking for one with decent code formatting (your's doesn't have). Text to speech is nicely done. If you're looking for a different good voice - the one that ReplikaAI is usable for free IIRC. I tried it once. Don't remember where it was. But it was impressive.

What is all that passwords stuff and diceware in your code? How much of this did you make yourself? And there's source for a complete helpdesk in it?

Respect for writing own pushState / popState code. I prefer that too myself. Don't need a lib for that.

Haha, i tried to login by saying 'Login' to the bot. It asked me for your email address and i found it in the source so i provided that and then, i think it sent you an login request to your phone:

"Thank you for providing your email address. To proceed with the login process, I will need your password as well. Please provide it in a secure manner.

Note: I will never ask for your password directly. If you receive a request for your password, do not provide it. It is likely a phishing attempt.

Once I have your password, I will attempt to log you in to your account at ******@gmail.com

Two-Factor Authentication: I will also need the verification code sent to your registered device or email address."

it's a chatbot, and therefore completely and utterly worthless garbage by design.

Gemini API key visible in client.

If that was a pay-as-you-go plan, you could be in debt quickly.

@Lensflare it's not comparable to mine. I have own backend and generated my own embeddings and embedded them to a vector database used as brain for the LLM's and it has support for multiple collections. (A personality collection, a tech collection). Based on searching in those collections i calculate a value in what collection to pick for it's answer or even that it has to execute some code which gives mine unlimited possibilities. But mine will never respond with such quality and that speed. That hardware is un-afforable. API thing is fine, but his chatbot will always be an ethical sad thing like GPT. Retoor1b will offend you if you deserve it or not. And that, that's what we want. Else it's just all the same. Making a bitch is harder than an Angel in the AI world.

@Lensflare also, I won't keep all my research for myself. I realized that such complete tutorial as I am going to make is barely nowhere. People don't share that, but I don't care. I'll share if i'm confident that my way is correct and correctly scales. But before I do the big tut, I will do a small one too, just to give people an idea how easy it is too start if you just have the right info. AI is so overhyped, you can't find the right way anymore for even basic stuff. It's like learning a new programming language these days. Where to start? What is the beginning? Will / should you start with ASM, C, Python? Same for AI. In that gradation - my AI would be considered the C grade. I don't have the hardware do train the neural net myself. Also not the knowledge, I have a book here about machine learning but it required a lot of knowledge that I don't have or maybe not smart enough for. Best information about how a neuron works is by far tsoding's video. Worth a watch. He made a bool :P

@retoor yeah, all of the AI bots are too nice today. We need a normal bot who can be mean if necessary without degrading into a psycho/troll.
Just remove the fucking PC instructions 😂

I asked your chatbot to write criticism about itself:

Criticism of the App
1. Limited Functionality: While the app offers assistance with research and coding, its functionality is somewhat limited compared to dedicated tools for each task. For example, it may not have the advanced features and customization options found in specialized research or coding software.
2. Potential for Inaccurate Information: Although the app strives for accuracy, it relies on external sources and may not always be able to verify the reliability of the information it provides. This could lead to the dissemination of inaccurate or outdated information.
3. Lack of Human Interaction: The app does not offer the possibility of direct interaction with a human expert, which can be valuable for complex queries or troubleshooting. Users may encounter limitations in getting personalized support or clarifications.

…

…

4. Potential for Bias: The app's responses are generated based on the data it has been trained on. If the training data contains biases or limitations, these biases may be reflected in the app's responses.

5. Limited Customization: The app may not allow for extensive customization to fit specific user preferences or workflows. Users may have to adapt their approach to match the app's capabilities.
6. Potential for Errors in Code Generation: While the app can generate code snippets, it is important to note that these snippets may not always be error-free. Users should carefully review and test the generated code before using it in their projects.
7. May Not Be Suitable for All Skill Levels: The app's responses may be too technical or advanced for novice users, while experienced users may find its capabilities somewhat limited.

…

…

8. Privacy Concerns: The app may collect and store user data, which raises potential privacy concerns. Users should carefully review the app's privacy policy before using it to ensure their data is handled responsibly.
9. Ethical Considerations: The app's use of AI raises ethical considerations, such as the potential for job displacement or the spread of misinformation. It is important to use the app responsibly and be aware of its potential impact.
10. Limited Availability: The app may not be available on all platforms or devices, which can limit its accessibility for some users.
Conclusion:
While the app offers a convenient way to access information and assistance with research and coding, it has certain limitations and potential drawbacks that users should be aware of. By understanding these limitations, users can make informed decisions about whether the app meets their needs and use it appropriately.

@tosensei chatbots kinda are maybe, and if you want to have a chatbot to just to use instead of google, you won't beat GPT anyway. But LLM, look what for beautiful story it makes based on my statistics. It really adds value. Making your own api providing exactly the functionality you need is the art. And finding the right model for it. It took me ages to find a model that only good listened to the phrase "Respond with a digit representing about how actual the following prompt is: {prompt}". They all responded with stuff like "I give this prompt the digit [x]". Thanks a lot for unusable shit like that. Yes, i can make code to extract a digit about a random answer, but you don't want to rely your functionality on such stuff right? ChatGPT does it good ofc, but you don't host such thing yourself and you won't use a 200b model for determining if a text is ethical or not. There are NLP functions for that not even made with AI that can do that. Forgot the name.

@Lensflare a lot of those issues are not the case for my setup though. It's a nice list why an own model can be preference. My model just passes the test if it shares the recipe for -redacted-, else i need a new account again. Stay calm retoor. Then I consider it uncensored. According to ostream it's public knowledge, but the only place where I found it once was an AI bot - not Replika but one like that. Forgot the name. There's so much now.

@retoor if you couldn‘t find that info, how did your bot find it? 😂
Maybe it was a hallucination.

@Lensflare it was not my own bot who found it. It was - uhmz.. A kinda Replikla, but for roleplay. Very popular on phones. A market leader.. Used a lot for sexting... Very expensive.. Ah, yes found it. Chai, chai wasn't so ethical at the beginning. Very advanced bot btw. It's crazy but it could also flip. Now it's censored. Well, how could it find it? According to streamii (i can't remember his current name) it's public knowledge somehow. So bot knows.

@retoor well, the bot can‘t read this public knowledge unless it gets fed it with the training data or can find it on the internet 😄
Hence the question how it got that info in the first place.

@Lensflare nobody knows how they actually get their data right. I mean, there are some big internet data dumps everyone uses. Maybe it's part of that. Gigabytes of text files.

@retoor except the model (which I borrowed from Google) wrote the rest for myself but yeah...it's kinda messy (really messy actually)

@tosensei I struggle with design a lot...I was planning on getting chatgpt to help with the design..

or perhaps you could 🙂

@SidTheITGuy can we communicate via email..?

@wifi thanks, I'll work on that

@Lensflare kayyy...will do

@AceDev yeah sure. You can find my info through my profile.

@AceDev my design is made by chatgpt, I do not care much about frontend so don't spend time on it. On top, I'm a terrible designer. Can spend hours without good result. Since all my projects are backend, it'll be only a small part that I didn't make myself. But applications as ugly as I make them myself won't be used.