Why is Lotted Smart Chat charged on a token basis?
Drop Chat uses OpenAI's GPT-3.5-turbo (GPT-4) model for processing，Their API is charged on a token basis，So because of the upstream charging mechanism，We also use the same charging model。
What is a Token?
Basically, a token is a word，Some words may require a combination of two tokens。Token is a thing that you can understand as a number，In this way, the content encoded in this way can only be understood by the GPT model，And the GPT model generates content，It also needs to be decoded by the corresponding token to become a text that humans can understand。
GPT 3.5 and GPT 4 What's the difference?
3.5 and 4 are all big language models of OpenAI，As of now， 4 It is the most advanced version。Compare the two，3.5 One step ahead in speed and price，cheap，And fast! However, in terms of derivation ability，4There has been significant progress。But in general，Most user feedback is on daily use，It was not visibly felt 3.5 and 4 of the difference。Other than that 4 Support for longer contextual support，Up 8000 token，whereas 3.5 can only be supported 4000 many。
Also due to GPT 4 The model is larger，Computing also requires more resources，So for now 4 The answer is slow，And the overall capacity is poor，It can only be supported per minute at most 200 times to ask questions，Comparatively，3.5 is supported per minute 3500 times to ask questions。（note，The number of questions asked here，is the global number of times，Not the number of times per user。）
What can you do with Drop Grid Chat? Pure chat?
Probably...... be。But，It is different from traditional chatbots。Because behind it is a GPT-3.5 large language model，So it can "understand" and process the text content，In addition to the dialogue，Tell stories，Generate text content (writing text、Write、Write poems and songs) outside，Can also translate、Retouching and polishing, analysis, summary and so on and so forth...
Of course，There are also interesting operations, such as pretending to be a terminal emulator，Pretend to be an input method，Take your logic directly to generate the code，Give you a creative idea about xx projects and so on。
Anyway，You imagine it as a real netizen，Word processing tasks that can be explained via text chat，This "netizen" can basically be completed。
⚠️ Note.，Since GPT-3.5-turbo (GPT-4) is an offline model，So the latest press materials, etc. are not available，You ask it and it will be made up。So，For some serious facts，Don't take it for granted，Do a good job of fact-checking。
Is there a difference between asking questions in Chinese and English?
Yes。
GPT is a large language model，It is trained from the network's public text content。Information not available online，It doesn't know。Less information online，The answer it gives is inaccurate。Compared to English，Chinese has relatively little internet content，So when using Chinese to ask questions，The accuracy with which GPT gives the answer、velocity、The degree of perfection will be relatively reduced。So if possible，Please try to ask questions in English to get a higher quality，More accurate results。Even if you don't speak English well，It's also smart enough to understand what you mean，Includes grammatical errors，Spelling mistakes and so on。(Of course, you can also let Drop Chat help you translate your Chinese questions into English.)，Then ask questions in English in a new conversation... ）
How is the networking function of Drop Chat Chat realized?
GPT is an offline language model，It itself cannot actively de-network (Otherwise, the era of Skynet will come）。Whenever you start a new session (for mobile and TG Bot，is the first question after clearing the conversation history)，Drop Chat will do an additional check，That is, whether the question involves factual information，For example, the definition of xxx is something like that，If yes，Then the model will generate relevant search terms to Google to search，We add the titles and biographies of the top three Google search results to our history，This is when GPT responds，There is a reference，In general，Most of the questions will not be made up。Since this operation is to inject search results into the chat history，If every message is processed like this, it is a waste of points，Of course, the significance is not particularly great。So we only have it in each sessionFirst questionThat's it，The results of the search are always saved to GPT，That way, even if you ask more in-depth questions later，It also has a certain probability of answering correctly。
⚠️ Note.，Although we try to make the GPT model answer as accurately as possible，But that doesn't hide the fact that it's actually semantically generated content，So if necessary，Please also customize the verification，You can't rely on any of the answers AI gives alone。
Why are some problems shorter but cost more points?
The number of points consumed does not depend on the length of a single answer by the model，but in this session，The total number of messages you sent and that the model answered。Drop Chat is context-friendly，And every time a message is sent，All history in this session is entered together，So that the model recalls what was before。
How can I save points?
Every time you send a message，Drop Chat will pass all the history history in this session as input to the model。This results in more context，The higher the cost of each answer。Model processing up to the maximum 4096 (GPT-4 is.) 8000 ) tokens，Roughly calculated，It is 0.08 Counted points。So try to keep the same topic in every conversation，There is no need to keep context on different topics。As a matter of fact，Most of the time，A single inquiry is enough to get the right answer。
Is there a limit to the history length of a single session?
Yes，As mentioned above, the maximum length of text processed by the model at a time is 4096 (GPT-4 is 8000 ) tokens，In fact, the conversion to words may not be enough 4000 word，So all the text for the entire session adds up to about the same at most 4000 A word is the maximum length that a session can support。Of course，This does not mean that the session cannot continue，If this limit is reached，Drop Chat automatically discards the oldest historical data to make the request conform to the model limit。
Where is my chat history stored?
Currently, there are two clients in Drop Chat Chat，They are the web side and the Telegram bot side。
- Web side：Your chat history is stored in the local storage of your current browser。There are also tokens that you log in to and so on，If you clear the browser storage，Then the data will be erased。Of course，When you ask a question to Drop Chat Chat，All history of your current session, including new questions, will be sent to Drop Studio's servers，However, this data is not stored on the studio's servers，They are immediately forwarded to OpenAI's GPT Model API， When you get a reply from the model，This data is also returned to the client immediately。The server of Drop Studio only forwards and records the number of tokens for billing。
- Telegram Bot 端：Your chat history will be replaced by [plaintext] stored in the Amazon Web DynamoDB database under the drop studio account，At most, it will remain 30 bar (Current settings，Message records may be dynamically increased or decreased in the future，Include messages that you send and that the model replies to，This makes it easier for you to pass the message to the model as a historical message after sending it。We strongly recommend that you use the /clear command to clear the background history after the question is over，That is, to protect privacy，It also saves points consumption，It can also be convenient for the next question。
Is my account information secure?
Safety。Rest assured, we do not store your password in plaintext ：）
As a matter of fact，Your account information is not in the database of Drop Studio。They are stored in the Amazon Web Cognito service in the United States，After you authenticate and log in through this service，You will get a JWT token，Whenever a server for Drop Studios is requested，Drop Studio's servers also go to the Cognito service for verification，This perfectly protects the privacy of your account。
What is your website and backend used for? Is it safe?
- site：Develop with React，Deployed in an object storage container in the Amazon Cloud，Use the CloudFront service for distribution using Amazon Cloud。
- server：Backstage is Flask，Run with Gunicorn，And use Caddy for inversion，Run on a Lightsail VPS in the Amazon Cloud。The data is stored in the Amazon Cloud's DynamoDB service，The entire backend server is stateless。
- Telegram Bot：The server is also running the Telegram bot，A webhook callback is used。
- Desktop client：Wrap your React App implementation using Electron technology。
- Mobile clients：Wrap your React App implementation using Ionic Capacitor technology。
Why does it sometimes stop talking halfway?
OpenAI's APIs sometimes do this，This situation is more likely to occur when the network quality is poor，The technical detail is "ChunkEncodingError" and the upstream framework is currently actively discussing the solution。So before there is an upstream solution，The drop of the smart chat will end early to avoid reporting errors，When you encounter this abrupt end，Just send a "continue"，“continue”，The model will continue to generate incomplete results based on historical data。
Does Drop Chat have a mobile app?
This one can have ：Drop Chat supports macOS iOS ChromeOS Windows Linux Android Client，Please move to https://im.logcg.com/chat-ai Download。
Of course，You can also use your mobile browser to open Drop Chat Chat https://chat-ai.logcg.com，iOS uses Safari，Open and click Share，Point Add to Home screen。This way when you open the drop chat from the Home interface，That's the app。This technique is called PWA，You can make a supported web app run like a native app。
Android uses Chrome based on any browser，Open the menu after opening，There should be an option to install the app，After clicking, it can be installed normally。The same is PWA technology，Does not take up space，Green environmental protection。
⚠️ Note.，Mobile is the same as TG Bot，Multiple session switching is not supported。
What features does Telegram Bot support?
Compared to the web version，Telegram Bot has an exclusive feature，This is to edit one of the messages in the history (provided you have not cleared the current session)，This allows the model to regenerate the answer。
- The web version and client support code highlighting，Bots are not supported。
- The web version and client support multiple session switching，Bots are not supported。
- The web version and the client store message records locally，Bot message records are stored synchronously on a background server。
- The web version and client do not support editing history messages，Bots can edit historical messages to regenerate answers (not if the background record is cleared)。
Can Drop Chat provide API access?
Drop Chat supports API access，But when you use API access，Advanced features of Drop Chat Chat，For example, the online search mechanism and the built-in prompt will not work，Everything is subject to the configuration sent by the client。This is done to improve compatibility with various clients。
What clients does the API support for Drop Chat Chat?
Theoretically, any third-party GPT client，As long as custom domain names are supported，can be accessed。Currently known supported clients are (to confirm support time ordering，Including but not limited to)：
- OpenCat
- Bob third-party GPT translation plugin
- AMA
- BlenderGPT
- Better ChatGPT
- Bob's official GPT translation plugin
Are there any usage examples?
- How do you say xxx in English?
- (Simple math)
- 12How much is USD in RMB?
- Can coffee grounds be used as fertilizer?
- Write a helloworld example in Python
- I want a Rust program，This program reads a csv file and adds 20 in the "value" column and outputs it。
- My code calls the xxx framework，But call xxx() When an error is reported:Xxx.，What could be the cause?– Is there a solution?– Can you elaborate on the meaning of the code in the example?
- My daughter is celebrating her 7th birthday，Any ideas for celebration?
- With rabbits、Turtle、Write a joke with birds
- I have an xxx product，Help me write a promotional copy。 — Change to the style of Little Red Book，Emoji with emoji
- Due to xxx reasons，Our logistics are delayed，Please write an email to the customer in English，The tone is tactful。
- Help me write a letter...
- This is the application letter I wrote，You help me polish it，The content is：......
- I'm sad now，Can you talk to me?
- What is the general idea of this article?
- ......
