Yearly Archives: 2023

Just a few days ago，OpenAI has released new models... Guys。Settler Zhichat also made the first adaptation，Let's take a look，What updates have been made to Roger Zhiliao： Model GPT-3.5 Now GPT-3.5-Turbo is merged with GPT-3.5 Turbo 16k，The new model has only one GPT-3.5 Turbo，The new model is an input-output asymmetry model，The input supports up to 16k，The output supports up to 4K。 GPT-4 has two new models：GPT-4 Turbo and GPT-4 Turbo with vision，apparently，Relative to the former，The latter supports the ability to read pictures。 It's a pity, though，Both models are still in preview，Especially the GPT-4 model with the ability to read pictures，Only 100/day number of requests is supported，It can only be tested by developers (probably not enough)。Fortunately, GPT-4 Turbo's request volume has now increased 10000 Plus per day 500 Per minute。 Of course，GPT-4 Turbo can read pictures in addition to that，Inputs up to 128k are also supported，But sadly, the output is relative to the previous 8k，Shrinkage to 4k。For now，It seems to be a little faster than GPT-3.5 in terms of speed，Of course, it may also be due to the limitation of the number of user requests。 Price The price of the new model has been reduced again! The price of the model of Lag Zhiliao has also been reduced in the same proportion，Since GPT-3.5 16k has been merged into GPT-3.5，about […]

Now，Recharge drop Chat opens 618 Group buying promotions，One person in a group，Group purchase price super discount! It's time to count a wave of points for you in the second half of the year。     Recharge Drop Grid Smart Chat，Add power to your work and life with the world's leading GPT4，Efficiency is a boost! Drop the smart chat point group purchase price： 1500 Points ← 618 Yuan.，Four folds! 200 Points ← 61 Yuan.，Triple fold! 50 Points ← 18 Yuan.，Thirty-six fold! 18 Points ← 6 Yuan.，Three-three fold! Such a lot of force，Words of conscience，I can't afford to play a few times a year。

Just today! OpenAI has released a new wave of model updates，Now，Drop Chat already supports the newly added GPT-3.5-turbo-16k，The model supports up to 16000 The amount of memory of the token，Cheap price makes more application ideas possible! From now on，You can use GPT-3.5-turbo-16k in the Drop Chat API，Including Smart Chat platform and API platform。   Other than that，This update adds the function call function，We also supported it。The API interface is now perfectly adapted，Including stream and non-stream request modes~ That's right，At present, the full range of models supported by Ledger Zhiliao have been updated to the latest 0613 Version，It's a pity，The API of Drop Smart Chat does not support specifying the old model，When you send a request，The specified version number is automatically replaced with the latest version。   Pricing is the same，OpenAI's impact on GPT 3.5 Serial models underwent a wave of price adjustments，We've also updated it as well：The original GPT-3.5 model，Unified by input and output 0.02 Point each 1000 Token price reduction is 75% off 0.015 dot，The output remains the same still 0.02 dot；For the newly joined 16k version of GPT 3.5 model，Enter as […]

  Now Drop Chat supports the second networking mode，Before that，ChatAI supports online search (Google search) on the first question of each session，to make some of the answers more accurate，This greatly reduces the scenario in which the model has hallucinations (fabrications).。 From now on，ChatAI can directly read the page you pass into，This allows you to ask questions about the content of the page、analyse、summary、translation and so on。Web links are matched and extracted using regular expressions，So when you use this function，It is best to separate the link from the body content with spaces，avoid extraction failures。 Other than that，ChatAI supports multiple links to read，There is no upper limit to the amount that can be extracted here，But for the sake of your points and model maximum processing length，It's better not to commit too much in a session。 Will reading a web page cost extra points? – No，But the content itself certainly costs points。 Is this feature also limited to the first question per topic? – No，This feature is unlimited，But reading many pages on the same topic will quickly fill up history，This leads to higher prices，Please make appropriate use of this function。   Of course，This feature also has some limitations，This needs to be clarified，Please be aware when using this function： Due to model handling content length limitations (GPT 3.5) and price issues (GPT 4)，Content reading is roughly limited to 800 non-english word (also English word) approximately 1000 token around，If the content is too long，The content behind will be cropped，It will not use the GPT model to summarise，This is also due to price and efficiency； Since it is the server that requests the web page content，Login required、Pages for authorization, etc. are unreadable — even if you give your account number and password (strongly discouraged)，It must be a publicly and anonymously accessible web page to be read by ChatAI； Since HTML itself has a lot of invisible content for functionality、Style and so on，These invisible texts are washed，to ensure that the token (i.e. in 800 word space meaningful body content in the word space as much as possible)，We currently only get [crayon-663a7e04ba5f4946045298-i/] and [crayon-663a7e04ba5fd677093975-i/] wrapped content，links will also be removed。     At last，I hope ChatAI can improve your productivity，Good luck in your work! Drop Studio chat-ai.logcg.com www.chatai.beauty www.chatai.lol

GPT-4 is already arranged! Right in 3 A day ago， OpenAI released smarter, more powerful (and more expensive) GPT-4 models，Drop Ge Smart Chat applied for API permissions at the first time。After receiving the mail this morning，Drop Ge Smart Chat supported the model for the first time。 Now，You can use the GPT-4 model on Drop Chat，The default is the latest model。--of course，Considering that the GPT-4 model is more expensive，And relatively slowly，We provide an option for model selection，If necessary，You can also switch models for a lower price，Faster GPT-3.5-turbo. This setting now supports cloud synchronization，Set it once，Telegram Bot will also take effect。   Limitation Since this is the first GPT-4 model，As far as the test results are concerned，It has some more obvious limitations： The model firmly believes that it is GPT-3.5-turbo and not GPT-4，However, you can tell from the amount of points consumed 3.5 still is 4； The model is slow，But it's actually acceptable，In the near future, OpenAI will improve speed optimization； Model API processing capacity is small，Currently, GPT-4 requests are limited to 200 Per minute，Than 3.5 target 3500 said，A lot smaller。If you keep getting errors，Please switch to GPT-3.5 for the time being； At the moment we only have GPT-4 […]

One fold! All right，The main reason is that the upstream (OpenAI) has cut prices。 OpenAI has released a new chat-specific API，It's called Turbo，The version went up to GPT 3.5. However, surprisingly，While greatly improving the speed of computing and response，Operating costs are also significantly reduced。The direct payout is that the price of this model is one-tenth of Davinci，That is, 0.002 Dollar 1000 Token。 I tested the model as soon as it was released，How much accuracy is improved is less clear，Anyway, the API format has changed a lot 😂。And then the first feeling was... Fast! React made a web app before it could hide elements based on state, and it was already shipped! - even Chinese is just as fast。 Of course，There are still some differences between this model and GPT-3，For example, it tends to ask for further instructions rather than answer immediately，This is what GPT-3 does not have，The latter will immediately begin to answer。 Anyway，Because a new model was applied，Now the drop grid wisdom chat is also folded in sync，As you read this，All changes have been implemented，Have a good chat!

Let's start with the main points：Register an account with Loge Smart Chat now，Charge a hundred get a hundred free，The event will end at any time，To charge is to earn! Since the launch of ChatGPT，This whirlwind of general AI has also been blowing for months，After trying out this epoch-making chatbot，I realized it right away：If you can't master such tools in the future，Then your productivity will be like the difference between using a computer now and using an abacus to do calculations in the past。Of course，For now, this model is only in the form of a chatbot，But soon，It will become a reliable tool。 After I found out that ChatGPT is not actually available in China，I decided to do a similar one - OpenAI opens up the model API anyway，And it is not forbidden to be used as a chatbot。 In just a few months，ChatGPT has sparked debate in all walks of life，I don't think there's a need to choke on food。opposite，multipurpose，Feel more，You'll see that as a tool, it's not a panacea — it may even give the wrong answer，The analysis given may also be copied ... But once we understand its advantages and disadvantages，It can be applied very well。For example, its answer is unreliable，Let's check it before using it；In addition to answering directly，It can also be summarized，Optimize article structure，Creative outlines and so on and so forth。As a programmer，I myself often use it to solve my code problems，Although many times the results it gives are not necessarily correct，But often there is a good idea that allows me to find a solution quickly。This is something that no other tool can do。 👍 Other than that.，I also found out since GPT is an offline model，So it's hard to answer some definitional questions accurately (e.g., everything you ask Siri or Google Assistant)，To this end, I tried to add a networking mechanism to Drop Chat Chat，When encountering such problems，It will first generate search terms Google，And then answer - without even loading into the body of the article，All you need is the website title and introduction，It will be able to distill the key knowledge points and answer your questions。(Obviously，New Bing is also such an implementation logic，Even the search term was shown to you) 🎉 Since the GPT-3 model is expensive，per 1000 token is required 0.02 Dollar，Therefore, Drop Grid Smart Chat can only use the pay-as-you-go charging model，Actually every 1000 I earn a little bit as server costs and currency conversion costs，You charge the money，In the end, it was all spent on OpenAI。 💻 For the convenience of everyone，Drop Chat supports two usage modes: Web and Telegram Bot，The web version can actually be added to the home page of the mobile phone and become a PWA APP，This will not display the browser's address bar and toolbar when opened，It is almost as local as it is used […]