The advent of ChatGPT has made the big model the focus of attention in the industry, whether it is a large factory, a small and medium-sized enterprise, or an independent development of open source programmers, they are riveted on the big model. In addition to focusing on the model itself, what areas do they also focus on about AI? The author of this article has analyzed this, I hope it will be helpful to you.
After the advent of ChatGPT, major companies have worked the "big model", and model capabilities have become the most concerned topic. Although ChatGPT directly defines the product type of generative conversational bot based on large models, the strength of its model also causes everyone to ignore its problems in product use.
In fact, whether it is a large factory of low-level model development, a small factory focusing on the application layer, or an open source independent developer, they are secretly competing in product experience outside the model ability, and participate in this dialogue revolution that can define a new way of human-computer interaction with different attitudes.
So besides the model, what are they "rolling"?
First, the big factory: focus on the model, the experience is slightly "simple", the main one is more
1. ChatGPT: The three-piece framework definition is minimalist
As the industry benchmark, ChatGPT directly defines the industry standard for generative conversational bots. In the early days, the free version of ChatGPT only had conversations, conversation management, and simple theme setting functions, and swept the world with model capabilities. With the increase in users and complaints, ChatGPT's product manager couldn't sit still and began to gradually add some features.
The most impressive is Upgrade to Plus, which firmly occupies the setting page in the lower left corner, and there are only two words at a glance, "top up". In addition, ChatGPT has also gradually added functions such as theme setting and privacy management, but it still highlights a restraint, quite a true biography of Xiaolong Brother.
The main differences of the paid version of ChatGPT are the addition of a button to switch models, and the addition of the Beta Features feature in the settings, which can use the networked search function Browsing or use other plugins.
2) Bard & Bing: Non-differentiated competition
The remaining two of the Big Three, Google Bard and Bing Chat, also have their own styles, but generally remain minimalist.
Bard's functional design logic is almost the same as ChatGPT's: dialogue function is the mainstay, integrating necessary conversation management, account management and interface setting functions. As a search master, Bard also introduces the function of search on the basis of dialogue, which can naturally introduce search according to the problem, and present the results of the fusion processing, and the experience is more up to date. For the weather, real-time information queries are all good.
Large models generate answers with multiple answers and corresponding weights, Bard's "View Other Answers" feature gives users more choices, and when I ask about the weather in Beijing, it even gives a text version, a table version, and a minimalist version, which is very amazing.
It is worth mentioning that Bard's answer generation is a one-time generation in terms of interaction rather than typewriter mode (text appears one by one, with typewriter effect, refer to ChatGPT), lacks the generation process, and the experience is not silky enough.
Unlike ChatGPT and Google Bard, Bing Chat's positioning is search-based, and it works by summarizing search terms from input to search and then summarizing the search results. Each response cites the corresponding source, which also leads to a lack of creativity, but the source is available and accurate.
Bing is also minimalist, and the overall product interface design is not much different from the other two, but the colors are obviously more lively. Functionally, there is not even a function of conversation management at the beginning, and you will empty it after chatting without leaving a trace. Later, the "Recent Activity" feature was added, allowing users to see their recent conversations.
Bing's only feature is that it sets three dialogue styles: more creative, more balanced, more precise, and the ability to determine whether the model is more accurate or more creative based on this parameter, which corresponds to the advanced parameter Temperature in the GPT model. However, because it is impossible to compare several modes at the same time, it is difficult for users to feel the difference in the user experience, even if it is more creative options, limited by Bing's more search-oriented positioning, the search results still lack creativity.
In contrast, the main competition of the three giants is still model capabilities, and ChatGPT is still relatively leading; Bard is a better combination of search and generation modes, and the experience is better, while Bing lacks the highlights of use, but a GPT-4 model is enough to attract most users.
Second, small and medium-sized factories: deep cultivation experience, ability to experience crazy rolls
Different from the model capabilities of large factories, small and medium-sized factories and individual developers pay more attention to the application layer on the model layer, relying on the underlying model capabilities of each large factory to soar in terms of function and experience.
1) Perplexity: What to do with search-based AI conversational products
The first thing to say is Perplexity, a conversational search engine that has exploded Google Bard and Bing Chat. Perplexity, which translates to "perplexity", is a common indicator used in the field of artificial intelligence to describe the effectiveness of language models, and the less confusion of the model, the more powerful the model. Perplexity is a small, interdisciplinary team with only 14 employees, but it can be said to be a complete explosion in terms of product power, which is a model of more is more.
Building on the basic ChatGPT-style conversational architecture trio, Perplexity combines vertical search (academic, YouTube, Reddit, Wolfram, etc.), search history, search hot lists, and Copilot (AutoGPT-like). The UI design adopts the traditional search layout, with the dialog box at the top of the screen, rather than the dialog frame input box below, so that people can know at a glance that this product is a more "search" product.
In this regard, Perplexity highlights its search positioning more than Google and Bing, with a clear banner and clear positioning. The logic of the product is similar to Bing, which is a summary of search results (summerize), and the citation of literature is more comprehensive through multiple uses, which is more creative than Bing. In multiple rounds of conversation, Perplexity also added related topics, predicting other questions you may have based on the answers, if it goes well, you only need to ask once, and the rest of the questions can be solved by clicking, and the experience is smooth.
For "how to be a minority author", the author tested Perplexity and Bing Chat respectively, and in terms of answers, Perplexity's content is more detailed, there are more reference sources, and the searched interface is better than Bing Chat experience in extended reading, follow up questions, and typography optimization. Ju Hard really makes products with feet.
2. Poe: The stitching monster of "American Zhihu"
Poe is the official AI chat application launched by Quora, which accesses GPT-3.5, GPT-4, Claude, Midjourney and other AI services, and is currently the most official "AI stitching monster".
Its biggest marketing point: GPT-4 and Claude Instant can be used for free, but click on it and you will find that GPT-4 can only be used once a day and 30 times a month. In addition, in addition to stitching, the product experience of its individual services is very rudimentary, almost only has dialogue functions, and the long-term use experience is poor.
Its official AI dialogue assistant Sage is not unique in the use experience, and it should also use a model such as GPT-3.5.
3. Forefront: The best alternative to ChatGPT
Free GPT-4 is the most effective way to advertise almost all shell apps, as long as you have this feature, even if you can only use it 1 time a day, you can scam a large wave of traffic (whip Poe).
Forefront is almost free of charge for GPT-4. It comes from a GitHub project, through reverse engineering Poe, Bing and other large factories that use GPT-4, provides free GPT-4 connections for ordinary netizens, and slashes 37,000 stars on GitHub, almost the fastest growing project recently.
Forefront has a wealth of built-in personality presets to meet the diverse needs of users. The personification preset comes from a classic prompt technique: let ChatGPT play a certain role so that its answers can be more accurate. Forefront can select super personalities with one click, Da Vinci, Jobs, software development engineers, etc., providing a wealth of scene templates, programming help, creative writing, academic research and other scenarios can be satisfied.
Third, the shell application: details explode, deep player gospel
After talking about the official workhorses, let's introduce the various shell products that rely on the official API. They mainly optimize the front-end interaction in terms of model capabilities and add some auxiliary features, and users can populate and use their own APIs.
This type of product is more suitable for in-depth experiencers and as a productivity tool, and there are many details that can be customized.
1. ChatBox: The king of multi-platform clients
ChatBox is currently the most mature multi-platform AI chat client, users can independently access ChatGPT, Azure ChatGPT services, Claude API, etc., and has obtained 117,000 stars on GitHub, sweeping the client world. It is also the best option on Windows for users who are obsessed with the client.
The architecture of ChatBox is also a three-stage architecture based on ChatGPT, but each part adds more functions to meet the diverse efficiency experience. It also sets a variety of preset pormpts in the conversation management function, including software development, personal assistant, kwakwart machine and other modes.
In the settings, ChatBox also supports advanced parameters and more detailed information display: first of all, you can customize the temperature parameter to adjust the randomness and creativity of the model answer; At the same time, it can display the token usage of the API, the estimation of the number of tokens that can be input and output, and adjust the text size and default language. It should be considered the best client experience on Windows.
2. MacGPT: If there can only be one GPT client, it's MacGPT
As for why ChatBox is only the best desktop client for Windows, there is a more volume product on the Mac platform: MacGPT.
Thanks to the features of the Mac system, MacGPT supports 5 modes: Web, API, taskbar mode, global outbound call, and Intext. Any scenario can meet the needs.
Web mode is equivalent to a short browser window, and the entire user experience is consistent with ChatGPT; The API mode experience is similar to ChatBox, equivalent to a local Mac client; The taskbar mode can support waking up from the taskbar and opening conversations at any time, relying on the taskbar to ensure that ChatGPT can be quickly started in any scenario, which is very efficient; Global outbound mode supports hotkeys to call out the top dialog bar and start the conversation immediately, and anyone who has used Alfred should know how smooth this experience is.
The Intext mode is even more amazing, and it is simply a killer for word workers. When you type /gpt in any text input environment (memo, word, etc.), the subsequent content will be used as ChatGPT input, and the answer will be generated directly in the current text environment, and you can use ChatGPT for Q&A without switching applications, and insert it directly into the document, which is simply the global version of Notion AI and must be blown up.
3. ChatGPT-Next: A cloud service AI assistant for everyone
If I have to choose a web-side ChatGPT shell client, then I would call ChatGPT-Next the king of personal assistants.
Its authors not only developed this web-side application, but also developed the ability to deploy to Vercel (front-end hosting server) with one click. This means that with just an API key and a few clicks, everyone can have their own ChatGPT client, which can be used by themselves, used by the team, or provided as a service to others, all as simple as breathing.
The author successfully sent the client I deployed to my parents and grandmother, bringing them a little AI shock, and my grandmother even happily made me a large bowl of braised pork.
In terms of product functions, ChatGPT-Next is also a masterpiece, preset 20 personalities and application scenarios, and supports customization; In terms of advanced parameters, model selection model, randomness tempeture, max tokens per reply limit, and presence penalty for topic freshness can be precisely adjusted according to the demand scenario. At the same time, the historical message length compression in dialog management is also a king-level function, which can summarize the context when the number of contexts reaches a certain token, and clear the previous memory, effectively reducing the token occupation and making the conversation more durable. Finally, its interface is the most customizable of all apps, fonts, voice, send previews, compact borders, all customizable.
In terms of feature richness and customization of shell products, ChatGPT-Next is the strongest in all directions, the best feeling after long-term use, and the speed of using APIs will be faster than the official speed of various services, making it the first choice for productivity players.
Fourth, domestic large manufacturers: rapid access to the ecology, start-up companies one step faster
DingTalk, Feishu and WeChat, as the three major office IM giants in China, have had unofficial open source AI robot access so far. Through the marathon project of the open source community, a domestic startup company developed an AI dialogue robot matrix based on DingTalk and Feishu, integrating multi-modal (ChatGPT, DALL· E+Whisper, Midjourney), image creation, table analysis, document export, multi-topic discussion, formula calculation, etc. can all be implemented. Even the API is provided for free, truly empowering the business and making domestic migrant workers the first batch of players to use AI seamlessly.
In terms of ToB functions, domestic startups should be at the forefront of the world, have completed the development of productization, and can be quickly deployed according to the situation of enterprises, relying on IM to achieve office AI efficiency. One-click rapid deployment, enterprise-level AI permission management, user import and export usage records, risk word blocking, etc., allow enterprise users to use smoothly.
WeChat also has a corresponding open-source robot, but due to the limitations of the WeChat platform itself, the function of the WeChat robot is more limited, and the basic official function is transplanted to the WeChat dialog box.
It is reported that Meituan is also accessing conversational robots to help employees improve work efficiency. Its Xiaomei assistant has a number of customized scenarios and prompts built in to help employees get started quickly. The overall experience is no different from ChatGPT.
The official clients of other domestic model manufacturers are still rolling models to catch up with the level of GPT-3, and they are still in a state of concealment, and the difficulty of obtaining experience qualifications is much higher than that of mature products, so I will not comment for the time being, but it should be possible to confirm that there will be no more surprising functions.
ChatGPT: App Store Top, Mobile AI Era Arrives
Finally, let's talk about ChatGPT Buddha-figures.
On May 18, ChatGPT launched on the App Store and quickly topped the iOS download chart. The ChatGPT on the mobile terminal is also prominently simple, but the overall interaction can be seen to put some effort: the conventional dialogue interface has added the vibration feedback of the reply, and it feels that the AI on the opposite side is really typing, and the experience is bursting; At the same time, the left-swipe interaction can call out the dialogue management function, and the right-swipe interaction can open a new dialogue with clear logic; The overall smoothness of use is also very good, much better than the web experience.
However, the current mobile application scenarios are very limited, and most of the people around them use it as a wiki Q&A, without the blessing of plug-ins and networking, there are not many application scenarios on the mobile terminal, and the productivity scene The Web side is more efficient and the collaboration is smoother - after all, I can't keep buttoning my phone when I can't go to work.
However, the iOS client solves the pain point of ChatGPT recharge plus difficulty, and can be subscribed directly through the App Store, so that many users can finally use GPT-4 conveniently.
The potential of mobile is of course huge, and now ChatGPT has just come to an end, but its strategic "I want it all" can be seen. In the case of Poe and Snapchat mobile, ChatGPT has undoubtedly made the competitive landscape of mobile different at once, and we will wait and see what it will look like in the future.
6. Summary
The AI revolution brought about by ChatGPT will profoundly change the direction of society, and now it is a chaotic chaotic situation, and everyone is trying to do something. But models are not something that ordinary people can do, so there are more products based on scene applications, and it also provides entrepreneurs and developers with many ideas for AI applications.
In addition, localization is another important topic, domestic manufacturers are obviously lagging behind in progress, but the application layer has taken the lead and has a lot of useful products, which will also be the most competitive and the most opportunities in the future. Ride this wave of AI and be a flying pig.
Finally, welcome to PandorAI, we are committed to helping AI entrepreneurs gain more insights.
This article was originally published by @PandorAI on Everyone is a Product Manager, and reproduction without permission is prohibited.
The title image is from Unsplash and is based on the CC0 protocol.
The views of this article only represent the author himself, everyone is a product manager, the platform only provides information storage space services.