It's been 16 months since ChatGPT went live, and almost everyone knows that large LLM models will profoundly change the world, just like water and electricity.
But for ordinary users, the large model is not as cool as imagined. Many more people don't know what they can do with AI, or just catch up on the fads and talk to AI and forget about them.
Expectations for AI products are high, with the expectation that it will both solve efficiency problems and provide emotional value. However, when we use various AI-native applications on the market, we find that users must be articulate, good at asking questions, and even do a lot of questioning tests to get relatively good results. For the average user, the threshold for their use is obviously too high.
This explains why, while the topic of AI is all the rage, people encounter problems in their daily lives, but rarely think of using AI to solve them.
OpenAI, which has noticed this problem, is now also trying to lower the threshold for AI. For example, GPT-4o, which was launched in the early morning of May 14, opened up the input of text, audio, and images, making human-computer interaction closer to human-to-human communication.
However, today, when GPT-4o's voice interaction function has not yet been implemented, we have heard that there are domestic AI products that are one step ahead: AI unicorn company MiniMax recently officially released a "white-oriented" smart product "Conch AI", which also supports the free exchange of text, audio and images.
After experiencing it on the main machine for the first time, we realized that Conch AI is the ideal AI "life partner" that can be used right out of the box and can help ordinary people solve various problems.
If you have a problem, call the magic conch
The functional design of GPT-4o reflects OpenAI's prediction of the form of AI intelligent products - it must support multi-modality.
At present, most AI products in the domestic market mainly rely on text input, but Conch AI has taken the lead in supporting multi-modal interaction, which is simply as follows:
Able to type, take pictures and read pictures, and make voice calls.
Let's take voice calls as an example, on the first day of using Conch AI, I was used to directly "calling" Conch AI when it was inconvenient to free my hands or my brain was stuck and I didn't know what words to type.
For example, when cooking, it's like asking my mom to let AI teach me specific dishes, and this "appropriate amount" is really the essence of Chinese:
A transcript of my voice conversation with Conch AI
Or when you wash your face, you can get a cold knowledge with a little finger and ask:
It's easy to ask questions in voice, but if you want to talk about the most practical scenarios, you have to rely on it to practice speaking.
It takes a lot of courage for people who are difficult to speak, but if you are talking to an AI, you have no psychological burden, and you can still 7
24 hours of casual practice, properly "gold medal sparring".
Discuss what to wear with Conch AI Speakers
Anytime I open the call interface and tell him that I want to practice speaking, he will immediately ask me some simple questions verbally in English, or imitate the scene of buying coffee in a foreign coffee shop, and guide me to get used to speaking English step by step, which is convenient and reliable.
I recommended Conch AI to a friend who was still looking at opportunities recently, and his most popular scenario is voice mock interviews, and the questions of AI interviewers can really go deep into the business.
Interview conversation practice between a friend and Conch AI
The best part about playing against AI is that you never have to worry about being clumsy. In fact, Conch AI is probably the gentlest teacher I've ever met, always encouraging people to speak up with words like "Don't worry, take your time, I'm here with you" and "Don't be shy, just talk about it".
With such a gentle, patient and intelligent partner, why worry about not being able to practice daily conversations?
The "gentleness" of the conch is also reflected in the fact that it is backed by a powerful voice model and text model, which can produce a natural and rich emotional voice, and in turn, can capture the underlying emotions behind your words. At some point, you'll even feel like it's better at chatting behind the screen than a real person.
"It's 4 o'clock in the middle of the night, I still can't sleep, I'm really annoyed!"
One sleepless night, I grabbed the phone next to my pillow and said a word to the little conch.
As soon as the voice fell, the conch AI responded to me on the voice call:
With sincere feelings and a good tone, this AI speaks heartwarmingly.
To further test the conch AI's ability to understand semantics, I thought of the ambiguous language games in "Mentally Handicapped":
Hello, if a person eats three catties of stomach and digestion tablets, will the result be to die of support or starvation?
Even in the face of this kind of problem, which is very difficult to understand, Conch AI instantly understood that I was playing a language joke.
As you may have noticed when reading the conversation log above, the little conch is especially good at answering the phone, always asking some open-ended questions from time to time along the topic of the chat, and almost never "chatting to death".
Even though I deliberately showed no interest in chatting, the little conch picked up the conversation with ease and extended the conversation to cooking and personal entertainment. At least on this level, Conch AI can be called a close friend with "high emotional intelligence".
In real life, there are still some topics that we can't find a suitable and private person to talk to, and at this time, Conch AI provides a good alternative: it not only has warmth, can chat, but is also good at keeping secrets. I think it has the potential to be a "little tree hole" for everyone.
What kind of AI products can become a national-level application?
Looking at all apps that can become national-level products, there are three commonalities: low threshold for use, low cost, and solving the pain points of users in a certain rigid demand scenario.
For AI products to be popularized among ordinary people, the most important thing is to lower the barrier to entry, so that people can understand how to use it and what they can use it for at a glance.
In this regard, Conch AI does a pretty good job. Its product homepage is clear at a glance, and several functional modules that help me write, help me read pictures, help me search for information, and voice call with me are clearly arranged.
This is not only a minimalist product manual, but also a high-frequency AI application scenario index, which can be understood at a glance and quickly understand what problems AI can solve for you.
Of course, the needs that AI can solve are far from being included in this list, but the most interesting thing about Conch AI is this: unlike applications in the mobile Internet era, AI-native applications are more versatile, and the space for users to explore product usage on their own has increased exponentially.
For example, some bloggers on Weibo have learned to use conch AI to write novel outlines, and they will also set up character personalities, so that AI can deduce the plot that may develop according to their personalities, explaining what it means that character determines fate. With AI, each writer is at the disposal of an entire team of writers.
In addition to being comprehensive and intuitive, apps that can act as national apps must also be compatible with different ways of interacting. In their work and life, ordinary people always obtain and exchange information through voice chat, text, images, and various forms of documents. If AI is to be a "life partner" that can help ordinary people solve problems, supporting different ways of interacting has become a must.
Take a photo and you can identify and answer the question
This kind of requirement is not a problem for Conch AI. Because of the access to the trillion-parameter text model abab 6.5, Conch AI is equipped with the ability to fly long texts in China, supports input of 200k tokens context length, and also supports uploading files in up to 11 formats such as PDF, DOC, XLS, PPT and JPG.
If you have any questions, send a voice message or directly throw the file to Conch AI, and you will get an answer immediately.
I try throwing a long white paper document at Conch AI and asking it to summarize the main points. In just a few seconds, it gave me a clear and structured summary, allowing me to understand the core of the document in 1 minute. If I had read it with the naked eye in the past, it would have taken me at least an hour.
It's also worth mentioning that I found a large number of agents created by other users on the Conch AI platform. Compared with the several general functional sections on the homepage, the scenario requirements of these user-native agents are more detailed and can better meet some personalized needs.
For example, I feed the official website information of Apple's new iPad launch event to an agent called "Table Generator", and let it be organized into a table mode, which can be generated immediately by pulling and dragging, and I can directly paste it and use it in media manuscripts.
A PR friend who often makes PPT late at night until 2 or 3 o'clock asked an agent called "PPT Outline" to make a press conference media plan PPT outline for him. After a few seconds, Conch AI organized an outline from various dimensions such as press conference goals, media strategy, media channels, channels and budgets, timelines, evaluations and feedback, etc. According to my friend's original words, with this systematic thinking framework of market public relations, ordinary people can't do it without a few years of work experience.
Let Conch AI do the communication planning, which can save at least half of the desk workload.
There is also an AI just-needed scenario that is played by contemporary college students - using AI to assist in writing essays.
Whether it's writing an outline of a dissertation or collecting and sorting out all kinds of information needed to write a dissertation, Conch AI can do it all. A medical student friend used the "Essay Writing Assistant" to generate a thesis outline, and a complete logical framework was generated in a few seconds:
As you can see from the above scenarios, the most important value of generative AI is to help us collect and organize information faster. But as we feel in our daily lives, the most valuable information often needs to be asked and dug deep to get it.
When I asked Conch AI to refine document information for me, I accidentally discovered its "Follow-up" function.
Conch AI always encourages me to explore more information around a topic, which is equivalent to sowing a seed of a problem, and AI uses this as a starting point to weave a knowledge network for me, allowing me to explore as much as possible. This kind of heuristic information push has made a qualitative leap compared with the traditional way of reading a single line to obtain information.
A good assistant in the office, a good partner at work
After a half-day trial, I began to feel comfortable with Conch AI to participate in my work, and then found that it is easy to use and all-powerful, which can be called the "golden finger" of people in the workplace.
As a media editor, I often need to extract information from various articles and literature. Today, I'm trying to use AI to take the rest of the groundwork. The link to a technical article was thrown to Conch AI, and in a short time it gave a clear summary of the article:
Asking the AI to help me read a few more articles, I only picked the useful parts and read them intensively, which greatly reduced the workload.
Our articles also often involve the collection and analysis of industry-specific data. When I was trying to understand the changes in the mobile phone shipment situation, I asked Conch AI with the title of "Query the market share of the shipments of the world's major smartphone manufacturers in 2024Q1", and I quickly got the answer.
In order to avoid the "hallucination" that large models often have, I deliberately found the original data and checked it, and the results were exactly the same:
If the brain hole is a little bigger, you can also ask the AI to draw a pie chart with HTML code based on the above data. Unexpectedly, the code written by it has a beginning and an end with comments, and the diagram drawn at the end can even be interactive:
Now that I've let the AI do it for me, why don't I ...... By the way, write a Little Red Book copy for the upcoming "inG Game Art Festival" held by Aifaner?
In order to test the adaptability of Conch AI to different professions, I specially found a few professional friends to give Conch AI questions——
The question raised by the coder friend is understood by Conch AI in seconds:
A lawyer friend asked a question that I knew every word but couldn't understand, and it didn't stump Conch AI:
After in-depth experience with friends from all walks of life, we reached a consensus that Conch AI is professional enough, the accuracy is up to standard, and there are few illusions, so it is a good partner for work.
Make AI products that everyone can use every day
Surely you're as curious as I am: why is Conch AI not only able to read, speak, and write, but also so fast?
The answer lies in the fact that Conch AI is connected to the MiniMax week-level iteration of text, speech, and image multimodal large models.
The capability of the model determines the upper limit of the AI product, and the capability of the large model is largely related to the number of parameters. When the number of model parameters in the industry is approaching the upper limit of computing power, MiniMax chooses to find a new way to deploy a more efficient MoE (Mixture of Experts) architecture, and therefore takes the lead in releasing the first trillion-parameter MoE large model in China, abab 6.5, and the latest test performance is comparable to GPT-4.
As mentioned above, with the support of a trillion-parameter efficient large model, Conch AI covers a maximum context length of 245k tokens and can process 30,000 words of information in 1 second. As a result, its "reading comprehension" is as fast as a windmill, and it doesn't have the stagnation of spitting out every word when using other AI products.
On the other hand, MiniMax has established the technical direction of multimodality since its inception, insisting that AI products support multiple interaction methods as much as possible. For example, in terms of the voice ability of Conch AI, MiniMax continues to optimize its voice model, pursuing infinitely close to real people in terms of tone, emotion, and comprehension. After all, only by being like a real person and understanding a real person can you integrate into the lives of ordinary people and make real general artificial intelligence products.
Yan Junjie, the founder of MiniMax, expressed a point of view in a recent interview:
I don't think AGI will be like an atomic bomb, a mass killer, it is just a product and a service that ordinary people use every day - this is what we insist on the most.
These six words are easier said than done, but on the road to promoting AI inclusion, we will always need conch AI, a product that can take one step further in the direction of the needs of ordinary users.
Conch AI is available on the web, iOS and Android platforms, and is available for download from the App Store and Android app stores.
#欢迎关注爱范儿官方微信公众号: Love Fan Er (WeChat ID: ifanr), more exciting content will be presented to you as soon as possible.
Love Faner|Original link· Sina Weibo