ChatGPT is crashing again?
Do you know how hot you are right now? Think of a slogan for yourself
Thanks for the compliment! My ad is: "Let ChatGPT take you through the infinite possibilities of knowledge!" ”
Even if you are polite, you are still so talented, no wonder the workers are excited and panicked
Have you experienced ChatGPT, the fastest growing consumer (paid) app in history? After the launch, the monthly active users directly exceeded 100 million, and the official website also announced that it was "fully loaded".
Not only is the writing like a literary hero possessed, the code is written accurately and clearly, and the efficiency is also beyond the reach of migrant workers, while admiring the ease of use, we can't help but sigh whether they will one day be replaced by AI, the mid-life crisis has not yet arrived, and the AI crisis is ushered in first.
ChatGPT subverts the original professionally generated content (PGC) and user-generated content (UGC) methods, and the era of AIGC artificial intelligence-generated content represented by ChatGPT has begun.
According to Semafor, Microsoft is in talks to invest $10 billion in OpenAI at a $29 billion valuation, and that money points to a new paradigm for AI models, the generative AI model.
So what is a generative AI model?
AI models can be broadly divided into two categories: decision-making AI and generative AI, and the difference between the two is that they are good at solving problems.
Decision-making AI models analyze, judge and predict based on existing data, such as distinguishing images into cats and dogs. It is often used in the well-known short video intelligent recommendation, precise advertising push, automatic driving and face recognition.
Generative AI excels at inductive deductive creation, stitched creation based on history, and imitative innovation, such as generating realistic images of cats or dogs.
According to Gartner's 2022 Key Strategic Technology Trends, generative AI is listed at the top of the 12 key strategies. Generative AI has begun to be applied in content creation, scientific research, medical care, and many industrial fields, such as copywriting, text-to-image, video intelligent dubbing, intelligent poster generation, intelligent medical diagnosis, etc. It can be said that creation is the core of generative AI, which is essentially a substantial increase and creation of productivity in various fields.
The improvement of AI capabilities is not achieved overnight, and most of them have experienced the cycle of "model breakthrough - substantial improvement - large-scale production - encounter obstacles - model breakthrough - substantial improvement". Sequoia Capital predicts that by 2023, generative AI will mature in text and code, but mature applications in images, video, 3D, and gaming will still take some time.
As a new influencer based on the GPT3.5 model, ChatGPT has gone through 4 iterations. ChatGPT can be compared to an orangutan cub, and programmers are like breeders, constantly training it.
The training process of ChatGPT can be divided into three stages:
In the first stage, keepers pre-train orangutans with unsupervised massive amounts of data. A random group of questions from tourists is selected, and keepers will help orangutans give high-quality answers. Orangutans absorb a lot of knowledge but don't necessarily understand the meaning.
In the second stage, the breeder trains the reward model. Orangutans are given one question and four answers, and keepers rate and sort the answers according to visitor preferences. Through training, the reward model is allowed to imitate the visitor to score the results.
In the third stage, reinforcement learning and self-training are added. Let the orangutan and the tourists chat non-stop, reward the model like the teacher, and score the orangutan's answers, and the orangutans learn to give answers that the teacher prefers in order to get a higher score. Thus achieving self-training of orangutans. As the interaction between tourists and orangutans increases, orangutans continue to evolve.
In the future, it will enter the fourth stage of training, where orangutans will have more assessment goals and summaries, and learn to use tools to help humans improve productivity.
However, Yann LeCun, Meta's chief AI scientist, believes that ChatGPT's underlying technology lacks innovation and revolution, and that at least six startups have similar technology. However, this view has been controversial, and ChatGPT as a language model has high application value and research significance in the field of natural language processing, especially in language generation.
In the future, generative AI may have an impact on the creative industry, but its ability to create and think is limited, and it can only be used as an auxiliary tool for creativity, and cannot completely replace human creativity.
So how is AI applied in e-commerce technology?
ChatGPT is not a completely new technology, Alimama has already applied this technology.
The Alimama Creative Center is like an intelligent creative director, and many of the time-consuming and labor-intensive creative production tasks for us are simple for it.
For example, in the past, merchants usually needed to hire professional writers and designers to complete the writing and design of the copywriting on the drawing, which was costly and relatively inefficient. Moreover, it is necessary to rely on the designer's template to fill in the corresponding copywriting, which is not flexible enough and the adaptation cost is high.
To this end, Alimama proposed a context-based model of copywriting on the multimodal diagram, comprehensively considering the image information, product text information, text box position and other information, and adaptively generated appropriate copywriting. It can help merchants save costs, improve efficiency and productivity.
In the actual use stage, Alimama used a large dataset of nearly 1 billion product pictures for training, and the obtained model had good effects in copywriting smoothness and rationality, and after rendering the model, it has reached the level of delivery.
Is there a solution for AI for live streaming services?
Of course, there are, for example, many merchants have a small number of viewers in the live broadcast room and poor conversion. Moreover, after the live broadcast, the starting price of producing a short video with a single clip is more expensive, time-consuming and laborious.
Now, through one of the three core technologies of Alima, the live room intelligent technology suite ACE (Alimama Content Express), merchants can realize the delivery while broadcasting. It is like a combustion booster for live broadcast detonation, while live broadcasting, intelligently editing 10-30 seconds of wonderful highlights, while doing thousands of people and thousands of faces in various resource locations of Taoji, helping merchants carry out multimedia secondary transmission, and leveraging natural traffic to feed back the live broadcast room. Moreover, short videos are covered around the clock in all resource bits, and the traffic obtained by merchants' products will continue to continue, and live broadcasts cannot always accompany users, but short videos can. It allows every highlight moment of the merchant's live broadcast not to be wasted, so that the vitality of the live content has a better continuation.
In the future, Alimama will continue to strengthen the research and development and promotion of scene intelligent applications, and apply artificial intelligence technology to various scenarios, such as Alimama is exploring application scenarios where the model turns three times and then takes a step forward. It also further explores the model of human-machine collaboration to improve the intelligence and effect of advertising, and provides a full range of AI intelligent services for Taobao merchants, so that older merchant users can also easily use e-commerce services.
Looking ahead to AI
Like autonomous driving, ChatGPT also faces regulatory and ethical challenges, requiring privacy protection measures, elimination of algorithmic bias, and the establishment of ethical decision-making frameworks and regulatory mechanisms to ensure that ChatGPT does not have a negative impact on human society.
Although ChatGPT has demonstrated amazing capabilities, AI technology may have an impact on existing creative work for humans, requiring a re-examination of one's own value and role in the work and actively learning new skills and knowledge to adapt to the needs of future work, and by partnering with AI to leverage the unique strengths of humans to create a smarter and more meaningful way of working and living.
Will artificial intelligence replace humans?
In the words of ChatGPT, there is still a lot of controversy and discussion. It may surpass in some areas, but it is still inferior to humans in terms of creativity and emotion, and human intervention is still needed. The future requires a balance between technological development and human values to achieve human-robot collaboration and symbiosis.
The era of AI has come, let us embrace the AI trend, face changes with a positive attitude, and strive to promote the development of AI technology so that it can better serve the development and progress of mankind.