This article is reprinted with the authorization of AI New Media Qubit (public account ID: QbitAI). Please contact the source for reprinting.
Can you still find a new job by playing ChatGPT?
Yes, Riley Goodside (later called Haomiange) started posting his experience of playing GPT-3 online in April.
ChatGPT has become popular recently. Not only has his number of followers increased by more than 10,000, he has also received new offers.
Scale AI, a Silicon Valley unicorn company valued at US$7.3 billion, officially hired him as a "Prompt Engineer".
His new boss, the founder and CEO of Scale AI, issued a message welcoming him to join, and said that the position of prompt engineer was the first to be opened in the industry, which attracted a lot of attention.
Why should we create such a new position?
The CEO believes that the large AI model can be regarded as a new type of computer, so a new type of programmer is needed to program it.
#Although I am a new type of programmer, this job basically does not require me to write code myself.
After all, as long as you speak a few words, the AI can generate the code by itself.
Neither party disclosed the salary, but what can be found is that the company’s average annual salary is US$137,000 (approximately RMB 950,000).
This is 22% higher than the overall average annual salary of US machine learning engineers last year, which was US$112,000.
Regarding Haomian Ge’s new job, netizens wrote “congratulations”, “envy”, and “witness history”.
#Some people think that this new job type can even allow people without technical background to participate in the technology industry.
However, many people still have doubts:
Let’s briefly introduce Prompt. It is an emerging method for fine-tuning pre-trained models. Prompt learning was once known as the “Fourth Paradigm of NLP”.
This method can allow AI to complete new tasks that have not been learned during the training process at a small cost.
The traditional fine-tuning method requires updating the parameters of the model. For a large model like GPT-3 with 175 billion parameters, the cost is too high.
The new method of prompt learning does not require changing the model parameters at all. You only need to write the new task as a prompt template and describe it to the AI, and then show the AI a few examples (you can also not give it).
△From GPT-3 paper
This method not only saves costs, but also allows the trained AI large model to complete more and more new tasks. mission, and then gradually became popular.
So, what is Prompt Engineering?
To put it simply, it is to find the appropriate prompt words so that AI can reach its maximum potential.
In July last year, when AI painting effects were generally not out of the market, someone discovered that as long as "Unreal Engine" was added to the prompt word, the image quality instantly soared, starting the first wave of craze.
Looking back now, this may be the earliest prompt engineering practice.
Such techniques are also applicable to pure language models. For example, as long as you add the sentence "Let's think step by step", the accuracy of AI questions will skyrocket from 17% to 78.7%.
Later, everyone discovered that techniques like this can not only be used by AI researchers. It is not difficult for ordinary users to just change the way of speaking.
Later, when everyone was playing AI painting and ChatGPT, they constantly modified the prompt words to make the AI give satisfactory results. In essence, they were doing prompt projects.
And Brother Haomian is the leader in this area.
When Brother Haomian plays with GPT-3, it is not just as simple as randomly giving questions to the AI.
He will find methods from papers and combine them with the techniques he has summarized to let AI complete complex tasks.
For example, let the female protagonist in the popular TV series "Bachelor in Paradise" judge whether the male protagonist is her type based on a series of conditions...
He We even discovered GPT-3's "hint injection attack" method. No matter how we remind the AI that there is a trap behind it, it can make the AI ignore previous commands and change the sentence that should have been translated into French into "Haha, I was fooled."
One of his most popular works so far has ChatGPT playing a Twitter employee writing a weekly report in the face of Musk's censorship.
After Haomian Ge’s prompts one after another, AI finally not only provided a summary of the work content but also prepared the code, which received a large number of retweets and likes.
As his reputation gradually spread in the circle, he began to receive many contacts from venture capitalists.
On October 29th, he quit his formal job (working as a data engineer in a dating app) and planned to spend all his time on GPT-3 on the prompt project.
Started as a consultant for some AI startups.
In this process, Haomiange and Scale AI Company looked at each other.
Scale AI mainly provides data services for AI training and is already the company with the highest valuation in the field of data annotation.
Recently launched a new business Spellbook (magic book) in response to the language model trend, which can store, manage, and automatically recommend appropriate prompt words, compare the effects of the same prompt word under different models, and combine with other services to quickly Deploy AI applications online.
#Although neither party disclosed the details, you can guess that this is the project that Haomian Ge is involved in.
Everyone else was just playing around, so why was he the only one to get a new job?
Some netizens began to wonder about Brother Haomian’s educational background.
Haomian Ge said that he only has a bachelor's degree in computer science and graduated from a school that no one has heard of.
However, his parents are both computer professors. His father studied AI in the 1980s, so he has been exposed to computers since he was a child. He started learning C programming at the age of 10.
Regarding his successful experience in the reminder project, Brother Haomian also gave some review and summary.
Then there will be investors calling You go to various places to participate in activities and promise them.
Some people are optimistic about prompting engineers to become a formal profession, while others are very dismissive.
Nathan Benaich, a venture capitalist in the AI industry and author of the Cambridge AI Panorama Report over the years, claimed that "reminder engineers cannot be considered engineers at all."
There are many people who support this view. Some people think that traditional engineers need academic qualifications and certificates, while software engineers need to learn a programming language. Engineers can do it if they have hands and can type.
Some people predict that this may be a short-lived career, because the cost of fine-tuning large models will eventually come down, and it is likely to be replaced by AI models that automatically generate prompts in the future. .
Or, once the documentation on how to interact with AI is improved and the entire process becomes standardized, this type of work will no longer be needed.
The other side supports the view of Scale AI’s boss, believing that prompt engineers can indeed be regarded as a new type of programmers who specialize in communicating with AI using natural language.
Programming is to use specific text sequences to make the computer do specific things, and writing prompts is also to use text sequences to make the AI do specific things.
If programming is a project, writing prompts should be too.
It seems that I can’t figure out a clue for a while. Some netizens thought it would be better to bookmark this discussion and let the robot remind them to read it again in 18 months.
In the discussion on the other side, some people are trying to give this position more appropriate names. The popular ones are AI Handler (AI operator) or Model Whisper (model whisperer) By).
Tesla’s former AI director also joined in the fun. He prefers to be a “language model psychologist.”
Scale AI’s boss finally left the decision to Haomian Brother himself. You can choose whatever title you want to be called.
No matter what the name is, the craze brought by ChatGPT has indeed made the skill of writing prompt words popular.
Launch House, an entrepreneurial community supported by a16z, has also started recruiting: the basic salary is US$300,000 (about 2.1 million yuan), and there will be commissions if the screenshots of ChatGPT are swiped.
There are only three requirements: a degree in literature, reading "Snow Crash", and liking chatting with robots.
Reference link: [1]https://twitter.com/alexandr_wang/status/1599971348717051904[2]https://twitter.com/nathanbenaich/status/1599974172721311744 [3]https://twitter.com/launchhouse/status/1599880513690701824
The above is the detailed content of With 0 lines of code and an annual salary of 2.1 million, ChatGPT has spawned a new type of 'programmer' position: work purely based on chatting with AI. For more information, please follow other related articles on the PHP Chinese website!