OpenAI announced GPT-4 on Tuesday. This is the latest version of the world-famous chatbot that has captured the internet’s imagination since its launch last November.
MetaNews took to social media to reveal what users have been doing with the upgraded technology and find out what the bot’s biggest wins and losses have been so far.
We present a large-scale multimodal model, GPT-4, showing best-ever results for function and consistency. https://t.co/TwLFssyALF pic.twitter.com/lYWwPjZbSg
— OpenAI (@OpenAI) March 14, 2023
Since the launch of GPT-4 Users are keen to share their wins with the chatbot, and the wins are piling up.
One of the big headlines since the inception of GPT-4 has been the amazing ability of bots to pass standardized exams with little difficulty. He is one of the bar exams that bots pass with flying colors (90%) that lawyer candidates must take to practice law. Other exams included the LSAT Law Exam (88%) and the GRE Quantitative math (80%).
Here are some of the big achievements of GPT-4:
Doodle to website
GPT-4 transformed a hand-drawn sketch into a functional website in one demonstration of its power. This website is indeed very basic, but it’s a solid proof of concept.
You’ve just seen GPT-4 turn a hand-drawn sketch into a functional website.
This is insane. pic.twitter.com/P5nSjrk7Wn
— Rowan Chan (@rowancheung) March 14, 2023
£5,000 and 2 weeks savings
one smart user relayed How we leveraged GPT-4 to code five microservices for our new product. According to the user, the “very good” developer put in a £5,000 estimate and said it would take him two weeks to complete the job. With GPT-4, the user was able to complete the job in just 3 hours.
Identify security holes in smart contracts
another application of GPT-4 Identifies security holes in Ethereum smart contracts. Misuse of this can lead to theft and the loss of large amounts of money.
Coinbase Director Conor Grogan demonstrated its capabilities via his Twitter account. Tuesday.
“We dumped live Ethereum contracts to GPT-4,” says Grogan. “It quickly highlighted a number of security vulnerabilities and pointed to surface areas where the contract could be abused. We then examined the specific ways in which the contract could be abused.”
We believe AI will ultimately help make smart contracts safer and easier to build. These are two of the biggest obstacles to mass adoption.
— Connor (@jconorgrogan) March 14, 2023
One of ChatGPT’s biggest losses came directly from its social media. The bot predicts 20 jobs that could be replaced in the near future. Their roles range from data entry clerk to recruiter to copywriter.
that job #GPT-4 is superseded and described by GPT-4. pic.twitter.com/aMrwQHnfwH
— ChatGPT (@ChatGPT_0penAI) March 16, 2023
Not so fast, GPT-4.
GPT-4’s capabilities may be impressive, but it’s still a long way from bots to replace skilled humans.Case in point: CNET. When a technical publication recently replaced a human writing staff with its own Copywriting AI, the articles it emitted were disastrous. Indeed, that bot was not his ChatGPT. chatbot To do human work almost unnoticed.
As for the idea that GPT-4 could replace “data entry people” or “recruiters”, this raises credibility to absolute breaking point. There is no GPT-4. No one is for this.
Here are other examples of GPT-4 failures reported by social media users:
GPT-4 is tired of your awful questions
One of the expected benefits of using bots to write code is that unlike hired software engineers, bots never get tired, slow, or boring. At least, that’s what I hope.
user report When we asked GPT-4 for a “long code segment,” the AI appeared “bored” and simply interrupted the task. A user who observed this behavior brightly remarked, “This object is becoming more human by the day…”.
MetaNews suggests giving GPT-4 more interesting projects and paying more.
victory over failure
I tried the following logic puzzle in GPT-4 without the additional prompt eng. GPT-3.5 failed spectacularly at this puzzle with infinite hallucinations, but GPT-4 failed less dramatically.
We still have a long way to go to achieve strong reasoning abilities, but it is progress.
— Sitar Shah (@sytelus) March 14, 2023
Yes, “It’s not so noticeable and it fails” might be the faintest of the faint praises, but it’s progress nonetheless. Perhaps this should be called “Ruined Victory” .
Small GPT-4 discovery: ChatGPT-4 can sort N=20 integers, often fails for N=21, and almost always fails for N=22. Can someone please tell me what this means.
— Adam (@traditionalboi) March 16, 2023
This means I have to go back to working with integers in my head.
Losing when you thought you won
This was to quickly identify an important use case for GPT-4, and not everyone had time to stop and really think if their win was really a win.
The phenomenon is the result of one over-excited person explaining how Visual-ChatGPT can be used to scan a picture of a refrigerator filled with fruit, cheese, meat, eggs, and other key ingredients into a chatbot. typified by users of The user then ordered from the ingredients he identified to have five recipes delivered all in just 60 seconds.
The user then confidently shared the GPT-4 output with five “pretty decent food recipes” he described. These recipes were fruit salads, cheese omelets, ham and cheese sandwiches, fruit smoothies, and cheese and fruit platters.
As the barbaric Internet quickly pointed out, most of these suggestions are hardly recipes of any kind, let alone decent recipes. or other containers.
To be fair to GPT-4, Chef wasn’t one of the 20 jobs he predicted GPT-4 could replace.
Still, if all this food talk whet your appetite, feel free to try GPT-4’s “pretty decent” recipe for a “ham and cheese sandwich.”