The Limited Times

Now you can see non-English news...

"It works very well, but it's not magic": this is ChatGPT, the new artificial intelligence that exceeds limits

2022-12-07T13:39:56.124Z

The model is the latest evolution of text generation systems. Its consistency surprises hundreds of thousands of users with credible conversations ranging from criminal law to linguistics



The ChatGPT artificial intelligence model has more than a million users in just five days after its launch.

The system answers any question not only in English, but also in correct Spanish.

That coherence in the dialogue makes it unheard of.

OpenAI, its creator and one of the leading companies in artificial intelligence, has also opened its use to any user, so the networks have been filled with screenshots of resolved programming requests, answers to deep questions, traps, tricks, but also errors. obvious.

"It works very well, but it's not magic," says Álvaro Barbero Jiménez, director of artificial intelligence at the Knowledge Engineering Institute (IIC).

"It's a tremendous innovation, I was also freaked out," explains Barbero Jiménez.

“I asked him to solve some programming exercises that I give to those who join my team and he did it perfectly.

But then I asked him about a programming project with a fucking bug, and he told me a made-up solution.

He told me to use a tool that didn't exist”, he adds.

Its mode of use is extremely simple.

You just have to go to the page chat.openai.com, open an account with email and password and start chatting with the robot.

#ChatGPT Input: Write a short article for a newspaper about the latest news about the vacancy of Peruvian President Pedro Castillo.

You must include three quotes from experts on the topic.

Also some related statistics.

Includes a headline.



Result: pic.twitter.com/0LJtKDfBVw

— Ernesto Cabral (@ErnestoCabralM) December 4, 2022

This ChatGPT comes in an incredible year for the progress of different artificial intelligence models.

In summer it was the time for illustrations created from text requests.

Before that, the generation of text from an initial sentence had arrived with the GPT-3 model.

Now the latest version of the GPT software surprises again with a human finesse.

That is when the debate between two groups begins: the enthusiastic ones who elevate the achievement to the prelude to a general artificial intelligence, which reasons and is conscious, almost human, and those who put it in context and cool the excitement.

No one disputes, however, that it is a technological feat.

Carlos Fenollosa, chief executive of Optimus Price, is one of the experts who see it with the most optimism.

“He is capable of

meta

-reasoning .

That is, he understands why he is giving you that answer and not another.

It's the first time I've seen it in an AI,” he says.

The model itself is somewhat more prudent with its explanation.

This is what ChatGPT in Spanish answers to the question “who are you?”: “As a language model trained by OpenAI, I can process language and speak in a similar way to a human being.

This means that I can understand questions and provide coherent answers using human language.

Unlike a human, however, I do not have self-awareness or the ability to think and reason autonomously.

I am simply a computer program designed to mimic human language."

It seems like a response given on purpose by a machine to calm down frightened humans with its ability.

Julio Gonzalo, vice-rector for research at UNED, was also surprised to start using the chat, but he is more prudent and admits that he has a long way to go: "He appears to know about any subject, and in a superficial conversation he can convince you, but if you dig deeper you realize that he is speaking from hearsay or, put more crudely, he does not know what he is talking about.

He has intuitively learned to hold conversations on almost any topic, and appears to have real knowledge and be able to combine it with rational thought.

But it is appearance: he has neither been taught to reason nor does he have explicit knowledge, ”he explains.

But both the most cautious and the most optimistic agree that it is a matter of time.

Few expected to be here in 2022. The difference between the two sectors is holding the hype or pushing it, but few doubt that this is a breakthrough with potential that is difficult to foresee.

“You have to polish many details and contexts where you give a bad response.

But it's not serious either.

ChatGPT can make mistakes just like a person, it doesn't have to be perfect either”, says Fenollosa.

I'm sorry, I simply cannot be cynical about a technology that can accomplish this.

pic.twitter.com/yjlY72eZ0m

— Thomas H. Ptacek (@tqbf) December 2, 2022

The enthusiasm was hard to stop when a user asked him, in English, to tell how to get a peanut butter sandwich out of a video in a biblical tone or when, in Spanish, he had a debate on criminal law.

The Complutense University professor Manuel Maroto was surprised by the results: “His level, taking into account the length of the answers, is good.

The ones I posted on Twitter are very complex questions that cannot be answered in a paragraph: but as a paragraph answer, it is very reasonable.

A separate thing is that he says relatively

incorrect things

: The fact that the fraud is in the guilt is an old question in criminal law, which decades ago was resolved in the opposite direction to what he says.

Hence what is funny in the eyes of a lawyer, that the AI ​​has positioned itself in a position that has been surpassed for a long time, and then denies having done so ”, he explains.

But just for a robot created at a Silicon Valley startup to be able to converse about criminal law in Spanish is devastating.

Twitter penalists: I've been chatting with ChatGPT, and I bring good and bad news.

The good news is that the machine is quite good at dogmatics, so everything indicates that soon we will be able to dedicate ourselves to more beautiful things in life.

The bad news is that AI IS CAUSALISTIC: pic.twitter.com/Z0UWux3HTx

– Manuel Maroto (@manumaroto) December 5, 2022

Maroto did not want, due to objections, to compare the answers with those of a hypothetical student or even with those of a teacher.

Because the law is not about accumulating knowledge, “but about a way of thinking and arguing”, but otherwise, she says, ChatGPT “is very impressive”.

The same impression was made by Elena Álvarez Mellado, a linguist at UNED.

“The fluency and phrasing and coherence and apparent ability to reproduce a linguistic ability that we as humans would only expect to find in another human is impressive,” she says.

She then gave him a couple of challenges.

First on ambiguity: “'On a list like coconut oil is made from coconut, olive oil is made from olive, what is baby oil made from?'

We know that it is not made of babies.

He responded well and even gave the ingredients, ”she explains.

Then Álvarez Mellado asked her to form sentences of a certain type: reflex passive with such elements.

There it failed.

We do not know why.

Perhaps it is that in Spanish there are not enough examples.

“These are exercises that some teachers put in high school and that a student in principle is capable of doing, but this system does not know.

I do not know why.

It's kind of niche but we know that this machine converts human instructions to python [programming language] well.

It's interesting that you can write python from statements but not a sentence,” he says.

“If ChatGPT is now used as a substitute for the university, it would be catastrophic, because chatGPT lies with the same confidence with which it says true things”, says Gonzalo, who adds a good comparison of the profile that the robot can have today: “A specialized communicator in crisis he told me that when there is a crisis to manage in networks (for example, a fire in an industrial plant), the most dangerous profile on Twitter is that of the

enteradillo

.

The one who seems to know, but really doesn't.

Because people attribute knowledge to it, and then the brown mess.

Well, chatGPT is the hypermegaentertainer”, he explains.

ChatGPT produces an illusion of irresistible rational thought.

But it is still a language model that *does not reason* nor does it have *reliable knowledge* about the world.

Here is an example of how he lets himself be carried away by the interlocutor and begins to ram.

pic.twitter.com/jITyRA3sz7

– Julio Gonzalo (@ JulioGonzalo1) December 3, 2022

Gonzalo posted an example on 19th century poets on Twitter.

The chat responses look like a brother-in-law but polite and courteous.

“I asked him to tell me Spanish poets of the 20th century, and he gave me a very decent list.

So I made up that I had made a mistake in one, and that it was actually from the 16th century (total false).

Well, he goes and agrees with me, apologizes and invents his dates of birth and death in the 16th century.

Imagine what a danger to replace teachers! ”, He explains.

The trades in which artificial intelligence can help or even end up replacing have expanded a lot.

The danger of machines is no longer limited to industrial plants.

This chat knows how to answer questions that inexperienced programmers, teachers or writers may answer less clearly.

There are class tasks such as essays or essays that are destined to die with this technology.

“The work of a programmer, with the tests we have done, many of us are clear that he will know how to do it very well,” says Victoriano Izquierdo, co-founder of Graphext, a data analysis company.

“It will make it possible to get down to the operating system level and do really complex tasks.

Perhaps now, paradoxically, it is professions that require more physical skill, such as being a dentist or a hairdresser, that seem more difficult to replace until hardware advances further and we are afraid of a mechanical Edward Scissorhands ”, he adds.

Fenollosa goes further with his list of jobs allegedly in danger.

“All copywriters

are charged

, including journalists (if they don't do reporting).

For example, rn press releases or current affairs summaries.

Also questions and answers, such as Stackoverflow [page of doubts about code, where they have had to block the use of the chat due to excessive errors] or Quora [generic questions], including the work of solving doubts from many professors.

And simple programming, design and illustration.

You can say to ChatGPT 'design me a website for a construction materials manufacturing company and include several stock photos of people visiting a construction site' and it will give you a starting point that would have taken a human a couple of days to prototype” , Explain.

The robot will even be able to make good jokes, says Fenollosa.

“I don't mean that I'm able to write funny stories by combining pieces from other stories.

ChatGPT understands irony and elements of humor, is capable of coming up with new jokes, ”she says.

“Of course it will impact professions.

But on the scale between nothing happens and it is the end of humanity, we are somewhere in the middle”, says Barbero Jiménez.

Although the fears are growing, it is reasonable not to be dazzled, as it happens in this exchange about a rental contract in Catalan.

It looks like it's impressive but it's actually not:

But he is not right.

They are not unfair clauses.

pic.twitter.com/THEanRpMMl

– Manuel Ríos San Martín (@mriossanmartin) December 6, 2022

This newspaper asked him for a couple of poems about journalists.

They were not very refined: “But in the midst of so many lies / the journalist feels tired and desperate / because sometimes it seems that the truth / has been lost in this gloomy place.

/ Yet he presses on / with determination and courage / to continue to search for the truth / and show the world what really happens.”

A small hope is that these tools are capable of solving well what until now was mediocre production.

All of this will be taken today as “automatic production”.

Adding value will be strictly human, like crafts.

You can follow

EL PAÍS TECNOLOGÍA

on

Facebook

and

Twitter

or sign up here to receive our

weekly newsletter

.

Source: elparis

All tech articles on 2022-12-07

You may like

News/Politics 2023-01-21T11:01:45.426Z
News/Politics 2023-01-27T03:57:27.094Z

Trends 24h

Latest

© Communities 2019 - Privacy