The researchers say that GEN AI resolution problems do not disappear anytime soon

Chatbots of artificial intelligence It is known that there are many mistakes. Let’s hope that you will not follow the Bojogle Amnesty International’s proposal Add glue to your pizza recipe Or eat a rock or two a day for your health.

These errors are known as the name HallucinogenicBasically, the things that the model compensates. Will this technology improve? Even researchers studying Amnesty International It is not optimistic that it will happen soon.

This is one of the results of a painting from Twenty artificial intelligence experts This month was issued by the Association for AI. The group also wiped more than 400 members of the association.

In contrast to the noise that you might see from the developers of years (or months away, depending on what you ask) away from improving artificial intelligence, this committee is from academics and industry experts seems more guarded about how quickly these tools are offered. This only includes obtaining the facts correctly and avoiding strange mistakes. The reliability of artificial intelligence tools must increase significantly if the developers produce a model Meet or overcome human intelligenceKnown as artificial general intelligence. Researchers seem to believe that improvements in this range are unlikely to happen soon.

“We tend to be a little careful and do not believe anything until it actually works,” Vincent KonitezerProfessor of Computer Science at Carnegie Mellon University and a member of the committee told me.

Artificial intelligence has evolved rapidly in recent years

The head of AAAI, the president of AAAI, Francesca Rossi, wrote in his presentation, which is to support research in artificial intelligence that produces technology that helps people. Confidence and reliability issues are dangerous, not only in providing accurate information, but in avoiding bias and ensuring that there are no unintended severe consequences. She wrote: “We all need to work together to progress in artificial intelligence in a responsible way, to ensure that technological progress supports the progress of humanity and is in line with human values.”

Artificial intelligence accelerated, especially since Openai Chatgpt In 2022, he was great, Konitez said. “In some respects, that was amazing, and many of these techniques work much better than most of us believed to do it,” he said.

There are some areas of artificial intelligence research where “noise has an advantage”, ” John SeikstonAssistant Professor of Computer Science at Cornell University told me. This is especially true in mathematics or science, where users can verify the results of the form.

“This technology is amazing,” said Seikston. “I was working in this field for more than a decade, and it shocked me how good it is and how good it becomes good.”

Despite these improvements, there are still important issues worth searching and looking.

Will Chatbots start obtaining their facts directly?

Despite some progress in improving the merit of confidence in information that comes from obstetric artificial intelligence models, more work must be done. accident A report from the press review in Colombia It is unlikely to refuse the chat that was found to answer the questions that they were unable to answer accurately, confident of the wrong information they provided and formed (and provided fabricated links) to sources to support these wrong assurances.

The AAAI report said that the improvement of reliability and accuracy “can be said that the largest field of artificial intelligence research today.”

The researchers noticed three main ways to increase the accuracy of artificial intelligence systems: precise installation, such as enhancing learning with human comments; The generation of retrieval, in which the system collects specific documents and withdraws its answer from these; And the series of thought, where the claim is destroyed to smaller steps that can achieve the model of artificial intelligence from hallucinations.

Will these things make Chatbot responses more accurate? It is not possible: “Realism is far from the solution,” the report said. About 60 % of those surveyed indicated that concerns about confidence or merit will be resolved soon.

In the manufacturing of obstetric intelligence, there was optimism that expanding the scope of current models would make it more accurate and reduce hallucinations.

“I think hope has always been very optimistic,” said Seikston. “Over the past two years, I have not seen any evidence that very realistic and realistic language models are just around the corner.”

Despite the ability of large language models such as Antarbur Claude or Llama MetaContezer said that users can mistakenly assume that they are more accurate because they provide confidential answers.

“If we see a person responding with confidence or words that seem confident, we consider that the person really knows what they are talking about,” he said. “Artificial Intelligence System, it may claim to be very confident about something completely nonsense.”

Lessons for artificial intelligence users

Awareness of the bonds of the artificial intelligence is vital to use it properly. SHIKETSTUN advice for users of models such as ChatGPT and Gemini from Google Simple: “You have to check the results.”

He said that the great public language models do a weak work in constantly recovering real information. If you ask for something, you must follow up by searching for the answer in the search engine (and not relying on the AI ​​summary of the search results). By the time you are doing, you may be better at the first place.

Sexston said that the way to use artificial intelligence models is to automate the tasks that he can do anyway and that he can verify accuracy, such as coordinating information schedules or writing code. “The broader principle is that I find that these models are very useful to automate the work that you already know how to do it,” he said.

Read more: 5 ways to stay smart when using Gen AI, as explained by computer science professors

Is artificial general intelligence just around the corner?

One of the priorities of the artificial intelligence development industry is a clear race to create what is often called artificial general intelligence, or AGI. This is a generally capable model at the human level of thought or better.

The report survey found strong opinions about the AGI race. It is worth noting that more than three quarters (76 %) of the respondents said that increasing artificial intelligence techniques such as large language models are unlikely to produce AGI. A large majority of researchers suspect that the current march towards AGI will work.

A great majority likels that systems capable of artificial general intelligence should be owned by the public if developed by private entities (82 %). This is in line with concerns about ethics and potential soldiers to create a system that can think of humans. Most researchers (70 %) said they are opposed to stopping AGI research until safety and control systems are developed. “These answers appear to indicate a preference for exploring the ongoing topic, within some guarantees,” the report said.

Smithston said the conversation about Aji was complicated. To some extent, we have already created systems that have a form of general intelligence. Big language models such as Chatgpt from Openai are able to do a variety of human activities, unlike the oldest artificial intelligence models that can only do one thing, such as Play Chess. The question is whether he can do a lot of things constantly at the human level.

“I think we are very far from this,” said Seikston.

He said that these models lack a compact concept of truth and the ability to really deal with creative tasks. “I do not see the way to make them working in a human environment using current technology,” he said. “I think there are many research developments on the way to get there.”

Conitzer said that the definition of what is exactly AGI is difficult: often, people mean something that can do most of the tasks better than humans, but some say it is just something able to do a set of tasks. “The tougher definition is something that would really make us completely extra,” he said.

While researchers are skeptical about it I come in the corner or lowerConitzer warned that artificial intelligence researchers do not necessarily expect the dramatic technological improvement that we have all seen in the past few years.

He said: “We haven’t seen the rapid change of things recently, and therefore you might wonder if we would see this will come if it continues to work faster.”