GPT-4 has evolved so much that nearly ten thousand people have signed a petition to block it! The White House held an emergency meeting, and ChatGPT concept stocks plummeted

The threat of GPT-5 has become overwhelming.

After the birth of GPT-4, the disruptive impact that AI may have on human society has already frightened many industry leaders.

Recently, it seems that the whole world is attacking ChatGPT.

An open letter jointly initiated by Musk, Turing Award winner Bengio and other big names a while ago to suspend advanced AI research and development has now grown to more than 9,000 signatures.

Immediately afterwards, the US non-profit organization Center for Artificial Intelligence and Digital Policy (CAIDP) also asked the Federal Trade Commission (FTC) to investigate whether OpenAI violated consumer protection rules and prohibits it from further publishing GPT-4.

However, Italy banned ChatGPT nationwide and only gave OpenAI a 20-day adjustment period. If it fails to come up with a qualified solution, it will have to pay a fine of up to 20 million euros.

The frenzy of blocking ChatGPT is sweeping the world

What’s interesting is that since Italy took this initiative, some other countries are also eager to try and prepare to take action on ChatGPT.

Privacy regulators in France and Ireland have contacted their Italian counterparts seeking more details about the investigation.

German Federal Data Protection Commissioner Ulrich Kelber said: "In principle, it is possible to adopt similar procedures in Germany."

In addition, The European Union has also proposed a draft plan to regulate ChatGPT as a high-risk AI system.

The United States also took action

And just today, US President Biden was also in the White House with Science and technology advisers held a meeting to discuss the risks and opportunities that AI brings to humanity, society and national security.

According to Reuters, during a meeting of the President’s Council of Advisors on Science and Technology (PCAST), Biden emphasized, “In my view, technology companies have a responsibility to ensure that their products are It is safe."

When asked whether AI is dangerous, he said, "It remains to be seen, there is such a possibility."

Biden said the harm that powerful technology can cause without protective measures has been evident on social media.

“Because of the lack of safeguards, we are seeing technology impact people’s mental health, causing them to feel low self-esteem and despair about their self-image, especially among young people. 》

He once again called on Congress to pass bipartisan privacy legislation that would limit the personal data collected by technology companies, ban advertising targeted at teenagers, and improve product development Prioritize health and safety.

Affected by this news, the stock prices of AI companies fell sharply before Biden convened this meeting.

The share price of ChatGPT concept stock C3.ai continued to fall, closing down 26.34% at US$24.95, with a market value of US$2.8 billion, the largest intraday decline since the stock was listed.

Previously, since Microsoft announced its multi-billion dollar partnership with OpenAI, C3.ai, like many companies in the AI ​​field, It has become one of the hottest companies on Wall Street recently, with C3.ai's stock price rising more than 130% this year.

Gates: Suspending AI research and development cannot "solve the challenge"

On April 3, local time, Gates publicly discussed the open letter signed by thousands of experts in an interview with Reuters,

Calling for a moratorium on AI development will not "solve the challenge". It is better to focus on how to best utilize the development of AI. It is difficult to understand a global moratorium on AI R&D How will this be achieved.

I don’t think asking a specific group to pause will solve the challenge, he continued. Obviously, these things have huge benefits... What we need to do is identify the tricky areas.

Gates is a strong supporter of artificial intelligence.

On March 21, the day before launching the joint open letter, Gates published a blog post saying "The era of artificial intelligence has begun."

It mentioned that artificial intelligence should be used to help reduce some of the world’s worst inequalities.

# He praised artificial intelligence in his blog, "The AI ​​era has begun, and the artificial intelligence revolution is no less important than the birth of mobile phones and the Internet. It is the only revolutionary technological advancement in his life."

So far, Microsoft has invested billions of dollars in OpenAI. OpenAI is building AI technology for Microsoft Bing, Edge, Microsoft 365 and other products.

In the interview, Gates questioned the feasibility of a moratorium on AI development and wanted to know how to implement it.

"The details of any moratorium would be complicated and difficult to enforce. I don't quite understand who they're talking about as being likely to stop, whether every country in the world would agree to a cessation, and why. Stop, there will be many different opinions in this field."

GPT-4 learns to "self-reflect" and its performance increases by 30%

However, even if the whole world Everyone is facing a formidable enemy against super AI, and if they want to rein in the situation quickly, it may be too late.

Because today’s AI has learned to evolve itself!

There is evidence that GPT-4 is improving itself!

Some time ago, former Google Brain research engineer Eric Jang discovered something crazy: GPT-4 can criticize its own results in a reasonable way. .

Instead of looking for the perfect hint for a large language model (LLM) (let’s think about it step by step), let the LLM evaluate its own output and correct its own mistakes immediately.

Even former Tesla AI director and OpenAI research scientist Andrej Karpathy praised this as a good example of why we haven’t seen GPT yet. -4 maximum ability.

In the example given by Eric Jang, I saw a Microsoft researcher tweeting that GPT-4 could not write "non-rhyming" words at all. poetry.

However, when asking GPT-4 again whether it has completed the task. At this time, after apologizing, it generated a non-rhyming poem! It can be said to be full marks.

This just proves that GPT-4 has the ability to "reflect".

Recently, researchers from Northeastern University, MIT and other institutions in the United States proposed Reflexion. This method gives the agent dynamic memory and the ability to self-reflect.

##Paper address: https://arxiv.org/abs/2303.11366

To verify the effectiveness of the method, the researchers evaluated the agent's ability to complete decision-making tasks in the AlfWorld environment, and to complete knowledge-intensive, search-based question and answer tasks in the HotPotQA environment.

The success rates in these two tasks are 97% and 51% respectively.

The overall architecture of Reflexion agent

Some netizens said, You can improve GPT-4's performance by a staggering 30% by asking it to reflect on "Why were you wrong?" and generate a new prompt for itself that takes that error into account until the result is correct.

In addition, the paper points out that Reflexion GPT-4 is also significantly better than conventional GPT-4 when testing the model's ability to write code.

It seems like Eric Jang said,

We can now consider a "Suffix Hint Project", not the clever "Prefix Hint Project". This will encourage LLM to find corrections and inconsistencies in previously generated solutions.

Regarding GPT-4’s ability to make itself smarter through self-improvement, a netizen made a video summary.

Meta researchers published a new paper Toolformer in February, directly targeting the "weakness" of ChatGPT, allowing large language models to learn to use tools on their own.

##Paper address: https://arxiv.org/abs/2302.04761

According to Meta's thinking, large language models like ChatGPT can make up for whatever they lack.

For example, if you don’t know how to do math, just take out a calculator; if you need the latest information, just connect to the search engine and search; you can even directly throw the content you want to translate to it.

The paper became an instant hit. Many AI researchers said that this paper opened up people’s minds and allowed large language models to become a The starting point of all things.

Similarly, the recent paper "Jarvis" HuggingGPT published by Zhejiang University and Microsoft Asia Research Institute has attracted great attention in the industry.

Paper address: https://arxiv.org/abs/2303.17580

In the paper, HuggingGPT proposed by the researchers uses ChatGPT as the controller to connect various AI models in the HuggingFace community to complete various complex tasks such as multi-modality.

Some netizens said that I strongly suspect that the first general artificial intelligence (AGI) will appear earlier than expected. It will rely on "glue" AI that can intelligently glue together a series of narrow artificial intelligence and practical tools.

Automatically write code

Recently, a developer made a project that allows GPT-4 to automatically generate code— —Auto-GPT.

You can even execute code. In subsequent improvements, Auto-GPT can use GPT-4 to write code and execute Python scripts. This allows it to recursively debug, develop and improve itself...

Auto-GPT allows people to further understand the powerful capabilities of GPT-4, Some netizens even made a meme:

##Andrej Karpathy said, in my opinion, the next frontier of prompt engineering is "AutoGPT ”. A GPT call is like an instruction on the computer. They can be concatenated into programs. Use prompts to define I/O device and tool specifications, define cognitive loops, page data input and output in context, and then run.

Netizen: The singularity is coming

Some netizens said that this is basically the definition of the singularity.

The only thing needed now for a self-improving AI is to be able to modify the model’s code and train it autonomously (by browsing the internet and labeling everything on its own).

Some netizens even believe that the rapid advancement of technology has brought us to a point of no return. By 2025, the world will have changed dramatically, for better and for worse.

Learning from mistakes is the foundation of intelligence.

It seems that GPT-4 defeated itself and wanted to become a super AI.

According to the latest news from a certain big boss, OpenAI is very clear that the recent rapid iteration of their AI products has caused extremely strong reactions, and they also welcome public supervision.

The person in charge said that the current risk per token has not reached the critical point, so the abuse of GPT will not cause much harm. The AI ​​alignment roadmap will be released within a few months.

The above is the detailed content of GPT-4 has evolved so much that nearly ten thousand people have signed a petition to block it! The White House held an emergency meeting, and ChatGPT concept stocks plummeted. For more information, please follow other related articles on the PHP Chinese website!