Shulou(Shulou.com)11/24 Report--

Is there a big reversal? Stack Overflow's "ChatGPT ban" was suddenly loosened, and moderators had to go on strike to protest, tearfully complaining that the spam generated by LLM would flood the community. As of press time, 227 users had signed the open letter. (and it's growing.)

As you all remember, when ChatGPT first came out, its code generation function made people feel very fresh.

However, with a flood of plausible and undistinguishable answers, Stack Overflow had to be banned overnight-ChatGPT, banned!

Specifically, Stack Overflow moderators have the right to block accounts suspected of posting AI-generated content such as ChatGPT, and delete posts directly.

However, Stack Overflow recently released a new rule:

In view of the fact that we can not accurately identify the content generated by AI at present, the probability of "false positive" is very high, and this arbitrary excessive prohibition may make the site lose a large number of contributing netizens.

Therefore, moderators can ban accounts only if they are truly verifiable. Whether it is subjective speculation such as writing style, or the results of the GPT detector, it can not be used as a measure.

As soon as this regulation was issued, it made the moderators explode instantly!

In their view, this is simply a blatant acquiescence to the wanton spread of the LLM illusion, a public destruction of the clean environment of the community, a loss of morality and a distortion of human nature.

Some of the moderators have gone on strike angrily.

To protect our community from the encroachment of AI, some moderators and users wrote an open letter angrily complaining that the new rules will completely destroy Stack Overflow, the holy place of code Q & An in everyone's heart.

They said indignantly that the content generated by AI posed a "major threat to the integrity and credibility of the platform and content". The operator's decision will undermine Stack Overflow's goal of becoming a "high-quality information base".

Once upon a time, moderators could delete posts or seal their own names as long as they determined that the content of the post was generated by AI.

But today, the standard has become very stringent. The usual methods of the past are no longer applicable.

These moderators believe that this will mean that the content generated by AI can be published freely on Stack Overflow, regardless of what the community thinks of it.

From then on, as soon as we open Stack Overflow, we are likely to see an eyeful of error messages and plagiarized content.

In addition, the new deal will also deprive the Stack Exchange community of the leeway to define its own policies.

Unfortunately, so far, operators and moderators have not achieved good direct communication, and the last thing moderators can do is to choose to strike and no longer censor posts on the platform.

In their view, their own strike is the last resort to save the community from the loss of full value.

After the full text of the moderator's protest letter was deleted from AI, Stack Overflow issued a new "ban" to allow incorrect information and unfettered plagiarism to be rampant on Stack Exchange. This poses a major threat to the integrity and credibility of the platform and its content.

The following signers are volunteer moderators, contributors and users of Stack Overflow and Stack Exchange.

From now on, we will launch a general strike against Stack Overflow and Stack Exchange to protest against the changes in policies and platform rules that Stack Overflow will impose on us now and in the future.

Our efforts to achieve platform change in a reasonable way have been ignored and our concerns have been ignored. Now, as a last resort, we want to withdraw from the platform that we have invested in caring and volunteering for more than a decade.

We firmly believe in Stack Exchange's core mission: to provide a high-quality information base in the form of questions and answers, and Stack Overflow's recent actions have directly undermined this goal.

Specifically, moderators are no longer allowed to delete AI-generated answers in some cases on the grounds of AI-generated answers. This allows almost all AI-generated answers to be released freely, regardless of the community's established consensus on such content.

This, in turn, makes incorrect information (commonly known as "hallucinations") and plagiarism rampant on the platform.

As Stack Overflow pointed out before, this undermines our trust in the platform.

In addition, the policy details sent directly to the moderator are very different from the guidelines outlined in public, and the moderator is prohibited from sharing these details in public.

These policies ignore the leeway previously given to each Stack Exchange community to decide their own policies, directly modify them without exchanging views with the community, overturn the consensus of the community, and refuse to think twice.

Until this matter is satisfactorily resolved, we will suspend these activities, including, but not limited to, the following:

Publish and process flag

Run SmokeDetector, anti-spam robot

Close or vote to close posts

Delete or vote to delete posts

Review tasks in the queue

Run robots that assist in auditing, such as detection of plagiarism, low-quality answers and rude comments

Before Stack Overflow withdraws this new policy to address moderators' concerns and allows moderators to effectively implement established policies on AI-generated answers, we call for a total strike as a last-ditch effort to protect the Stack Exchange platform and users from a loss of full value.

At the same time, we would also like to remind Stack Overflow that a network that relies entirely on volunteers cannot always ignore, abuse and point the finger at these volunteers.

Stack Overflow's new policies and regulations state that Stack Overflow officials have conducted a series of analyses of the current content review methods and found that the content review mechanism generated by AI is not accurate.

In other words, the judgment on whether some content is generated by AI is not necessarily correct, and the responsible moderator or volunteer may make a misjudgment.

Officials worry that the move will encourage prejudice against users in certain regions or countries, or shut out a large number of legitimate content publishers.

To solve this problem, the official requires the moderator responsible for reviewing the content to use very strict evidence standards to determine that the content posted by the user is generated by AI before being blocked.

Under the new rules, the useful standards commonly used by moderators, such as user writing style and forum behavior, will no longer be used because it is not "strict".

Officials say the method commonly used by moderators cannot be ruled 100%.

At the same time, GPT detectors are not officially desirable. The misjudgment rate is too high to be accepted and can not be used as a reliable index of judgment.

In short, the new rules officially limit the power of moderators. Only conclusions drawn from a rigorous review of evidence can be adopted and further action can be taken (such as closing accounts, etc.).

Any way of relying on intuition and guessing should no longer be used.

Why do we have to strike? A moderator's blood and tears accused a moderator of posting on Stack Overflow, explaining in detail why the moderators had to go on strike.

Ta says people are doing this more or less for the benefit of the community as a whole.

In late November 2022, with the launch of ChatGPT, a large number of users began to move up the answers generated by AI frantically.

The answers generated by ChatGPT are very similar to those of human users.

For example, the following netizen said that he first created a new account, then posted the random questions without answers into ChatGPT, and then po the generated answers.

Within an hour and a half, ta gave nine responses to po and got one acceptance, three approvals, and 62 popularity values.

Subsequently, these random copies and pastes from ChatGPT's answers quickly had a huge impact, and the netizen was looking forward to the follow-up.

The intrigued netizens attracted the attention of the audit team, who sounded the alarm and immediately made new rules for Stack Overflow.

The new rules make sense. Since its establishment in 2008, Stack Overflow has relied on community volunteer audits to maintain its operation.

But the answers generated by ChatGPT usually look good, and countless netizens are frantically posting these answers, which puts a huge burden on volunteers.

As a result, the community had to issue a ban.

The moderator said that after the ban was issued, the moderators worked conscientiously and directly participated in the community management team to help implement the new policy in this community in all aspects.

And the process of dealing with posts is also very troublesome.

The audit team will use JavaScript scripts to help with the audit, and when they need a tool, they will write it directly. They will also expand the functionality of the existing website.

Every day, moderators deal with a large number of posts and experience situations such as being abused.

As a result, moderators were very tired after receiving the new ban on Monday. After all, everyone is a volunteer, not a regular employee. They've done enough for the community.

Strikers' demands what do strike moderators want? In the final analysis, the problem is solved.

There are some issues that need to be personally acknowledged by officials, and some issues that employees need to discuss openly to figure out exactly what to do.

ChatGPT rules are made by the whole community and have been widely supported.

Employees should not just use private channels (chat rooms and moderator teams) to make large-scale rule changes, which means that the majority of users will not know.

Moderators are asked to deal with flag according to a rule that has never been made public. In fact, everything on the site shows that the old policy is still in effect.

As a result, moderators will still receive flag that cannot be handled in accordance with this policy. Eventually, the community is sure to notice the difference.

A small number of moderators who post about the policy changes before the announcement is made public are actually risking losing their rights and not being able to access the moderator team. This is very unfair to the moderator.

In a post angrily denouncing the new rules, Philippe said the possible racial discrimination that officials were worried about was a naked accusation. But it is clear that no one wants to be accused of racial discrimination without cause.

In the worst case, the new rules reflect employees' perception of moderators and communities as opposed to each other.

Employees are very vague about policies (as mentioned earlier, there are differences between public and private policies) and do not provide specific examples of what they consider to be incorrect.

They want moderators to stop using GPT detectors as the only criterion, of course, but for someone who has posted more than a dozen long answers in two hours, it's clear that they're not all written by themselves, they're just copying and pasting questions.

According to policy, this is unacceptable, but how should moderators deal with it?

Can moderators use testing tools without direct evidence of plagiarism?

However, the reality is that no one tells the moderator what they should do.

Reference:

Https://openletter.mousetail.nl/

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.