作者 | 褚杏娟、核子可乐
Starting today, ChatGPT's registration-free service will be available in some markets and will be rolled out to the rest of the world. This means that accessing chat.openai.com will no longer require the user to sign in – but they can do so if they want to. Not only can you talk directly to ChatGPT, but you can also use the same basic model as the logged-in user.
Train with your data, with stricter content restrictions
You can join the chat as much as you want, but keep in mind that the feature set used without registration is different from that of a registered account: users who are not registered users will not be able to save or share chats, use custom commands, or other features related to persistent accounts.
In addition, unless explicitly opted off, all chats for unregistered users will still be used as training data. IT'S NOT JUST CHATT, DALL•E IS THE SAME.
This means that you can still choose not to use your chat history for training by clicking on the little question mark in the bottom right corner, then clicking on "Settings" and disabling the option. OpenAI also gives the following GIF for users' reference:
But some people are skeptical, after all, if it doesn't even participate in the training, why would OpenAI open ChatGPT for nothing?
In addition, OpenAI has made a point to emphasize that we will not use content from business products such as ChatGPT Team, ChatGPT Enterprise, and API platforms to train our models.
What's more, this free version of ChatGPT will have a "relatively stricter content policy." What does this mean? A spokesperson for the company gave a lengthy but largely substantive-free reply:
The no-sign experience will benefit from security mitigations that are now built into the model, such as denying the generation of harmful content. In addition to these existing mitigations, we've implemented additional protections specifically designed to address other forms of content that may not be suitable for the registration-free experience.
Based on our understanding of GPT-3.5's capabilities and the risk assessments that have been completed, we have examined various possible behaviors for the use of registration-free services in an inappropriate form.
We don't know exactly what the more restrictive policies are, but with the influx of non-registered users on the site, we may soon find out. "We recognize that additional iterations may be needed and welcome feedback," the spokesperson noted. "As you can expect, there will be not only feedback, but quite a lot!
On this issue, foreign media also asked if they have a plan to deal with the almost inevitable unprecedented attempts at service abuse and model weaponization. After all, even billionaires can't afford to raise such a pure money-burning platform. To this day, inference is still expensive, and even GPT-3.5 models, which are less resource-intensive, require a lot of power and server space. And as soon as the registration-free is opened, people will definitely swallow it up with traffic in an instant.
To this question, OpenAI also gave a long-winded and unnutritious response:
We've also carefully considered how to detect and stop abuse of the registration-free experience, and the team responsible for detecting, preventing, and responding to abuse has been involved throughout the design and implementation of the experience and will continue to guide the design going forward.
As you can see, there's still any real specifics missing here. The spokesperson may not know exactly what the outcome of this decision will be, so he can only sit back and wait for the dust to settle.
The official website is hanging?
It's unclear which regions or groups will be the first to receive the purely free ChatGPT service, but the release of registration-free information may bring more traffic, and some users have difficulty opening the official website and wonder if it is down.
Some netizens who opened ChatGPT also encountered situations where ChatGPT made mistakes and could not respond:
Someone else asked for help on Twitter: Can you still use ChatGPT? Not sure if it's my network problem or a network failure. Most of the people below the post also said that it could not be used.
It is reported that ChatGPT is currently used by more than 100 million people every week, covering 185 countries. As more people poured in, this put OpenAI's underlying infrastructure to the test. Some netizens commented: "Google search traffic enters ChatGPT"
But not everyone is optimistic about OpenAI's decision.
Ethan Mollick, a professor at Wharton, said, "I think OpenAI made a mistake by making 3.5 available without logging in. This could make it harder for people to see what AI can do, as it lags far behind GPT-4. He stressed that GPT-3.5 won't even make it into the top six or seven.
"That's exactly what I've noticed in many of the people I've talked to. They abandoned the big model because they had only tried GPT-3.5. I tend to give them some trial shots of GPT-4-Turbo and Claude 3 Opus, and they're even more impressed. Mira Hurley said.
However, some netizens pointed out that although the use of AI systems has increased, the speed of adoption has not kept up with the speed of innovation. This means that there may be an oversupply in the AI market.
Simon Willison, co-founder of Django, had other ideas, "I wonder how their prevention scraping works? I guess the temptation to abuse this free 3.5 API will be huge. ”
However, for more people, people seem to be more concerned about when GPT-4 will be available for free and when GPT-5 will be released than without registration, after all, a significant number of people have already used ChatGPT.
"I wonder if they'll make GPT-4 a free version. Presumably the end result will be lower running costs, and obviously they will have a better model to attract paying customers. So is it only a matter of time before 3.5 is abandoned, even if it hasn't been released for a long time?"
An outsider "melon"
At the same time, this company's "Gong Dou" melon can also eat a wave.
Reuters reported on April 1 that according to a filing with the Securities and Exchange Commission, OpenAI adjusted the governance structure of its venture capital fund that supports AI startups, and the current high-profile CEO Sam Altman no longer owns or controls the fund.
The change, documented in a March 29 filing, was also raised about Altman's unusual structure, which previously owned the OpenAI Venture Fund. While the fund operates in a similar way to a venture capital arm within a business, the fund itself was raised by Altman from outside limited partners, who make the investment decisions himself. OpenAI said that although the fund is owned by Altman, he does not enjoy the financial benefits of the fund.
Foreign media Axios first reported the news of the change of ownership of the fund on Monday. An OpenAI spokesperson noted in a statement that the fund's initial general partner (GP) structure was temporary, and "this change further clarifies the situation."
The $175 million in investment currently held by the OpenAI Venture Capital Fund comes mainly from OpenAI partners such as Microsoft, but OpenAI itself is not a funder.
Control of the fund has been handed over to Ian Hathaway, who has been a partner of the fund since 2021, according to the filing. Altman will no longer serve as the fund's general partner.
According to OpenAI, Hathaway oversees the fund's accelerator program and has led investments in companies such as Harvey, Cursor and Ambience Healthcare.
Altman himself is the former president of startup accelerator Y Combinator and has previously been closely watched for building a vast investment planning system outside of OpenAI (including crypto startup Worldcoin and fusion company Helion Energy) and fundraising in the Middle East.
OpenAI said that after receiving a "raid" oust from the company last November, Altman had undergone a follow-up independent investigation that concluded that he had no wrongdoing in terms of product safety or OpenAI's finances.
Original link:ChatGPT makes the official website hang without registration?Professor Wharton: OpenAI made a wrong decision_AI& large model_Chu Xingjuan_InfoQ Selected articles