Article reprint source: AI Trends
Original source: Quantum Bit
Image source: Generated by Unbounded AI
There is new news about the ChatGPT crash that caused global netizens to panic.
OpenAI announced that it found records of its servers being hacked during this incident.
The already high traffic after the developer conference, coupled with the hacker attack, caused the server to be overwhelmed, and the original plan to fully launch GPTs on Monday was forced to be postponed.
But the good news is that just now, after some emergency patching by OpenAI, this new version that supports customization was finally officially unveiled.
OpenAI hacked
At 7:49 p.m. Pacific Time on the 8th, or noon Beijing Time on the 9th, OpenAI's status monitoring website added a record of being attacked by DDos.
Simply put, a DDoS attack is when hackers use a large number of controlled devices to frantically send access requests to the target server, causing the server to overload and crash.
A hacker group has already claimed responsibility for the incident, but some people speculate that the attack was carried out by OpenAI's competitors.
As of 1:21 p.m. local time on the 9th (5:21 a.m. Beijing time this morning), OpenAI finally announced that its service status had returned to normal.
20 minutes before the official announcement of the return to normal, OpenAI CEO Altman also brought another exciting news -
GPTs, which has been delayed due to a server crash, is finally going to be available to paying users.
Now that the storm has passed, let’s take a look at what exciting content ChatGPT has updated.
New web version
"Customized GPT" is fully launched
First, GPTs, which can customize robots, are fully open to paying users.
At the developer conference earlier, Ultraman grandly introduced this feature.
Users can deeply customize the robot's "identity", language characteristics, etc., and can also establish their own knowledge base.
The customization process is also very simple. You only need to describe it in natural language under the guidance of GPT Builder.
GPT Builder will automatically complete most of the settings except for its own knowledge base based on our description. These settings can be adjusted through instructions or directly in the Configure menu.
The generated customized version of GPT can be shared through a link. OpenAI’s developer conference also announced the launch of the “GPTs store”, allowing creators to share their own customized robots, but there is currently no entrance in ChatGPT.
However, OpenAI has also prepared some customized robots for users on the GPTs page, including math tutors, cooking assistants, negotiation experts, and more.
There is no third-party content on this platform yet, but some netizens have posted their own customized works on social media:
All in GPT-4
In addition to GPTs announced at the developer conference, the UI of the ChatGPT web version has also undergone significant changes.
A GPTs-related "character selection" has been added to the sidebar, and the model version (3.5/4) switch has also been moved from the center above the chat box to the upper left corner.
GPT-4, DALL·E, web browsing, and data analysis functions have also been merged into the same entrance. The three functions are connected with the basic version of GPT-4, while the plug-in function remains an independent entrance.
Earlier, OpenAI conducted a small-scale test of this integrated entrance and added an "Alpha" option between the buttons of GPT-3.5 and 4, but the overall UI was not adjusted.
After this update, we also tested this comprehensive entrance and designed tasks that can use these functions at the same time:
Retrieve the monthly visit data of ChatGPT since its launch (networking), make a table and draw a line graph (data analysis, source code interpreter), and then draw a promotional poster with this title (DALL·E-3)
Previously, when using the Internet function alone, ChatGPT would often ignore the user's language and answer questions directly in English.
In the comprehensive entrance, the probability of ChatGPT replying in English becomes greater.
Ignoring this, ChatGPT did a good job overall, retrieving the data and drawing the charts as required.
The poster was also successfully drawn, but there were still some detail issues in the processing of the text.
If we combine the newly added multimodal capabilities of GPT-4 and DALL E-3 painting, we can achieve drawing with pictures in ChatGPT.
In addition, we also found that after this UI update, the button for viewing the code in the "problem-solving process" has also changed. After clicking it, the code will be displayed in the form of a pop-up window.
If you are still more accustomed to the way each module is placed separately in the old version, you can also find the independent version provided by OpenAI in Explore.
The "50 items in 3 hours" prompt disappears
In addition, we also found that in the ChatGPT interface, the prompt that GPT-4 can only send 50 messages within 3 hours disappeared.
But this reminder has not been removed from OpenAI's user plan comparison page.
OpenAI has not yet responded to this, but this prompt disappeared briefly in a previous update.
Some netizens on Reddit said that they still encountered a system pop-up prompt of exceeding the limit when playing with customized GPTs.
We can continue to wait and see whether OpenAI will relax this restriction.
Reference link: https://www.bloomberg.com/news/articles/2023-11-09/openai-suggests-cyber-attackers-behind-persistent-chatgpt-outage
