Starting today, this important feature is fully open to ChatGPT Plus users!

Estimated read time 5 min read

Summer is getting closer and closer, and the countdown to the debut of GPT-4.5/5 is expected to begin. New news about ChatGPT is starting to emerge.

OpenAI CEO Sam Altman still maintains his usual practice and throws out a few riddle-like posts on social platform X from time to time, which really whets our appetite.

Then let’s get straight to the point and take a look at some recent news about OpenAI that is worthy of attention (no fidelity~).

Will ChatGPT understand you better than your family?

Early this morning, OpenAI officially announced that it is launching a memory function for ChatGPT Plus accounts.

Simply put, the feature allows users to tell ChatGPT things they want them to remember in a chat. In this way, ChatGPT can not only remember what you chatted with it last time, but also bring up those old memories in new conversations.

Of course, this feature can be turned on or off at any time.

For example, OpenAI once said that if you tell ChatGPT that the children at home like jellyfish, then when you ask it to help create a birthday card one day, it will suggest using a jellyfish-patterned party hat.

At this rate of development, ChatGPT may know you better than your family in the future.

It is worth mentioning that OpenAI said that the memory function is currently “unavailable” in Europe and South Korea. As for the specific reasons, OpenAI did not say. In the future, this feature will also be rolled out to team and enterprise customers, as well as other versions of GPTs.

In fact, this function has been quietly open to some users for testing as early as February, and today it is further expanding the scope of the open testing.

Netizen @ai_for_success reminded that although this function automatically saves your personal preferences and other detailed information, it may also slow down the operation of GPT-4, so it is recommended to turn it on with caution.

OpenAI’s AI search engine is coming

Just as Perplexity and Google are competing, OpenAI, the “leader” in generative AI, seems to have dropped a blockbuster – the AI ​​question and answer search engine SearchGPT.

According to netizens, similar to Perplexity, SearchGPT can not only search the entire network through chat Q&A, but also perform image searches and even provide various widgets, such as weather, calculator, sports, finance and time zone.

Moreover, it can also conduct interactive questioning, and it can also summarize the content of the web page after entering the URL. It can be called an all-round player.

In addition, SearchGPT uses multiple models such as GPT-4 Lite, GPT-4 and GPT 3.5, and combines multiple search engines such as Bing and Fortis.

Some netizens speculate that GPT 3.5 may be used for fast searches, while GPT-4 Lite may be used to handle more complex tasks, such as encoding. All these functions can be integrated into a single model if the user wishes.

Previously, Sam Altman said in an interview that the world does not need a replica of Google. There may be a new and better way to help people find, utilize and integrate information.

Now it seems that the launch of SearchGPT is exactly the new method he said.

When the generative AI referee goes off to play football in person, Google is expected to have another red alert, but before that, competing product Perplexity may be trembling.

What’s even more interesting is that as the largest sponsor of OpenAI, will Microsoft’s Bing also directly compete with SearchGPT? By then, the rift in the relationship between the two is expected to further widen.

GPT-4.5 model leaked in advance?

After Claude 3 and Llama 3 “exploded” one after another, what we are most concerned about is the new model of ChatGPT.

No, recently a new model called gpt2-chatbot has been pushed to the forefront. Its performance is close to the GPT-4 level, and even surpasses GPT-4 in some aspects. However, gpt2-chatbot has a limit of only 8 messages per day.

Reddit netizens believe that this is at least as good as GPT-4 Turbo, and even the system information dug up shows that gpt2-chatbot is a large model trained based on the GPT-4 architecture.

Netizens also found that gpt2-chatbot can not only do what GPT-4 can do, but it is also better at following the user’s will and freedom.

Moreover, its performance in drawing ASCII graphics is also impressive. Whether it is a unicorn or other animals, it can draw a simple image. gpt2-chatbot also seems to be better than Claude Opus or the latest GPT 4-Turbo on complex code manipulation tasks.

Netizens pointedly pointed out that there is a substantial difference in the expressions of GPT-2 or GPT 2.

While everyone was talking about it, the person involved, Sam Altman, also joined the ranks of “eating melons” and posted on social platform X that “I really have a soft spot for GPT-2”, adding to the popularity of gpt2-chatbot. A fire.

Attached is the experience address: https://chat.lmsys.org/ (Click “Direct chat” in the upper right corner and select the model)

I have to say that OpenAI knows how to stimulate our emotions.

In an interview with the media, Sam Altman once predicted that from the establishment of OpenAI to the time we create AGI, there will definitely be some crazy and explosive things happening, and there may be more crazy and explosive things happening in the future.

Now it seems that his prediction may be coming true step by step.

You May Also Like

More From Author

+ There are no comments

Add yours