A new GPT is born in 1 minute! Customized GPT exploded within 3 days, and the ideal boyfriend and scientific research tool were all over the Internet.

Article reprint source: AI DreamWorks
Source: New Wisdom
In just three days, the world has seen an explosion in GPT applications, and the prototype of a blockbuster product has emerged.
A new GPT is born in 1 minute!
In less than a week, various customized GPTs exploded around the world, and the growth rate has completely exceeded everyone's imagination.
This just confirms what Altman said at the developer conference: "We are breeding new species, and they are proliferating rapidly."
The entire Internet is predicting the pessimism of Agent startups, and even the research direction of developers is about to undergo earth-shaking changes.
 
On the other hand, a tweet from Karpathy added fuel to the flames and exposed OpenAI's ambition to make the GPT model an operating system for large models.
In a joking way, he also gave the specific "technical specifications" of the big model OS:
• LLM: OpenAI GPT-4 Turbo 256 core (batch size) processor @ 20Hz (token/s)
• RAM：128K token
• File system: Ada002
By the way, Sam Altman just announced that the new version of GPT-4 Turbo is also available on ChatGPT, which means that everyone can use 128k context.
After GPTs was opened, people's creative enthusiasm was instantly aroused, and its potential lies in the creative outburst of genius. This form of UGC will always produce a hit of the era.
The most complete series summary of customized GPT
It turns out that ChatGPT is actually just a trial version, and GPTs is the real AI product launched by OpenAI.
After the release of GPT-4, it took OpenAI just over half a year to transform itself from the world's best basic model provider to the world's best AI application provider.
There are now more than 3,000 GPTs applications listed on various online platforms!
In GPTs, the most creative people from all walks of life around the world are using their experience and data to help OpenAI develop AI applications.
From accurate information services in professional fields to answers to life's questions.
From being a powerful assistant for academic research, the GPTs that appeared in just three days have an application scope that covers almost any field you can imagine.
I really didn’t expect that the famous saying of the AI ​​tycoon, “All applications need to be reconstructed by AI”, would happen in just 3 days.
We have selected the most popular GPTs on the Internet right now to show you how deep this rabbit hole goes!
C-RPG game generates GPT
A GPT application called Retro Adventures. It can create a pixel-style C-RPG text game for you on the spot.
Just give it a specific theme and it will generate a specific game. We chose "Twenty Thousand Leagues Under the Sea" and this is the beginning it generated:
After introducing the background of the game through a picture and a paragraph of text, the game is completed. The gameplay is very similar to C-RPG or mud games, and the game content is experienced through dialogue.
After choosing to drive a submarine to explore the mysterious waters, the game continues:
Players can choose one of the three options generated each time to advance the game, or they can directly enter what they want to do. Retro Adventures can generate very interesting game content for you.
This time, I did not follow its advice and chose to advance the game according to my own ideas. I replied: I found a hot pot restaurant on the bottom of the sea and went in to try it.
As a result, it really followed my thoughts and took me in to experience what it was like to eat hotpot at Twenty Thousand Leagues Under the Sea!
Then we reopened Westworld, which was also a very interesting gaming experience.
Dream Sticker Maker
There is also the "Dream Stickers" GPT officially produced by ChatGPT, which allows users to directly turn the content of their dreams into usable stickers and have them delivered directly to their homes!
According to the introduction, we created a very "wild" dream:
  The ideal boyfriend is here
In response to the calls from netizens, the well-known blogger @dotey made a "Boyfriend GPT".
This boyfriend is very tolerant.
Characteristics: Humor without losing depth, childlike innocence without losing sense of propriety, idealism without losing simplicity, and most importantly, always remaining stubborn and kind in the face of fate.
https://chat.openai.com/g/g-IlNu7BVYQ-your-boyfriend-alex/c/b9d330d5-a4e5-4b76-a0ba-7a7690446f75
Let’s all test it out and see if this boyfriend is qualified.
Shop with GPT and enjoy hassle-free shopping
Quinn Leng, an engineer from Databricks, created a "Shopping GPT".
Its power lies in the fact that any object you take a picture of can be automatically identified, compared with prices across the entire network, and provided with purchasing suggestions.
 
For example, you can enter a picture of fitness equipment, and then "Chua Shou GPT" will start a full-network search and summarize a specific list for you.
A powerful tool for scientific research, AI paper polishing experts are online
"AI Paper Polisher Pro" is simply an AI scientific research tool.
It provides straightforward suggestions for improving AI conference papers, focusing on paper structure, technical precision, and LaTeX code for visual elements.
It can even analyze screenshots of your paper and provide feedback on different levels, including overall layout and structure, as well as detailed writing suggestions.
Some netizens said, this is what I want!
Too many GPTs, one question to solve
A domestic developer said that there are now thousands of GPTs emerging and he really doesn’t know where to start, so he built a GPT specifically for “recommended GPTs”.
https://chat.openai.com/g/g-iD7sLuO9S-gptofgpts
For example, if you want to learn English, you don’t have to look hard for it, just ask “GPT of GPTs”.
Chinese Little Chef, one-click creation of various classic Chinese dishes
GPT address: https://chat.openai.com/g/g-VrmZ5hQPP-zhong-hua-xiao-chu-shen
In addition to telling you how to cook, it can also recommend various dishes to you based on your lifestyle.
Want to make your own GPT but don't know how? Let GPT teach you
GPT address: https://chat.openai.com/g/g-Z0f6pPPGc-gpt-builder-builder
3D Printing Master, Get Started with 3D Printing in 5 Minutes
GPT address: https://chat.openai.com/g/g-W0lCzVAZH-3d-print-master
After a simple search based on the content, we found that these 4 recommended models are indeed quite reliable.
GPT-4 Turbo performance stuck at 73K
On the first Developer Day, Altman upgraded the GPT-4 API to GPT-4 Turbo, and updated the time to April 2023.
Most importantly, the maximum context length can be 128k, which is equivalent to 300 pages of text content.
Its price is about 1/3 of GPT4, and its speed is almost 4 times faster.
Just feel how fast it is.
But to find out the real performance of GPT-4 128K, an engineer conducted a stress test on it.
The results show that the recall performance of GPT-4 begins to decline when the prompt exceeds 73K tokens.
The recall performance is related to the position of the recalled fact, and the performance is low when it is between 7% and 50%. If the fact is at the beginning of the document, it will be recalled regardless of the context length.
This means that GPT-4 cannot guarantee 100% accuracy in retrieving facts, so don’t assume that it will retrieve all the facts.
So what is a good solution?
As we all know, less context = higher accuracy. Providing less contextual input can improve the accuracy of GPT-4's recall of facts.
In addition, the position of the fact in the document is also important. Adding the facts to be recalled at the beginning and the lower part of the document will help the model recall more.
The test process of the engineer Greg Kamradt is as follows, using Paul Graham's 218 articles as the "background" token.
Then, a random sentence is placed at different depths in the document, using the fact "The happiest thing in San Francisco is sitting in Dolores Park and eating a sandwich on a sunny day."
Ask GPT-4 to answer this question using only the context provided, use LangChainAI to eval another model (GPT-4) to evaluate GPT-4s answer.
Next, this process is repeated for 15 document depths (from 0% top of document to 100% bottom of document) and 15 context lengths (from 1K to 128K tokens).
Two tests were run with larger context lengths to derive higher performance.
Finally, Kamradt said, "While I think this is directionally correct, more testing is needed to more clearly understand GPT4's capabilities."
The cost of this test was about $200 (1457 yuan), and the cost of a single call with 128K input tokens was $1.28.
The API call cost for this test was ~$200
The source code of this test project has been open sourced, and those who are interested can test it.
https://github.com/gkamradt/LLMTest_NeedleInAHaystack
GPT-5 starts grayscale testing?
Some netizens discovered that OpenAI has begun grayscale testing of two new features: Magic create and Gizmo.
Many people may be familiar with the name of the Gizmo model, which is the legendary GPT-5. Its training time ends in January 2022.
What do you think?
References:
https://twitter.com/rickawsb/status/1723217817531801992
https://twitter.com/dotey
https://twitter.com/lencx_/status/1723226804671135958