MLNLP community is a well-known machine learning and natural language processing community at home and abroad, covering NLP master’s and Ph.D. students, university teachers, and corporate researchers.

The vision of the community is to promote communication and progress between the academic and industrial circles of natural language processing and machine learning, especially for the progress of beginner students.

Reprinted from | Xi Xiaoyao Technology Talk

Author | Fu Milk Tea

How powerful is GPT-4o? Just released for two days, netizens have shared various evaluations. Milk Tea has collected some cool uses discovered by netizens.

Next, Milk Tea invites everyone to appreciate these cool applications!

The Terrifying Stock Selection Tool

A Twitter user shared their experience using GPT-4o for stock selection, claiming it is “terrifyingly powerful” and “efficiency beats GPT-4”. Let’s see how they did it:

They used GPT-4o to automatically rewrite over 200 lines of stock selection indicators into an automated stock selector, outputting charts and data archiving. After one round of detailed modifications, they claimed “efficiency beats GPT-4”.

Although Milk Tea is not yet clear about the meaning of each column of indicators, judging from the output charts, True or False may be the judgment results given by the model.

Alright, if you can’t select stocks, you no longer need to force it, because your strength has arrived!

According to netizens, it seems to have indeed brought returns:

Design Modeling Master

It seems that GPT-4o is also quite skilled in the design field! It appears capable of creating new ideas and concepts from completely unrelated images. For example, it can transform two images into an advertising concept, such as “Sad Frog Head Cereal”, which is quite eye-catching.

Some netizens also tried these two images:

GPT-4o created a new species–Frog Turtle [Dog Head Emoji].

In 3D modeling, GPT-4o has also shown impressive capabilities. It can convert 2D images into 3D models, such as performing a 3D reconstruction of the OpenAI logo, with smooth and natural effects.

Even more surprisingly, GPT-4o can model based on text requests without needing images, such as creating a table with four legs and adding some random features:

Perfect Tutor

GPT-4o also demonstrates powerful capabilities in tutoring:

Indeed, GPT-4o can explain geometric math problems to children with great patience and accuracy. This feature will be a huge lifesaver for many parents!

Personal AI Assistant

Sam Altman has also promoted the GPT-4o assistant:

When working on the computer, you have to constantly switch between different pages to find information. But if you use GPT-4o as a personal AI assistant, you can ask it questions anytime, allowing you to stay focused.

Master-Level Simultaneous Interpretation

At OpenAI’s press conference, GPT-4o’s real-time translation capabilities were demonstrated, allowing for smooth conversion between English and Italian without delays or stuttering. GPT-4o uses a new tokenizer, which not only improves efficiency in English (tokens reduced by 1.1 times) but also enhances efficiency in many other languages.

Master of Reading Between the Lines

GPT-4o’s emotional capabilities are also quite astonishing, such as the ability to interpret micro-expressions on faces:

Okay. You look a bit tired, which may indicate mental or emotional fatigue. A smile suggests you are trying to appear positive or approachable, but there is an underlying sense of pressure or concern. The way you furrow your brow and your gaze may suggest you are deep in thought or dealing with some challenges. Overall, you appear calm and composed, but underneath, there are more complex emotions.

(The above recognition results for the embedded text in the image come from GPT-4o)

In addition, through responses in situational dialogues, it can be observed that GPT-4o is more inclined to express its own emotions and awareness, showing that it pays more attention to emotional expression and subjective experience when simulating human conversation.

Summary

The main research and development team of GPT-4o, Omni, is led by Prafulla Dhariwal, who expressed gratitude to team members James Betker, Jamie Kiros, and others on social networks, revealing that the development of GPT-4o began a year ago. This model is OpenAI’s first native multimodal large model.

In addition to the aforementioned innovative applications, the strength of GPT-4o is also evident:

Moreover, GPT-4o can solve problems that were previously unsolvable:

Everyone is welcome to explore together and share your interesting discoveries with us!

References

[1]https://x.com/rackslabs/status/1790372310442000555?s=46[2]https://x.com/minchoi/status/1790396782200987662?s=46[3]https://x.com/sugurukun_ai/status/1790286932305641533?s=46[4]https://twitter.com/eviljer/status/1790421640683352203[5]https://twitter.com/mckaywrigley/status/1790088880919818332

Technical Exchange Group Invitation

△ Long press to add the assistant

Scan the QR code to add the assistant WeChat

Please note: Name-School/Company-Research Direction

(e.g., Xiao Zhang-Harbin Institute of Technology-Dialogue System)

to apply to join Natural Language Processing/Pytorch and other technical exchange groups

About Us

MLNLP Community is a grassroots academic community jointly established by machine learning and natural language processing scholars at home and abroad. It has now developed into a well-known machine learning and natural language processing community at home and abroad, aiming to promote progress between the academic and industrial circles of machine learning and natural language processing and a wide range of enthusiasts.

The community can provide an open communication platform for practitioners’ further studies, employment, and research. Everyone is welcome to pay attention to and join us.