OpenAI Announces 2024 Plans: GPT-5, Open Source, and a More Powerful ChatGPT!
-
At the end of the year, amidst the festive Christmas atmosphere, OpenAI has outlined its development plans for 2024.
OpenAI co-founder and CEO Sam Altman announced on social media that AGI (a bit later), GPT-5, better voice models, and higher rate limits are on the agenda.
Better GPTs, improved reasoning capabilities, control over wake/behavior levels, video models, personalization, enhanced browsing, and open-source initiatives are the goals OpenAI aims to achieve in 2024.
Most of the product plans listed by Sam are capital-intensive. According to Bloomberg, OpenAI is seeking a new round of financing at a valuation of $100 billion or more, while also negotiating a multi-billion-dollar chip collaboration plan with G42.
Therefore, OpenAI's announced 2024 product plan is not just empty talk but a serious commitment to substantial action.
Below, the "AIGC Open Community" interprets OpenAI's major product plans for 2024 based on their past product releases and technological trends.
GPT-5
The earliest information about GPT-5 dates back to July 31 of this year. At that time, a trademark attorney named Josh Gerben represented OpenAI in filing trademark applications for GPT-5.
According to information displayed by the United States Patent and Trademark Office, OpenAI submitted the application on July 18. The functionalities provided by GPT-5 include natural language processing, text generation, comprehension, speech transcription, translation, prediction, and analysis, though the actual released features may vary.
Additionally, before filing the trademark application, OpenAI took an important step by reaching an AI safety agreement with authorities, voluntarily committing to managing various risks posed by AI. The company pledged to adhere to three core principles—safety, security, and trust—in future product releases and usage processes.
In other words, this paves the way for OpenAI to safely develop GPT-5, removing the shackles.
On November 9, OpenAI announced the formation of a 'Data Collaboration Alliance,' primarily focused on text, images, videos, audio, and especially those ultra-long, cross-language, non-public private data that deeply express human intentions and emotions.
Sam once stated that besides a stronger technical architecture and parameters, massive high-quality and private data will be a key foundation for developing GPT-5.
This is because it helps large models deeply understand the operational models of various industries worldwide, the languages/cultures of different ethnic groups, and human work thinking/intentions.
On November 14, the Financial Times reported that OpenAI is seeking a new round of funding from Microsoft to develop products like GPT-5. This once again proves that developing GPT-5 will be a highly capital-intensive project. Therefore, it is clear from the above news that OpenAI's commitment to GPT-5 is not just empty talk or an unrealistic vision, but a solid step-by-step advancement of the entire plan.
Better GPT
If there's one product from OpenAI's 2023 releases that left the deepest impression, I'd choose Custom GPTs.
Custom GPTs were unveiled by OpenAI at its first global developer conference on November 7th this year.
Users can build custom GPTs without any coding knowledge, using only visual click operations. Simply provide ChatGPT with conversation instructions or additional knowledge data, then choose whether to enable multimodal features like web search, data analysis, and image generation. Custom GPTs enable non-programmers to participate in generative AI while allowing them to share their vast amounts of high-quality data with others.
In less than two months, users have created over hundreds of thousands of specialized ChatGPT assistants through custom GPTs, demonstrating their immense popularity.
If OpenAI launches the "GPT Store" in 2024, it is likely to spark another wave of application enthusiasm.
Open Source
In fact, OpenAI has been continuously working on open source projects. For example, the speech recognition model Whisper on GitHub has garnered over 53,000 stars;
On GitHub, the text-to-3D model Shap-e has over 10,000 stars, along with other projects like the Consistency Decoder designed to enhance the coherence of text-to-image models.
Whisper open source address: https://github.com/openai/whisper
Shap-e open source address: https://github.com/openai/shap-e
Consistency Decoder open source address: https://github.com/openai/consistencydecoder
None of these can meet our needs, what we really want is ChatGPT!
Not long ago, 57 renowned technology and academic research institutions including Meta, Oracle, Intel, Meta, AMD, IBM, Sony, and Dell formed an 'Open AI Grand Alliance'.
This alliance has R&D expenditures exceeding $80 billion, student resources from academic institutions exceeding 400,000, and total membership exceeding 1 million people.
This resembles the "Eighteen Route Warlords" in Romance of the Three Kingdoms, aiming to compete with closed-source products like ChatGPT, Midjourney, and Gen-2. Therefore, under immense pressure from the open-source community, OpenAI's so-called "open source" in 2024 likely refers to the GPT series of large language models used by ChatGPT.