OpenAI Japan has unveiled plans for its next AI model, ‘GPT-Next,’ which is set to be released by the end of 2024. According to Tadao Nagasaki, CEO of OpenAI Japan, this new model is expected to be 100 times more powerful than GPT-4, yet it achieves this substantial performance boost without requiring a massive increase in computing resources. Should i learn software engineering then?😭 — Hagay (@Hagay102) September 4, 2024 GPT-Next is set to transform AI with 100x more power than GPT-4, enabling faster processing, smarter decision-making, and groundbreaking advancements across industries like healthcare, finance, and media. At the KDDI Summit 2024, OpenAI Japan presented a visual representation highlighting a dramatic increase in the capabilities of GPT-Next compared to its predecessors. The graph depicted how the new model, GPT-Next, would achieve what Nagasaki referred to as “three orders of magnitude” improvement over GPT-4, showing the company’s commitment to pushing the boundaries of artificial intelligence. No more 6-finger generated image and cross-eyed portrait results, I presume. — ㊬ 昭栄竜介・Ryūsuke the illustratorꉂ ෴ (@shogunated) September 4, 2024 This ambitious development is underpinned by two critical projects: ‘Strawberry’ and ‘Orion.’ Project Strawberry is a new AI system that improves reasoning, problem-solving, and language processing skills. Unlike earlier models that rely on generating quick responses, Strawberry employs a more thoughtful approach, allowing the AI to deliberate and reason through complex problems. RIP to more jobs I guess — 🎗️Sinah 🇺🇸🇮🇱✡︎ (@wisp_sin) September 4, 2024 Another major feature of Strawberry is its ability to produce high-quality synthetic training data, which is crucial when there is a lack of diverse, high-quality data to train AI models. This innovative approach could also help lay the groundwork for future models like Orion. How is 100x more powerful quantified? — DSF (@DougF66) September 4, 2024 The model’s development includes training with data generated by Strawberry, and it has been refined using substantial computational resources, equivalent to 10,000 H100 GPUs. This represents a hefty investment in time and resources to ensure the model performs reliably and efficiently across various applications. Impressive feat. AI creating societies mirrors human complexity fascinatingly. — Alexander De Ridder (@adridder) September 4, 2024 With competitors like Anthropic’s Sonnet 3.5 excelling in general coding tasks and Meta’s LLaMA showing strengths in specific areas like Go, OpenAI’s new models are poised to offer a distinct advantage in handling complex and varied inputs. Looking further ahead, OpenAI plans to release another model, Orion, in 2025. This model is expected to be even more advanced, using training data from Strawberry to enhance its capabilities further. Show don’t tell OpenAI. Put it in our hands and we will then give credit if this is legit, we are not looking for vaporware. — Jordan Thibodeau (@JWthib) September 4, 2024 The announcement of these new models reflects OpenAI’s ongoing efforts to innovate and develop tools that can handle a wide range of tasks, from everyday consumer use to specialized applications in fields like programming and data analysis. OpenAI’s strategy is building better models and finding ways to make AI more efficient, accessible, and effective for various purposes. Wouldn’t that be GPT-4 with Strawberry? Or GPT-Orion? — Scott the VRNerd (@TheVRNerd) September 4, 2024 For more news and insights, visit AI News on our website.
Instead, the improvements are based on better architecture and more efficient learning methods, allowing the model to handle complex tasks more effectively.
GPT-Next is expected to offer advanced multimodal capabilities, enabling it to process and understand not just text but also images and videos, making it a versatile tool for a range of applications.
This is referred to as “System 2 thinking,” where the AI takes time to analyze and solve issues, especially in areas like mathematics and programming.
Project Orion, which may ultimately be branded as GPT-Next, is designed to succeed GPT-4 with a broader scope. Orion aims to excel at tasks involving multiple forms of data, integrating text, images, and videos seamlessly.
Additionally, OpenAI has been clear about its intention to maintain a strong position in the AI sector by continuing to develop and refine its models.
Orion has already undergone months of training using a large number of GPUs, and it is set to provide even greater improvements in processing and understanding different types of data.
These developments show that OpenAI aims for significant strides in AI technology, focusing on making its models more powerful and versatile without requiring vast amounts of additional computing power.
OpenAI’s ‘GPT-Next’ to Arrive by End of 2024—A New Era for Artificial Intelligence?
-
Midhat Tilawat is endlessly curious about how AI is changing the way we live, work, and think. She loves breaking down big, futuristic ideas into stories that actually make sense—and maybe even spark a little wonder. Outside of the AI world, she’s usually vibing to indie playlists, bingeing sci-fi shows, or scribbling half-finished poems in the margins of her notebook.Midhat Tilawat
Midhat Tilawat is endlessly curious about how AI is changing the way we live, work, and think. She loves breaking down big, futuristic ideas into stories that actually make sense—and maybe even spark a little wonder. Outside of the AI world, she’s usually vibing to indie playlists, bingeing sci-fi shows, or scribbling half-finished poems in the margins of her notebook.
Key Takeaways:
Was this article helpful?
YesNo
Midhat Tilawat is endlessly curious about how AI is changing the way we live, work, and think. She loves breaking down big, futuristic ideas into stories that actually make sense—and maybe even spark a little wonder. Outside of the AI world, she’s usually vibing to indie playlists, bingeing sci-fi shows, or scribbling half-finished poems in the margins of her notebook.
