Is GPT-5 on the Horizon? Exploring OpenAIs Project Strawberry

OpenAI Delays Launch of GPT-5 Following Poor Results and High Costs

Concerns about a model significantly more powerful than GPT-4 have been raised from very early on. Shortly after the release of GPT-4, a petition signed by over a thousand public figures and tech leaders has been published, requesting a pause in development on anything beyond GPT-4. Significant people involved in the petition include Elon Musk, Steve Wozniak, Andrew Yang, and many more. It should be noted that spinoff tools like Microsoft Copilot are being based on the latest models, with Copilot secretly launching with GPT-4 before that model was even announced.

• Following this trend, the next step for GPT-5 could be the ability to output video.
• The competition in the AI industry is heating up, with strong contenders emerging to challenge OpenAI’s dominance.
• He specializes in reporting on everything to do with AI and has appeared on BBC TV shows like BBC One Breakfast and on Radio 4 commenting on the latest trends in tech.
• Over the past year, the talk of the town has shifted from $10 billion compute clusters to $100 billion clusters to trillion-dollar clusters.
• Instead, reports suggest it could be rolled out initially for OpenAI’s key partners, such as Microsoft, to power services like Copilot.
• In a Reddit AMA with OpenAI’s Sam Altman, Kevin Weil, Srinivas Narayanan, and Mark Chen, Altman blamed compute scaling for the lack of newer AI models.

That’s a challenge for current models, which can produce errors with apparent confidence, known as hallucinations. But the progress is not great enough to justify the new model’s huge cost. Based on public and private estimates of various aspects of the training, a six-month training run could cost about half a billion dollars in computing costs alone. OpenAI has been working hard to refine the model, and the excitement around it shows how much people expect from this technology.

Great Barrier Reef has reached ‘catastrophic’ coral bleaching levels, study warns

Well, in Sam’s telling of the story, we could expect GPT-5 to handle five-hour tasks. This is congruent with Eric Schmidt’s argument that in the next five years, these machines will be able to undertake tasks that have 1,000 discrete steps. It is strikingly plausible that by 2027, models will be able to do the work of an AI researcher/engineer. Aschenbrenner goes into lots of detail about geopolitics, security, energy use, and more. Instead, I want to explore the implications of his argument about the trajectory of model capabilities. The report also details the various staffing problems OpenAI has been dealing with since Sam Altman was ousted and rehired in November 2023.

However, that changed by the end of 2023 following a long-drawn battle between CEO Sam Altman and the board over differences in opinion. Altman reportedly pushed for aggressive language model development, while the board had reservations about AI safety. The former eventually prevailed and the majority of the board opted to step down. Since then, Altman has spoken more candidly about OpenAI’s plans for ChatGPT-5 and the next generation language model. Yes, OpenAI and its CEO have confirmed that GPT-5 is in active development. The steady march of AI innovation means that OpenAI hasn’t stopped with GPT-4.

The internet is running out of training data

Any delays should actually be beneficial in the grander scheme of things. Whatever the case, it seems OpenAI is gearing up to launch its next big model soon. Anthropic recently upgraded the Claude 3.5 Sonnet model which gets even better at coding and other tasks. Amid many researchers and executives leaving OpenAI, the company is in a tight spot to keep up with the momentum. The key thing that’s going on now is we’re moving very quickly through the capability ladder steps.

This would help support tasks in mathematics, science, and other areas that require more accuracy and logical reasoning. Sam Altman, the CEO of OpenAI, has an exciting vision for the future of AI. He sees AI evolving from being just digital assistants to becoming highly capable colleagues who can work alongside us, enhancing our productivity and creativity. This vision is not just about making tasks easier; it’s about creating a new kind of partnership between humans and AI.

The problem is that it’s too slow to properly interpret multiple images quickly enough, meaning that video input is currently out of the question. It would be great if OpenAI made headway in this area, allowing GPT-5 to truly take video inputs into account. I’m hopeful, particularly as the company has been making decent strides in video AI in general, particularly with the reveal of Sora. Google is also working on video input with Gemini 1.5, and it’s looking promising. Alternatively, the power demands of GPT-5 could see the end of Microsoft and OpenAI’s partnership, leaving the Copilot+ program without even a basic chatbot. If GPT-5 is 100 times more powerful than GPT-4, we could get AI that is far more reliable.

Whatever the case, the figure implies OpenAI made big improvements to ChatGPT, and that they might be available soon — including the GPT-5 upgrade everyone is waiting for. If you’re using Microsoft Copilot Pro or Google Gemini Advanced, you’re probably aware of the tool integrations that you get with either service. Copilot Pro has full Microsoft 365 integration, and Gemini Advanced has full Google Suite integration. Those are pretty big advantages to have over competitors, to the point that Copilot Pro is simply a better purchase for most people over ChatGPT Plus.

According to the report, OpenAI has conducted two large training sessions for GPT-5, where each session went on for months and took massive amounts of data. However, the company reportedly faced unexpected issues each time which stopped it from getting the desired results. Notably, a six-month long training period for GPT-5 is said to cost the company around $500,000,000 (roughly Rs. 4,260 crores). The Wall Street Journal reported that the GPT-5 project, which is also codenamed Orion, is currently running behind schedule, and there is no clarity on when it might work, if at all. To overcome these challenges, researchers are exploring “test-time compute,” a technique that enhances existing AI models during the so-called “inference” phase, or when the model is being used. For example, instead of immediately choosing a single answer, a model could generate and evaluate multiple possibilities in real-time, ultimately choosing the best path forward.

As the field continues to evolve, these technical and resource obstacles are shaping both the pace and direction of AI advancement. The lessons learned during this process will likely influence AI development methodologies and expectations well into the future. The company’s strategic pivot toward developing advanced reasoning models represents a fundamental shift in approach. These new models focus on sustained critical thinking and problem-solving capabilities, requiring less specialized training data but introducing new layers of developmental complexity. This reorientation signals a broader evolution in how AI systems are conceived and developed.

Even the likes of Samsung’s chip division expect next-gen models like GPT-5 to launch soon, and they’re trying to estimate the requirements of next-gen chatbots. Perhaps the most interesting comment from Altman was about the future of AGI – artificial general intelligence. Seen by many as the ‘real’ AI, this is an artificial intelligence model that could rival or even exceed human intelligence. Altman has previously declared that we could have AGI within “a few thousand days”.

Hugging Face Introduces Compact AI Models for Everyday Devices

Each company is racing to develop smarter, more versatile, and reliable AI systems. Its reasoning abilities will allow it to provide insights into potential outcomes or suggest strategies based on historical data. Teachers could use GPT-5 to create customized lesson plans for different learning styles.

It is a more capable model that will eventually come with 400 billion parameters compared to a maximum of 70 billion for its predecessor Llama-2. In machine learning, a parameter is a term that represents a variable in the AI system that can be adjusted during the training process, in order to improve its ability to make accurate predictions. It will feature a higher level of emotional intelligence, allowing for more empathic interactions with users. This could be useful in a range of settings, including customer service. GPT-5 will also display a significant improvement in the accuracy of how it searches for and retrieves information, making it a more reliable source for learning.

While GPT-5 is likely to automate many tasks, it is not expected to replace all jobs. Instead, it will assist professionals in areas like writing, coding, customer service, and more, allowing them to focus on higher-level tasks. While GPT-5 is expected to expand its multimodal capabilities, OpenAI has not confirmed whether it will include advanced image or video generation. “Hallucinations” refer to incorrect or fabricated responses from the AI. GPT-5 is expected to significantly lower the occurrence of these errors by refining the model’s architecture and using better training data.

It might process and analyze such media, but full creative generation like current specialized AI models may not yet be implemented. OpenAI typically releases its newest models behind a paywall, reserving free access for older versions or limited features. Doctors could use GPT-5 for quick access to medical research and case studies. Its advanced reasoning will allow it to suggest treatment options based on patient symptoms and medical history. A Step Closer to AGIWhile the world eagerly awaits the launch of GPT-5, reports indicate that the AI model is likely to arrive no sooner than early 2025.

OpenAI’s GPT-5 reportedly falling short of expectations – TechCrunch

OpenAI’s GPT-5 reportedly falling short of expectations.

Posted: Sat, 21 Dec 2024 08:00:00 GMT [source]

This slowdown suggested that larger training runs would not only demand more time but also incur significantly higher expenses. Considering that most strawberries in that image are ripe, Altman might be telling the world that Strawberry could launch soon. The “summer” detail can indicate a summer release for this ChatGPT upgrade. Imagine a scenario where GPT-4 is integrated into a diagnostic system for analyzing patient symptoms and medical reports. A hallucination could lead the AI to confidently provide an incorrect diagnosis or recommend a potentially dangerous course of treatment based on imagined facts and false logic. The consequences of such an error in the medical field could be catastrophic.

• Instead, it will assist professionals in areas like writing, coding, customer service, and more, allowing them to focus on higher-level tasks.
• When she isn’t writing about the latest laptops and AI software, Madeline likes to throw herself into the ocean as a PADI scuba diving instructor and underwater photography enthusiast.
• Sign up for the most interesting tech & entertainment news out there.
• Even though GPT-5 may not be there yet, it will give us a glimpse, and if you ask me, that’s something to look forward to.

For his part, OpenAI CEO Sam Altman argues that AGI could be achieved within the next half-decade. Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards. For the finale of its 12 Days of OpenAI livestream event, CEO Sam Altman revealed its next foundation model, and successor to the recently announced o1 family of reasoning AIs, dubbed o3 and 03-mini. Arthur has been a tech journalist ever since 2013, having written for multiple sites.

Despite access to vast quantities of internet data, the model struggled to achieve the sophisticated understanding and reasoning capabilities that OpenAI had envisioned. This revelation highlighted a critical distinction between data quantity and quality in AI development. OpenAI might use Strawberry to generate more high-quality data training sets for Orion.