A naukri.com initiative
Analyticsindiamag
1M
188
Image Credit: Analyticsindiamag
OpenAI is So Doomed if Inference Time Scaling for o1 Fails
- OpenAI's recent progress from GPT-4 to Orion has slowed according to a report. However, Team OpenAI researchers have clarified that the article inaccurately portrays the progress of OpenAI's upcoming models or rather misleading. There are two key dimensions of scaling for models like the o1series training time and inference time. Scaling laws focusing on pre-training larger models for longer are still relevant, but there's now another important factor. The introduction of this second scaling dimension is set to unlock new capabilities which offers a more dynamic and thoughtful approach to solving problems.
- Orion is partially trained on AI-generated data or synthetic data produced by other OpenAI models, including GPT-4 and recently released reasoning models. The information report stated that Orion outperforms previous models. While Orion's performance improvement is less dramatic than the leap from GPT-3 to GPT-4.
- OpenAI chief Sam Atman suggested that AGI could emerge as soon as 2025. OpenAI has yet to release o1 fully. Many believe that o1 could be the first commercial application of System 2 thinking. In EpochAI's FrontierMath benchmark, it was revealed that only 2% of the hardest and unpublished math problems were successfully solved by LLMs.
- While others remain uncertain, OpenAI chief Sam Altman is confident that artificial general intelligence (AGI) is closer than many think. In a recent interview with Y Combinator's Garry Tan, Altman suggested that AGI could emerge as soon as 2025.
- Apple recently published paper titled 'Understanding the Limitations of Mathematical Reasoning in Large Language Models', which said that the current LLMs can't reason. The researchers introduced GSM-Symbolic, a new tool for testing mathematical reasoning within LLMs because GSM8K was not accurate enough and, thus, not reliable for testing the reasoning abilities of LLMs.
- OpenAI released o1-mini and o1-preview, mentioned in their blog post that o1's performance consistently improves with more reinforcement learning and with more time spent thinking. NVIDA CEO Jensen Huang recently said that the company is currently facing in computing inference time scaling which involves generating tokens at incredibly low latency.
- OpenAI senior researcher Jason Wei explained that the traditional reasoning used by AI models like GPT was more of a mimicry than a true "thinking" process. In this paradigm, the chain of thought reflects more of an internal reasoning process, similar to how humans think. The model engages in an "inner monologue" or "stream of consciousness," where it actively considers and evaluates options, a process that is dynamic and thoughtful.
- OpenAI researchers were quick to correct the narrative asserting that the article inaccurately portrays the progress of OpenAI's upcoming models-or rather misleading. Gary Marcus recently remarked that the LLM improvements have hit a wall.
- It appears that OpenAI has exhausted all available data for pre-training the model and is now exploring new methods to improve o1. Regarding inference time scaling, OpenAI said, "The constraints on scaling this approach differ substantially from those of LLM pretraining, and we are continuing to investigate them."
- Google DeepMind recently published a paper titled 'Chain of Thought Empowers Transformers to Solve Inherently Serial Problems'. While sharing his research, Andrej Karpathy suggested that next-token prediction frameworks could become a universal tool for solving a wide range of problems, far beyond just alone text or language.
Read Full Article
11 Likes
For uninterrupted reading, download the app