A naukri.com initiative
Tech Radar
1w
213
Image Credit: Tech Radar
ChatGPT is getting smarter, but its hallucinations are spiraling
- OpenAI's latest AI models, GPT o3 and o4-mini, exhibit more hallucinations compared to their predecessors, raising concerns about AI reliability.
- The increased complexity of GPT o3 and o4-mini may lead to more confident inaccuracies and errors in their responses.
- These models are designed to mimic human logic and reasoning, but the higher error rates suggest issues with accuracy.
- During benchmark tests, GPT o3 and o4-mini showed hallucinations in 33% and 48% of tasks, respectively, emphasizing the rise in inaccuracies.
- When tested on general knowledge questions, the hallucination rates increased to 51% for o3 and 79% for o4-mini.
- The AI research community theorizes that reasoning models like GPT o3 and o4-mini have more chances to go astray when attempting complex evaluations.
- The line between theory and fabricated fact becomes blurry for AI as it speculates on possibilities, often leading to hallucinations.
- Increased hallucinations in AI models contradict the goal of providing reliable assistance, raising concerns about their real-world applications in various sectors.
- Sophisticated AI systems, despite being impressive, pose risks when hallucinations go unnoticed and could lead to errors in critical environments.
- It is advised to approach responses from AI models like ChatGPT with caution due to the prevalence of hallucinations and inaccuracies.
Read Full Article
12 Likes
For uninterrupted reading, download the app