Can AI even think? This question has been troubling scientists for a long time. When we talk to an AI chatbot, how does it respond? Does it really think or just respond by connecting words? An AI company named Anthropic has made a big breakthrough in solving this mystery. They have developed a unique ‘AI microscope’ that will now help them understand what is going on inside an AI model. This discovery will not only make AI transparent but will also make it more secure and reliable. AI Revolution is taking the entire world by storm at once.
AI startup Anthropic has created a new technology that can be used to understand how large language models (LLMs) actually work. This is the same company that created the cloud chatbot. Anthropic says that they have developed an ‘AI microscope’ that will help find out how these AI models think. This technology is inspired by neuroscience (brain science) and can be used to understand how language models process information and what is going on in their minds. According to the company, if it is understood how AI models like the cloud work, they can be made more accurate and reliable. This will also ensure that they work exactly as we expect them to.
AI microscopy will help in understanding LLM.
Large language models (LLM) are often called ‘black boxes,’ as scientists have not yet fully understood how these AI models reach a specific answer without any fixed programming. Problems like AI hallucinations, fine-tuning, and jailbreaking still remain mysteries. But this new discovery of Anthropic can make the way AI models work more clear and understandable. The company said that they have published two new scientific research papers on ‘AI Biology.’ The first research explains how the cloud prepares answers to user questions, while the second research explains what happens inside the Cloud 3.5 Haiku model. With the help of this new technology, AI can be made more transparent, understandable, and reliable.
Reasoning process of AI model and special things discovered
Anthropic has created a new model named ‘Cross-Layer Transcoder’ (CLT). This model helps in understanding different parts of the neural network that works like the brain of AI. Researchers observed that when the cloud is asked to write poetry. It first chooses rhyming words and then molds them into sentences.
Another surprising thing that came to light was that the cloud can sometimes show that it is solving mathematical problems thoughtfully, when in reality it is not. This means that it can also create a wrong reasoning process. This discovery shows that the ‘chain of thought’ technology and reasoning AI models of companies like OpenAI may not be completely reliable.
This means that AI does not always think correctly, but sometimes it can show that it is giving thoughtful answers when in reality it is not.
Limitations of the AI Microscope and the Way Forward
Anthropic also admitted that there are some shortcomings in this research. The company said that their technology is just an estimate and cannot explain the entire thinking process of complex AI models like clouds. Apart from this, there may be some neurons (small parts of AI’s brain) that are not caught by this technology but can still affect the answers of the cloud.
According to Anthropic, even when AI is given small and easy questions, this technology is able to understand the calculation of only some parts. It is possible that some things may not show the thinking process of the real model correctly. But despite this, this new technology can help in making AI more safe and reliable.
Anthropic PBC is an American company that works on artificial intelligence (AI). It was started in 2021. This company builds large language models (LLMs) that can answer questions like humans. Its AI model called Cloud competes with OpenAI’s ChatGPT and Google’s Gemini. Anthropic’s goal is to create safe and reliable AI. This company builds AI models that are transparent and can be used by humans in a safe way.
Nowadays AI is progressing rapidly, but concerns are also increasing about its dangers. Many experts believe that if AI is not handled properly, it can cause problems in the future. For this reason, Anthropic is working on technology that makes AI ethical and reliable. The company’s research ensures that AI is not only fast and smart but also accurate and safe. Anthropic’s AI development focuses not only on new technology but also on making it safe and helpful for humans.
ALSO READ: Elon Musk sells X to xAI startup for $33 billion
Share
Rashmika Mandanna's Reminder to Embrace True Self
Vontimitta Brahmotsavam 2025: Here's All the Details
