In 2017, some Google researchers made Transformers that are the sole reason why most of the AI products work nowadays. Even ChatGPT uses transformers to chat with the user. When ChatGPT, an LLM (Large Language Model), was released last year in November, everyone predicted that it’s going to start an era of AI models. Bill Gates also agreed to that statement saying that ChatGPT is the first step to the age of AI. Now that many LLMs are in use, researchers are aiming for Artificial General Intelligence (AGI). The purpose of AGI will be to create an AI model that will be as intelligent as humans.
Even though behind every AI are transformers that are powering them, Google researchers say that the transformers cannot power AGI as AI hasn’t reached the point where it could compete with human abstractions, thought processing and predictions. For instance, ChatGPT only responds to the prompts that users ask it to. It cannot make anything of its own. Even the free version of ChatGPT cannot tell about anything that has happened after 2022. These kinds of instances are making researchers think whether the dream to make human-like AI is even possible or not. Researchers are also trying out different other technologies that can work instead of transformers.
What we can expect in the future is a better AI model than ChatGPT. Albert Gu, assistant professor at the machine-learning department of Carnegie Mellon and Tri Dao who is a chief assistant at TogetherAI, submitted research to ArXiv which highlights a model called Mamba. Mamba is an SSM (State Space Model) that works even better than Transformers. There isn’t much difference between SSM and LLM. SSM also replies to commands like LLM but it uses more mathematical states. Dao also wrote on X that Mamba can reply 5 times faster than transformers when given a prompt.
If tests like Mamba keep on happening, there can soon be an AI model that can compete with human intelligence. ChatGPT works well but it is nowhere near human-like intelligence. We will soon get a model that will be able to generate responses similar to humans.
Read next: ChatGPT Gave the Wrong Answers for Nearly 74% of Medical Questions
Even though behind every AI are transformers that are powering them, Google researchers say that the transformers cannot power AGI as AI hasn’t reached the point where it could compete with human abstractions, thought processing and predictions. For instance, ChatGPT only responds to the prompts that users ask it to. It cannot make anything of its own. Even the free version of ChatGPT cannot tell about anything that has happened after 2022. These kinds of instances are making researchers think whether the dream to make human-like AI is even possible or not. Researchers are also trying out different other technologies that can work instead of transformers.
What we can expect in the future is a better AI model than ChatGPT. Albert Gu, assistant professor at the machine-learning department of Carnegie Mellon and Tri Dao who is a chief assistant at TogetherAI, submitted research to ArXiv which highlights a model called Mamba. Mamba is an SSM (State Space Model) that works even better than Transformers. There isn’t much difference between SSM and LLM. SSM also replies to commands like LLM but it uses more mathematical states. Dao also wrote on X that Mamba can reply 5 times faster than transformers when given a prompt.
If tests like Mamba keep on happening, there can soon be an AI model that can compete with human intelligence. ChatGPT works well but it is nowhere near human-like intelligence. We will soon get a model that will be able to generate responses similar to humans.
Read next: ChatGPT Gave the Wrong Answers for Nearly 74% of Medical Questions
