Skip to main content

What are LLMs (Large Language Models) ? - Science Feed

Image
By

LLMs are large language models which are kind of deep learning models where it predicts single token or sequence of token that are relatable to each other. A token can be a word, a subword or a character. 

These are built on a type of neural network architecture called transformers. Currently, ChatGPT, Gemini, Perplexity etc. all are based on it. These transformers are very large in size because of parameters billions in size. It uses encoders and decoders. Transformers use self-attention techniques where self is for input sequence and self-attention weighs the relation between tokens in the input sequence. 

Training and Fine-tuning are other factors that contribute to better working of LLMs. NLPs (Natural Language Processing) are also important for the LLMs. 

Issues with LLMs

LLMs have issue of hallucinating, biases and using a lot of computational and energy resources. AI companies are massively using these which are creating shortage for common people. 

Popular LLMs

GPT-5, GPT-4o, Deepseek -v3.2, Gemini 3.0, Claude Opus 4.5 and many more. These are also coming up with Agentic AI. 

Follow and Subscribe our blog more content. 

Popular posts from this blog

How does a NPU, Neural Processing Unit, works? - Science Feed

By
A NPU (Neural Processing Unit), also knows as AI accelerator, is a computing unit specially designed for the running artificial intelligence and machine learning models efficiently and effectively on a single motherboard with other computing units.  There has been significant rise in the use of NPUs in personal devices like smartphones, laptops, PCs etc. to do personalized recommendations and offline AI processing.  NPUs are capable of parallel processing, low precision arithmetic, hardware acceleration.  These have silicon architecture to kind of simulate the human brain working like human where instruction set of AI efficiently ran. NPUs are not for computing traditional problems as it lacks important physical capabilities. Because of these, AI systems are able to learn faster without wasting time and energy. The parallel computation feature give wings to matrix operations ease, activation functions of neural networks, dot product.  NPUs have got dedicated compute ...
Read more

What is Dark Energy? - Science Feed

By
There are different forms of energy. But, there has been speculation of an hidden form of energy that has been kind of hiding from us which constitutes a large part of our universe.  The universe has been found to be expanding faster than the speed of light. The traditional form of energy that we have found has been to be not able to provide that much of push. So, there must be something. This type of energy is called Dark Energy. Dark Energy is that form of energy that is not understood by us but it is fundamental in nature and drives the process of the universe. In Theory of General Relativity, there is concept of cosmological constant which says there is a negative of gravitation or empty space that is all over the interstellar space.  The researchers of DES and others have found that the universe is expanding not alone of known form of energy but some mysterious form of it. This is also related to Big Bang. Dark energy is told to be constituent of 68 to 70% of total univer...
Read more

Quantum Computer at Google solves real world problem 13,000x faster than current Supercomputers - Science Feed

By
On 22 October 2025, Google announced that they have solved a real world problem using their Quantum Echoes algorithm that did the work 13,000 times faster than current supercomputers.  This was the first time it has ever done where they used out of order time correlator OOTC algorithm. This  algorithm is verifiable since it can be ran again and again to get consistent result. They have released a scientific paper over this where they were able to do study of molecular structure using their latest Willow quantum computing chip. This chip used 105 qubit arrays to simulate the molecular behaviour. In this task, they disturbed one qubit from the simulated structure which lead to the formation of butterfly effect inside kind of like wave. This was first done in forward operations. After this, they ran the same operation but from the reverse side.  Both the sides waves start interacting with each other and give useful insights about the quantum nature.  According to Google...
Read more