Traditional Excel sheet vs Excel sheet with AI
Working with an Excel sheet without ChatGPT / AI is not recommended, and also the traditional approaches, by using them, we can excel our work/tasks completion within less time.
eg: How to integrate AI with an Excel sheet?
Overview: AI (Artificial Intelligence) uses Microsoft 365 Copilot in Excel to help us by speed up tasks regarding spreadsheets, that are “adding formulas, columns, formatting tables, and data analysis”.
First, learn the capabilities of AI:
* Day-to-day tasks can be performed efficiently
* High Productivity can be achieved
* Help to automate the repetitive tasks.
* will help to do perfect Data Analysis and Data transformation.
* No Manual work or a minimal amount of work will save a lot of time.
eg:
1. AI can help to identify the best quality Apples from a bunch of Apples.
2. Sales managers can use the AI to get the Best Analysis report from Quarterly sales by quickly.
Different AI’s
* GPT-5
* DeepSeek
* Gemini
* Claude
* Grok
* Kimi
Background of GPT-5
1. What is GPT?
* GPT stands for Generative Pre-trained Transformer.
* It is an advanced AI language model created by OpenAI.
* GPT can understand and generate human-like text based on the input it receives.
—————————————————————————————-
2. Core Ideas Behind GPT
2.1 Transformer Architecture
* Introduced in 2017 by Google in the paper “Attention is All You Need.“
* Uses self-attention to process words in context, making it excellent at understanding
long sentences.
2.2 Pre-training + Fine-tuning
* Pre-training: Model learns from massive text data (books, websites, articles).
* Fine-tuning: Adjusted with specific tasks or safety guidelines.
2.1 Generative Power
* GPT doesn’t just answer questions-it can
summarize, translate, code, write stories, chat, and reason.
—————————————————————————————-
3. Evolution of GPT
* GPT (2018): Proof of concept, 117M parameters.
* GPT-2 (2019): 1.5B parameters, surprised everyone with its fluency.
* GPT-3 (2020): 175B parameters, huge leap in quality and reasoning.
* GPT-4 (2023): More advanced reasoning, multimodal (text + images).
* GPT-5 (2025): Current generation, with deeper reasoning, better context
retention, and improved accuracy.
—————————————————————————————-
4. Capabilities
* Answering questions
* Writing & summarizing
* Generating code
* Translating languages
* Creating educational, marketing, and technical content.
* Assisting in research & creativity.
—————————————————————————————-
5. Limitations
* Doesn’t “know” facts beyond its training or updates.
* Can sometimes make mistakes (hallucinations).
* Needs guardrails for safe and ethical use.
—————————————————————————————-
In short/summary
GPT AI is like a highly advanced “text prediction engine” that can simulate conversation, reasoning, and creativity-powered by deep learning and huge datasets.
—————————————————————————————-
Background of DeepSeek & Overview
* Introduction
DeepSeek is a cutting-edge artificial intelligence (AI) company and research initiative focused on building large-scale foundation models and intelligent reasoning systems. Known for its innovations in natural language processing (NLP), multimodal AI, and efficient training methods, DeepSeek is recognized as one of the fastest-growing players in the global AI landscape.
—————————————————————————————-
* Founding & Vision
* Founded:2023
* Origin: China
* Mission: To create AI systems that combine powerful reasoning abilities with efficient compute usage, making advanced AI accessible for businesses, developers, and researchers.
* Core Philosophy: Unlike traditional “brute-force scaling,” DeepSeek emphasizes optimization, sparsity, and efficiency, ensuring high performance at lower infrastructure costs.
—————————————————————————————-
* Key Developments
1. DeepSeek-V1 (2023):
* Early large language model.
* Focused on multilingual NLP tasks.
2. DeepSeek-Coder (2024):
* Specialized in programming and code generation.
* Trained with massive amounts of open-source code.
* Outperformed several industry leaders in software engineering benchmarks.
3. DeepSeek-LLM (2024):
* A general-purpose large language model.
* Showed competitive performance against GPT and Claude on reasoning and creative tasks.
4. DeepSeek-V2 (2025):
* Introduced Mixture of Experts (MoE) architecture, making it highly computationally
efficient.
* Balanced performance, cost, and scalability.
* Positioned as a strong competitor to OpenAI’s GPT-4 and Anthropic’s Claude.
—————————————————————————————-
* Technology & Innovations
* Mixture of Experts (MoE): Uses specialized subnetworks (experts) for different tasks,
Reducing unnecessary compute use.
* Efficient Training: Optimized data pipelines and distributed training techniques.
* Multilingual Focus: Strong capabilities in Chinese, English, and cross-lingual tasks.
* Open Collaboration: Provides open-source models and encourages research
contributions.
—————————————————————————————-
* Impact & Position in AI Industry
* DeepSeek is often seen as China’s response to OpenAI and Anthropic.
* Attracted attention for cost-efficient training – proving that state-of-the-art AI does not
always require massive spending.
* Widely adopted in enterprise solutions, research labs, and developer communities.
* Sparked global AI competition, encouraging more innovation in efficiency-focused AI
models.
—————————————————————————————-
* Future Directions
* Expansion into multimodal AI (text, image, video, and speech).
* Enhancing reasoning capabilities to push closer toward artificial general intelligence (AGI).
* Building developer ecosystems with APIs and enterprise integrations.
—————————————————————————————-
* In summary:
DeepSeek is a next-gen AI company making waves through its efficient training methods, coding-specialized models, and competitive LLMs. By focusing on intelligence + efficiency, it has positioned itself as a strong rival to Western AI labs while pushing the boundaries of what’s possible with less compute.
—————————————————————————————-
DeepSeek vs. OpenAI (GPT)
A Comparative Background of Philosophy, Technology, and Impact.
| Founding Vision & Philosophy | |||
|---|---|---|---|
| S. No | Vision/Philosophy Type | OpenAI (2015, USA) | DeepSeek (2023, China) |
| 1 | Goal | * Started as a non-profit aiming to ensure that AGI benefits all of humanity. | * Founded with goal of building efficient, reasoning-oriented AI. |
| 2 | Strategy | * Train ever-larger general-purpose models with huge compute budgets | * Focus on cost-efficiency, optimization, and sparsity rather than brute force scaling |
| 3 | Belief | * Scale is intelligence -> More parameters + more data = smarter AI. | * Intelligence = Efficiency + Specialization, not just raw size. |
| Technology Approach | ||
|---|---|---|
| S. No | OpenAI (2015, USA) | DeepSeek (2023, China) |
| 1 | * Dense Transformer models: Every parameter active for every input | * Mixture of Experts (MoE): Only a subset of parameters are user per task, making training & inference cheaper. |
| 2 | * Trained with massive datasets (hundreds of billions of tokens) | * Heavy emphasis on efficiency (using fewer GPUs and less energy |
| 3 | * Relies heavily on reinforcement learning from human feedback (RLHF) for alignment. | * Specialized models (e.g.,DeepSeek-Coder) for coding and software engineering, not just general text. |
| Model Scale & Efficiency | ||
|---|---|---|
| S. No | OpenAI (2015, USA) | DeepSeek (2023, China) |
| 1 | * GPT-3 (175B parameters), GPT-4 (multi-expert, but dense in deployment). | * DeepSeek-V2 uses MoE architecture activating only ~10-20% of parameters at a time |
| 2 | * Trained with budgets in the hundreds of millions of dollars | * Delivers GPT-4-class performance at a franction of the cost |
| 3 | * Often criticized for being closed-source and costly | * More open in releasing models for research and enterprise use. |
| Global Positioning | ||
|---|---|---|
| S. No | OpenAI (2015, USA) | DeepSeek (2023, China) |
| 1 | * Leader in Western AI. | * Seen as China’s answer to OpenAI. |
| 2 | * Integrated into Microsoft ecosystem (Azure, Copilot, Office, GitHub) | * Strong adoption in Chinese enterprises and research |
| 3 | * Strong presence in consumer apps (ChatGPT). | * Positioning itself as a global low-cost competitor for businesses and developers |
| Strengths & Weaknesses | ||
|---|---|---|
| S. No | OpenAI (2015, USA) | DeepSeek (2023, China) |
| 1 |
* OpenAI Strengths: * World-class research & infrastructure. * Best-in-class alignment and safety. * Deep ecosystem integration with Microsoft. |
* DeepSeek Strengths: * Efficient & affordable AI training/inference. * Specialized coding model (DeepSeek-Coder) that outperforms GPT on benchmarks. * More accessible through open-source releases. |
| 2 |
* OpenAI Weaknesses:
* Very expensive models. * Closed-source -> limited transparency. * Slower to release specialized domain models. |
* DeepSeek Weaknesses * Alignment & safety research still catching up. * Less global brand recognition compared to OpenAI. * Heavier focus on Chinese/Asian markets ->slower Western adoption. |
| Future Outlook | ||
|---|---|---|
| S. No | OpenAI (2015, USA) | DeepSeek (2023, China) |
| 1 |
* OpenAI: * Pushing toward AGI via massive scale + alignment. * Strengthening ties with Microsoft enterprise products. * Likely to keep models closed-source but API accessible. |
* DeepSeek: * Expanding into multimodal AI (text, image, video, speech). * Competing by offering cost-effective GPT-4-level performance. * Could democratize AI by making powerful models cheaper & open. |
In summary:
* OpenAI = scale-first, closed, premium AI (high cost, deeply integrated).
* DeepSeek = efficiency-first, semi-open, affordable AI (cost-effective, specialized).
* Together, they represent two different philosophies of building the future of AI
* Brute-force scale (Open AI) vs. Optimization + specialization (DeepSeek)
