Comparing AI LLMs

ChatGPT, Claud, Perplexity, and Gemini Compared.

When comparing large language models (LLMs) and AI tools like ChatGPT, Claude, Perplexity AI, and Gemini, it's essential to consider their core functionalities, strengths, limitations, and ideal use cases. While all are powerful in generating human-like text and engaging in conversations, they often have distinct focuses.

Please note that CBS covers the costs for all of these apps; if you haven't logged in and tried one yet, most, if not all, can be accessed using Sign In with Google,

Here's a breakdown:

ChatGPT (OpenAI)

Core Functionality: A versatile conversational AI that can generate human-like text, answer questions, summarize, translate, and assist with creative writing and coding.¹ It's built on OpenAI's GPT (Generative Pre-trained Transformer) models.²

Key Features:

• Natural Language Processing (NLP): Excellent at understanding and generating natural language.³

• Contextual Understanding: Can maintain context over extended conversations, leading to more coherent and relevant responses.⁴

• Content Generation: Capable of producing various forms of text, from articles and stories to emails and code snippets.⁵

• Code Assistance: Can generate, debug, and refactor code in multiple programming languages.⁶

• Multimodal (with GPT-4o): Latest versions (like GPT-4o) can process and generate content across text, audio, and images.⁷

• Customization: Enterprise versions offer customization options and data analysis.⁸

• Integration: Widely adopted API for integration into various applications.⁹

• Web Search (for paid versions): Paid versions can access real-time information from the web.¹⁰

Applications:

• Customer support chatbots

• Content creation for marketing, blogs, and media¹¹

• Code generation and debugging

• Interactive learning tools

• Language translation

• Brainstorming and problem-solving¹²

• Data entry and management (e.g., structuring raw input)¹³

Limitations:

• Factual Accuracy: Can sometimes "hallucinate" or provide inaccurate/outdated information, especially the free GPT-3.5 model with a knowledge cutoff (September 2021 for the free version, April 2024 for GPT-4). Human review is often necessary.

• Bias: May exhibit biases present in its training data.¹⁴

• Common Sense Issues: Can struggle with real-world common sense and logical reasoning.¹⁵

• Difficulty with Long-form Structured Content: Can sometimes repeat points or struggle with complex formatting without explicit guidance.¹⁶

• Lack of Real-time Web Access (for free version): The free version cannot access current events or real-time data.

• Usage Limits: Paid versions have usage limits depending on the subscription tier.¹⁷

Claude (Anthropic)

Core Functionality: Developed by Anthropic, Claude prioritizes safety, ethical AI use, and responsible development.¹⁸ It's designed to be helpful, harmless, and honest, adhering to "Constitutional AI" principles.

Key Features:

• Safety and Alignment: Strong emphasis on ethical AI, aiming to provide reliable and less biased outputs.¹⁹

• Large Context Window: Known for its significantly larger context window (e.g., 200,000 tokens for Claude 3 Sonnet), allowing it to process and remember much more information in a single conversation (equivalent to hundreds of pages of text).²⁰

• Multimodal (with Claude 3): Newer versions (like Claude 3) can process text, image, and audio inputs.²¹

• Document Analysis: Excellent for summarizing, analyzing, and extracting information from long documents (e.g., PDFs, Word documents, legal contracts).²²

• Long-form Text Generation: Capable of generating detailed and coherent long-form text.

• Code Generation and Review: Can generate and review code snippets.²³

Applications:

• Sensitive customer service environments

• Internal knowledge management requiring accuracy and safety

• Applications demanding transparent AI behavior

• Legal and technical document review and summarization

• Research and question-answering for large texts

• Ethical discussions and content moderation

Limitations:

• Limited Image Generation: While multimodal for understanding, its image generation capabilities are more limited compared to dedicated image generation models or other LLMs.²⁴

• No Internet Browse (in some versions): Some versions may lack direct internet Browse capabilities, though newer models might integrate this.

• Can Be Overly Cautious: Its safety guardrails might lead to less creative or more conservative responses in certain situations.

• Usage Limits: Like other models, it has message limits, especially for free users, which can be affected by message length and complexity.²⁵

• Regional Restrictions: Claude's services may not be accessible in all regions (e.g., Mainland China, EU).

Perplexity AI

Core Functionality: Primarily a question-answering and search assistant tool that integrates real-time web data to provide concise, sourced answers.²⁶ It acts as an AI-powered search engine, emphasizing transparency and citation.

Key Features:

• Real-time Web Search: Its main differentiator is the ability to search the internet in real-time, providing up-to-date information.

• Citation and Source Tracking: Every answer includes footnotes linking to the original sources, allowing users to easily verify information.

• Concise and Sourced Answers: Excels at delivering direct, factual answers rather than extended conversational prose.

• Context Awareness: Understands the context of queries to provide relevant information.

• Copilot Mode: Offers an interactive mode for more guided research.

• File Analysis: Can upload and analyze files (PDFs, CSVs, images).²⁷

• Multi-model Access (Pro version): The Pro version can leverage multiple underlying LLMs (e.g., GPT-4, Claude 3).

Applications:

• Researching current events and factual information

• Verifying facts and data

• Summarizing industry reports and academic papers

• Getting quick, cited answers to specific questions

• Content creation requiring factual accuracy and sources

Limitations:

• Less Suited for Open-ended Conversations/Creative Content: Its focus on factual answers means it's not as strong for highly creative tasks or lengthy, open-ended discussions.

• Limited Conversational Flair: Responses can be more factual and less conversational or empathetic.

• Limited Problem-Solving/Reasoning: While good for information retrieval, it may have limited advanced problem-solving or deep reasoning abilities.²⁸

• Accuracy and Reliability Concerns: Still susceptible to inaccuracies or "hallucinations" from underlying LLMs, requiring user verification.²⁹

• Response Consistency: Can sometimes provide overly brief or inconsistent answers, and formatting can be difficult to enforce.

Gemini (Google)

Core Functionality: Google's multimodal AI model, designed to be highly capable across various modalities (text, code, audio, image, video). It's integrated within Google's ecosystem and aims for advanced reasoning.³⁰

Key Features:

• Multimodal: Can understand and operate across different types of information, including text, images, audio, and video.

• Advanced Reasoning: Designed for strong reasoning capabilities, including complex problem-solving.³¹

• Google Ecosystem Integration: Seamlessly integrates with Google products like Gmail, Docs, Drive, Maps, and Chrome.³²

• Deep Research (with paid plans): Can browse and analyze hundreds of websites in real-time to generate comprehensive research reports.³³

• Video and Image Generation: Can generate videos (with Veo) and images, transforming words into visual content.³⁴

• Large Context Window: Offers a substantial context window (e.g., 1 million tokens for Google AI Pro/Ultra), allowing it to process large amounts of text and code.

Applications:

• Generating content across text, image, and video

• Summarizing documents and emails within Google Workspace³⁵

• Researching complex topics with real-time web access

• Creative tasks involving multiple modalities (e.g., storytelling with visuals)³⁶

• Planning trips (integrating with Maps, Flights, Hotels)³⁷

• Coding assistance and debugging³⁸

• Educational assistance with detailed, cited responses (e.g., via OpenStax app)

Limitations:

• Hallucinations: Like other LLMs, it can generate factually incorrect or nonsensical outputs.³⁹

• Bias Amplification: Potential to amplify biases present in its training data.⁴⁰

• Language Quality (for non-English): While multilingual, performance might be less effective for some non-English languages or dialects that are underrepresented in training data.

• Domain Expertise: May lack the depth of knowledge for highly specialized or niche technical topics, potentially leading to superficial information.⁴¹

• Feature Maturity: As a newer offering in direct competition, some features might still be evolving or have regional limitations (e.g., Gemini in Chrome for US only).

• Usage Limits: Also has usage limits based on subscription tiers, which can be affected by prompt complexity and conversation length.⁴²

Comparative Summary

Feature/Aspect	ChatGPT (OpenAI)	Claude (Anthropic)	Perplexity AI	Gemini (Google)
Primary Focus	General conversational AI, content generation	Safe, ethical AI, long-form text/document analysis	AI-powered search engine, factual answers with sources	Multimodal AI, advanced reasoning, Google integration
Web Access	Yes (paid versions), No (free GPT-3.5)	Limited/No direct Browse (varies by version)	Yes, real-time web search with citations	Yes, real-time web search (Deep Research)
Multimodality	Yes (GPT-4o: text, image, audio)	Yes (Claude 3: text, image, audio)	Yes (file analysis: PDFs, images)	Yes (text, image, audio, video)
Context Window	Up to 128K tokens (GPT-4 Turbo/o)	Very large (up to 200K tokens for Claude 3 Sonnet)	Varies (focus on current query, less conversational memory)	Large (up to 1M tokens for Google AI Pro/Ultra)
Safety Emphasis	General safety, evolving	High, "Constitutional AI" principles	Transparent sourcing, but relies on underlying LLMs	High, responsible AI principles
Best For	General tasks, creative writing, coding, chatbots	Long document analysis, sensitive applications, ethical considerations	Factual research, quick answers with sources, current events	Multimodal creative tasks, research, Google ecosystem users
Limitations	Hallucinations, data cutoff (free), potential bias	Can be overly cautious, limited image generation, regional restrictions	Less creative/conversational, potential for LLM inaccuracies, sometimes brief responses	Hallucinations, potential bias, feature maturity/regional limits

Choosing the "best" AI depends entirely on your specific needs.

• If you need a versatile tool for general content creation, coding, and dynamic conversations, ChatGPT is a strong contender.

• For handling large documents, prioritizing safety, or working in sensitive environments, Claude excels.⁴³

• If your primary need is real-time, cited factual information, Perplexity AI is highly effective.

• For a comprehensive multimodal experience deeply integrated with Google services and advanced reasoning, Gemini offers a compelling solution.⁴⁴