ChatGPT: enthusiasm or concern?

Introduction

ChatGPT has dominated headlines as an advanced conversational bot developed by OpenAI, still in prototype stage. It aims to answer questions across diverse domains including law, programming and general knowledge.

Logic tests

Initial experiments with logic puzzles prove convincing:

• The black and white balls enigma: ChatGPT partially solves this logical deduction problem involving three people wearing coloured balls, requiring inference based on observations.
• The Monty Hall variant: ChatGPT correctly provides probability calculations and explains how the variant differs from the original problem.

These first tests prove genuinely impressive.

JavaScript and code

ChatGPT demonstrates strong programming capabilities:

• Generates correct JavaScript email validation functions
• Rewrites the same function in COBOL and Assembler without difficulty
• When bugs are identified (missing null checks), ChatGPT acknowledges errors and immediately corrects the code

ChatGPT’s fundamental limitations

1. An extraordinarily large model

GPT-3 contains 175 billion parameters, requiring approximately 700 GB of RAM just to load. Training consumed enormous resources: a single V100 GPU would need 288 years; Microsoft’s Azure infrastructure parallelised this over several months, at an estimated cost of $4.6 million in computational power alone. The carbon footprint is substantial.

2. Already obsolete upon creation

GPT-3’s training data ends in 2021. Current events and recent developments are not reflected. When asked about Twitter’s current CEO, ChatGPT names Jack Dorsey — outdated information. Real-time updates would require prohibitively expensive retraining.

3. Hallucinations and false information

ChatGPT exhibits “ultracrepidarianisme” — confidently stating incorrect information. Despite OpenAI’s data quality precautions, the system occasionally produces false information, potentially spreading misinformation if treated as an authoritative source.

The competitive landscape

Google is not directly threatened. ChatGPT cannot replace Google Search’s real-time indexing due to prohibitive costs. Google develops LaMDA and the PATHWAYS architecture, which updates models continuously as new data arrives, enabling dynamic adaptation across multiple coexisting models.

Agora Software’s distinct approach

Agora does not seek to compete with ChatGPT’s omniscience. The company develops targeted conversational interfaces through GAEL, its proprietary Natural Language Processing system.

Key differences:

• Decentralised deployment: GAEL distributes across dedicated containers, not centralised servers
• Learning classes: Clients create proprietary training data through supervised learning classes via Agora’s admin console
• Lightweight: Training completes in under one minute; models run on Raspberry Pi with a minimal carbon footprint
• Scope limitation: GAEL answers questions only within connected applications’ functional boundaries
• Security and privacy: Dedicated, fully secured environments minimise inter-project data sharing
• Omnichannel and multilingual: Integrates with Teams, WhatsApp, Messenger, SMS and other platforms where users already communicate

Conclusion

ChatGPT represents a cutting-edge technical achievement with spectacular results demonstrating remarkable progress. However, it cannot truly replace humans — it learns only from text, missing the contextual nuance and non-verbal elements essential to complete understanding.

These systems will increase productivity and assist diagnosis across sectors, but they offer surface-level understanding. As Jacoby Browning and Yann LeCun note: these systems are “destined for superficial understanding that will never approach complete human thought”.

ChatGPT is a powerful tool — like a hammer, useful for building or dangerous if misused. The technology will change lives without replacing human judgement.

This article was not written by ChatGPT.