Introduction
Artificial Intelligence (AI) tools are flooding the market, but what many people don’t realise is that many of these tools are powered by a few core, foundational models—large language models. Understanding these key players in the AI space can empower you to directly tap into their capabilities and fully utilise the AI landscape.
The Big Three
OpenAI’s Chat GPT (3.5 and 4.0)
First up is OpenAI’s Chat GPT, available in two versions: 3.5 and the more powerful 4.0. Both models aren’t connected to the internet, meaning you must provide a prompt or instruction manually. Companies can access these models through an API, acting as a surface layer while the model does the heavy lifting.
Microsoft’s Bing Chatbot
Bing offers three modes: Precise, Balanced, and Creative. Both Creative and Precise modes utilise GPT-4, the more powerful version. The advantage of Bing is that it’s connected to the internet and can read and work with documents, offering capabilities beyond those of Chat GPT.
Google’s Bard
Bard is Google’s contribution to foundational models. At present, Bard might not be as capable as the other two but is catching up fast. It’s freely accessible, connected to the internet, and its speed makes it an attractive option.
Using These Models
All three models are versatile, but their functionality can differ slightly depending on the prompts and tasks you’re interested in. They can perform creative tasks like writing poems, handle data, write analyses, and more. You need to learn how to refine your prompts to get the best out of them.
| Feature | ChatGPT | Bing | Bard |
|---|---|---|---|
| Powered By | OpenAI | Microsoft (uses GPT-4) | |
| Internet Connectivity | No (as of recording) | Yes | Yes |
| Free Access | Limited | Yes | Yes |
| Speed | Varies (3.5 is faster) | Varies | Generally Fast |
| Model Versions | GPT-3.5, GPT-4 | GPT-4 | Palm 1, Palm 2 |
| Creative Writing Abilities | Strong | Strong | Strong but different |
| Custom Prompts | Yes | Yes | Yes |
| Data Privacy Options | Chat history toggle | Not specified | Not specified |
| Multiple Modes | No | Precise, Balanced, Creative | Not specified |
| Accessibility in Countries | Limited | 169 Countries | Not specified |
| Can Create Images | No (as of recording) | Yes | No |
| Real-time Suggestions | No | Yes | No |
| Integration with Work Tools | API available | Coming Soon | Coming Soon |
Strengths and Weaknesses
Pros
- Good at ‘human tasks’ like writing, creative work, and data analysis.
- Versatile, as you can turn them into virtual teachers, editors, or office assistants with the right instructions.
Cons
- They can generate incorrect or misleading information.
- The output may not be consistent over time.
- Despite their capabilities, they aren’t infallible. Don’t expect Warren Buffett-level investment advice or a Marvin Gaye-like song from these models.
Other conversational tools
There are several other tools and services similar to the ones mentioned above to consider:
- Dialogflow – Developed by Google, widely used for creating conversational agents integrated into various applications, websites, and services.
- Amazon Lex – This service by Amazon Web Services (AWS) allows developers to build conversational interfaces.
- Watson Assistant – IBM’s offering in the conversational AI space, designed to be implemented across various channels like websites, messaging platforms, and IoT devices.
- Rasa – An open-source alternative for building conversational AI.
- Wit.ai – Acquired by Facebook, this tool provides natural language processing for applications.
- SAP Conversational AI – This solution targets enterprise-level conversational applications and is often used in customer service scenarios.
- ChatterBot – A Python library for creating conversational agents, this is a simpler tool ideal for learning and experimentation.
- Pandorabots – One of the oldest platforms, it allows the creation of conversational agents using its proprietary language called AIML (Artificial Intelligence Markup Language).
- Mitsuku – A highly advanced conversational agent that has won the Loebner Prize Turing Test multiple times.
- ELIZA – One of the earliest examples of a conversational agent, created at MIT. Though not on par with modern systems, it is still an interesting example of early natural language processing.
Conclusion
To really grasp the full capabilities of these foundational AI models, you need to spend time experimenting with them. Be aware of their strengths and weaknesses and adapt your strategy accordingly. While they aren’t magic wands, understanding these models can give you a strong foundation in the ever-expanding world of AI.
This post includes a summary of the video by Wharton Interactive’s Faculty Director Ethan Mollick and Director of Pedagogy Lilach Mollick on Large Language Models (LLMs), which can be accessed on YouTube.
#ChatGPT #Bing #Bard #Comparison #LanguageModels #Features #CreativeWriting #DataPrivacy #RealTimeSuggestions #WorkToolIntegration
Amir-Homayoun Javadi, PhD
Founder and director at 0&1 LTD
