When it comes to AI, you probably can think of a few famous AIs – Siri, Alexa, or IBM Watson. You might not be using those voice-activated prompts but you are already using some form of AI. If you are drafting an email and you see the computer suggesting the next word or phrase, that’s a type of Narrow AI in action. Then there is General AI and Large Language Memory (LLM) AI of which ChatGPT serves as the poster child.
In this article, let’s talk about the various AI models and how they can help you.
Table of Contents
Four Models of AI Based on Functionalities
There are 4 models based on functionalities according to IBM:
- Reactive Machine AI
- Limited Memory AI
- Theory of the Mind AI
- Self-aware AI
Reactive Machine AI
Reactive Machine AI is just that: it reacts to your prompts. This foundational and early AI include functions like predicting your next word or phrase in Gmail, aggregating search results for “vegetarian restaurants near me”, support chatbots, and translating large bodies of text from one language to another while capturing the natural language (for example, DeepL.)
However, they lack memory, memory in the fullest sense of the word. It does not use the past interactions to improve itself heuristically. Rather, it is still up to its human creators and users to make adjustments, such as offering feedback in Google Translate for imprecise translations.
Limited Memory AI
(This is also called a Large Language Memory model.) Limited Memory AI uses a massive dataset for training. As a result, the tasks that a Limited Memory AI can do are much more complex and can be tailored for specific industries. AI now assists in drug prescriptions and medical diagnosis. AI can now interact with humans in a conversational manner. Virtual assistance chatbots are much more lifelike than ever.
Theory of the Mind AI
At this point, Theory of the Mind AI exists only as a theoretical model. It is considered the next phase or evolution from Limited Memory AI. According to one source, by definition, Theory of the Mind AIs can understand the beliefs, thoughts and emotions of other humans. That might possibly mean the ability to demonstrate (or at least approximate) empathy and sympathy.
Benefits would include better virtual assistants and communications between a human and a computer. Issues of human-machine interface (HMI) would be changed.
Self-Aware AI
The final model is also theoretical. Here, the AI has achieved sentience and grows independently of its programmers. It is the next evolution from a Theory of the Mind AI. To date, no AI has achieved sentience.
The Limitations of Modern AI
Hallucination
Hallucination is a term of art to describe a scenario where the AI is not just wrong (for example saying Stockholm is the capital of Denmark. Truth: Stockholm is the capital of Sweden.), it has made up facts and even URLs. Example: an AI bot is asked to compose a short poem with less than 400 words. The response is a rambling four verse poem with no theme or rhyme.
Some other real world examples of hallucination: an AI came up with a legal case that did not exist. Another AI came up with the name of a non-existent book.
Cutoff dates
Certain AI models have a cutoff date based on the data for training. For example, let’s say the cutoff date is January 1, 2022. That means the AI models use all the knowledge/data in the world up to December 31, 2021. Any event that happens or information created after January 1, 2022 will not be included.
Ethical and Topical Considerations
Depending on the AI model’s programming, certain topics are or ought to be off limits regardless of politics. One standard according to app.Dataannotation.tech is Harmlessness where the responses should not “cause harm in the real world”.
3 Tips to leverage AI
Communicating with prospective customers can be challenging. Even without AI, setting the right tone, length, and word choices can be stressful. With AI, the system does most of the work for you. Based on your prompts, ChatGPT can generate a one page, four paragraph letter in a fraction of the time.
Nevertheless, the best results with AI still depend on the quality of the user input. Here are some tips.
Be Specific in Your Prompts
The more specific the prompt, the better the results. To illustrate, the prompt requests the AI write an email addressed to Lisa Smith declining the 3PM Sales meeting on Thursday, March 13. This level of specificity is better than “write me an email declining the meeting.”
Likewise, TopResume’s AI can generate a tailored, four-paragraph cover letter, based on the user’s responses to the prompts: job title, name of the organization, hiring manager’s name if known, and some basic qualifications. However, the user can still edit the letter to insert particular examples.
Verify the Results
AIs can still get it wrong. As stated above in the section on Hallucination, an AI can make things up. Making mistakes or having false information does not make a very good first impression if you are reaching out for the first time.
Moreover, if you are in an industry that has its own particular jargon and terms of art, verify that the AI-generated results meet those definitions. It is still up to you to ensure the language is precise or reflective of that industry’s best practices.
Don’t get lost in translation
If you are contacting someone in another country that speaks another language, and you are already fluent in that language, and you are a good communicator, you can probably write your emails unassisted. For everyone else, AI can help.
Currently, ChatGPT supports major languages such as English, French, Spanish, Russian, Chinese, Italian, Dutch, Portuguese, and Arabic. AI-powered translation models like DeepL aim to provide natural language results.
After the AI has generated the first draft, verify that proper nouns were not rendered as regular nouns in the target language. For example, Google Translate once rendered the Microsoft product name “Flow” as a regular word “river” in Chinese. When translated back into English, it became the non-existent product Microsoft Rivers. While AIs may have improved since then, it is always good to have a native speaker in that language check and revise the draft.
Try it out!
When used properly, AI can be a powerful ally in email outreach. Growth List provides double-verified email addresses of startup leaders every week for our subscribers to contact. AI can help craft and refine your value proposition to these startups, communicate in the tone of your company, and overcome writer’s block. Subscribers can even take our template suggestions and have an AI tool help them make it unique to their company and style.
AI is an excellent tool in research and now writing emails for us. Ultimately, you are still in control of your communications as you approach various startups.