Now that ChatGPT has dominated the headlines for a while, what’s your verdict?
Two main questions to answer in the thread:
Of course, feel free to add in any other thoughts and opinions you have.
I started out by using it to be more concise in email structure and brainstorming possible subject lines. Have also found it useful to make templates for note-taking on discovery calls and account research.
My favourite prompt: “Imagine you are…insert your role and you are trying to insert problem here ?”
Biggest concern: ensuring not to share sensitive information! Always keep things generic.
It’s useful for condensing emails and building compelling subject lines for emails. I try and limit what I input in there due to privacy concerns. Interested to see how others use it in the sales motion.
I use it to get past writers block or to approach the email from a different perspective. I’ll take what it generates and revise it to my liking.
I am a huge fan of ChatGPT. Sometimes when I am meeting with a new prospect and their title can be confusing, I ask what ChatGPT what they do within a specific category, for example, a large enterprise bank. It’s amazing how accurate it is. I also have it revise cold outreach emails.
I think it’s really useful to ask it to generate compelling subject lines or provide great calls to action. It’s also helpful to shorten an existing email or make it sound more casual (less salesy).
Has it ever backfired on you yet in any way, Kate?
I have been using Chat GPT to research the top 10 priorities (in bullet point format) of business titles that I may be less familiar with i.e. CISO (Cheif Information Security Officer). This has worked out well.
Although not exact, but within the ballpark, here are a few things I use Chat GPT to cut down research time:
Understand how the company make money and the distribution
Understand the company’s corp level metrics
I am new to using it, but found some good value so far.
I personally am using the Bing ChatGPT experience for understanding the associated companies to a parent company for Enterprise clients. It isn’t always obvious from their business pages and you get more folks to prospect into. Some subsidiaries run independently.
@Matt.Conley Definitely have had outdated or factually wrong information. The LLM based AI tools are great at analyzing and corelating data, building out logical and organized templates for emails, presentations, and bringing in data from different sources but the training of the model, the amount of data and depth of data it is pulling from matter in how it responds.
I am not an expert with AI but i am learning and listening to people far smarter than me on it. Reinforcement learning is the key to these models or tools getting better and more accurate. You cannot replace humans with these tools but the human oversight and guidance of these tools elevates the output substantially.
I have tried Google Bard to generate a pitch for a new company and it was pretty decent. I would still work on edits and try to refine the messaging further. So far, I find the generative AI tools to be as useful as search - for generating initial ideas - it gives you an easy way to start. I would still opt to keep refining the message personally and in combination with AI with deeper prompts.
ChatGPT is a decent writing partner. I mostly use it to bounce ideas off of by asking it to “reword” concepts or emails I’ve written, and then taking the best ideas. I’d say it fails in being too creative, but it’s a decent place to start with.
@jacob.welch When you got wrong info, how impactful was that? Did it affect the progress on any deal in any way?
I’m hesitant to rely on anything (even if it’s an efficient shortcut) that could be factually wrong. Am I overthinking it?
It has not hindered a deal because I always confirm the data within it. folks also have to make sure that proprietary data is not being pulled into messaging. Alot of the usefulness of ChatGPT is in prompting. How you prompt will determine how good it is. Human input is essential. Look at LinkedIn for Prompt cheat sheets. there are very good ones on there. I will try to share some on here as well.
Like you said, LLMs are highly fluent and conversational, but contextual grounding is a major issue and the reason why I wouldn’t trust it for facts specific to understanding a company or an executive. I usually feed it databook insights for grounding and then use single and few prompt for things like condensing emails.
It’s important to remember that LLMs do not “lookup” the answer in an explicit discrete knowledge source, instead they probabilistically generate the answer based on its statistical knowledge (which has been learned by training on extremely large volumes of explicit textual content).
When you ask the LLM a question, it’s answering by probabilistically “recalling” relevant information based on it’s breadth of context.
However, due to design, there are some reasons I wouldn’t depend on it:
Can’t tell what’s real or not (look up: "hallucinations) – Since LLM’s probabilistically generate answers, they suffer from the hallucination problem, where the model generates information that appears factual, but is completely made up. It also can’t point to a source. You would need to manually double check to confirm on the quote or information that you’re pulling.
Lack of Control for Search and Ranking-- The LLM’s search/ranking algorithm is a black-box (implicitly done by probabilistic generation) and hence there is no way to control:
@jacob.welch good article on it. Would be interested in your thoughts on how you check your info? https://thenewstack.io/how-to-reduce-the-hallucinations-from-large-language-models/
I would say that in my experience, ChatGPT is a lot better than Bing Chat (even though Bing is using GPT4 models). No idea why.
Having said that, have a great experience using it to edit my LinkedIn Posts, draft some ideas. Remember the age old saying garbarge in garbage out? Prompts are one of the most important thing to learn here to ensure that you generate the right response.
