The Broken Promise of Chatbots - And How LLMs Can Fix It

I keep doing it, I keep clicking on that little icon in the bottom right corner.
I can't resist the temptation...

The temptation that just maybe, this time, fingers crossed, the chatbot could resolve my issue.
But I'm let down each time and I don't learn.

Why Do We Keep Trying?

Why do so many people still gravitate toward the chatbot whilst 99% of them suck?

I think it's because of the promise of what could be.
The hope that the company I'm dealing with can infer my intent or my problem and resolve it immediately.

This promise was betrayed by the first wave of chatbot technology, and it's easy to see why.

The Current State of Chatbots

The majority of chatbots contain some combination of the following:

1. Keyword matching
   Picking up a word in your sentence and following a deterministic preprogrammed answer

2. Conversational menus
   Just a menu system masquerading as conversation

3. Self-help deflectors
   Incessantly suggesting articles instead of solving problems

4. Broken context
   Capturing information wrong (e.g., name = "No you misunderstood me")

We can summarize chatbots into two categories:

• A search tool (redirecting to articles)
• A menu tool

Hardly anywhere is there actual issue resolution.

The LLM Revolution

But we're now in a new age - the age of LLMs.
LLMs bring us into a position where we can finally fulfill this promise of chatbots.

Alas, I think it might be too late to rescue the word "chatbot" from its sullied past - so at Stubber we call them "Chat Agents".

How LLMs Change the Game

1. Inferring Human Intent

LLMs can understand your reason for chatting within 1-2 sentences.
This eliminates:

• Keyword matching
• Menu-based programming
• Data parsing issues

2. Actual Problem Resolution

LLMs can:

• Read documentation directly
• Work through issues with you
• Provide real solutions (not just article links)

The Big Picture

What are humans doing when building menus/apps/portals?
We're trying to anticipate user "intents" and build navigation systems.

With LLMs, we can skip this unnecessary work.
We can use natural language - our default human communication mode - to get to intent immediately.

Author: Werner Stucky