80 lines
3.6 KiB
Markdown
80 lines
3.6 KiB
Markdown
---
|
|
type: blog-post
|
|
title: Why your AI Chatbot sucks
|
|
description: Please don't burn too much money
|
|
draft: false
|
|
date: 2023-10-24
|
|
updates:
|
|
- time: 2023-10-23
|
|
description: first iteration
|
|
tags:
|
|
- "#blog"
|
|
---
|
|
|
|
The title may seem editorialized. But honestly it is true. The current landscape
|
|
for AI bots, doesn't deliver anything, aren't interesting to use. And no, I
|
|
won't use your shitty AI bot to ask questions about my groceries.
|
|
|
|
## Yet another AI chatbot
|
|
|
|
AI bots like the IPhone apps of yore, are being churned out like a warm bread
|
|
out of an oven. And funnily they all look the same, even if they may have
|
|
different seasonings. It is actually crazy how much AI bots mimic how we built
|
|
apps. When apps first became a thing. In the beginning everything was generic
|
|
cookie cutter stuff. Such as yet another:
|
|
|
|
- Downloader for ringtones.
|
|
- Instructuals,
|
|
- Tic/Tac/Toe
|
|
- Horrible company marketing app with no functionality
|
|
|
|
Like with the AI bots we fall straight into the 4th item. Putting a small prompt
|
|
on your AI bot doesn't make it useful, or interesting. Not unless you're
|
|
creative, like Wendies were on Twitter, until every fast food brand caught on,
|
|
and did the same.
|
|
|
|
Or useful. Can your AI bot actually do stuff. Can I actually buy groceries
|
|
directly using my chatbot? No, then why in all that is holy would i spend the
|
|
time to phrase a question for it? If your AI bot doesn't deliver a better
|
|
experience or original take on your product, why would anyone use it? If it is
|
|
just yet another user interface for your product, people will just use what is
|
|
most familiar to them.
|
|
|
|
For new products like this, either it has to be extremely useful, original. If
|
|
neither fits, you won't get any traction.
|
|
|
|
A lot of these AI bots feel like talking to a generic receptionist, whom likes
|
|
to talk, way, way too much. Not that I can relate :,)
|
|
|
|
## Gain traction before it becomes a commodity
|
|
|
|
If you can get the same product everywhere, you're gonna choose what you're most
|
|
familiar with. Get users before this happens, otherwise you would just have
|
|
burned a lot of money on compute and R&D.
|
|
|
|
Already you can tell that all these companies racking in billions of dollars in
|
|
investment building foundational models (the stuff that powers your favorite
|
|
chatbot). Will go bust in the next few years. Building these models are
|
|
expensive, risky, and very short-lived. If these companies cannot out-compete
|
|
their rivals, then they will just have burned a ton of money for no avail.
|
|
|
|
Building these models looks a lot like chip manufactoring. It requires tons of
|
|
capital, but in just a few years, that old investment is pretty much worthless.
|
|
All the users and money is given to the ones at their forefront of their field.
|
|
Whether that is in monthly users, cheapest operations, originality etc.
|
|
|
|
I fear for those playing in the generic foundation model field. You may be able
|
|
to buy 100 A100s now, but in two years that investment is pretty much obsolete,
|
|
so unless you get proper mindshare in that amount of time, you are doomed. We're
|
|
probably gonna see some interesting attempts from some of these companies over
|
|
the next years, trying to diversify their models, into specific fields, or
|
|
products. Because right now, being a generic general purpose chatbot, is a
|
|
difficult journey, especially when up against giants such as Google, OpenAI
|
|
(Microsoft), Apple etc.
|
|
|
|
GPT4 and rivals are already proving to be very reliable. In a few years ones the
|
|
next models come out, and they're further optimized for speed, and cost. GPT4
|
|
and its fellows will become a commodity. There will probably still be a marked
|
|
for the best and newest models, like with phones. But for most people the older
|
|
cheaper model, will probably be just fine.
|