Greg Kamradt Profile picture
Aug 7 10 tweets 2 min read Twitter logo Read on Twitter
Here are the 5 AI niches I would pick if starting today:

1. B2B CompanyGPT Interface
2. ETL for Retrieval-As-A-Service
3. AI Characters
4. Unstructured to Structured Data
5. Around The Org - Enterprise Edition

The breakdown of each one: Image
1. B2B CompanyGPT Interface
50% of employers don’t pay for ChatGPT

This tells me:
1. Companies haven’t enabled their workforce with LLMs yet
2. There’s an opportunity for a closed-source interface for LLMs

Basically the ChatGPT you wish you had for your employees:
* No API Keys, seamless model selection
* Living knowledge of documents
* Update LLM configuration files with company docs, tone, mission statement

For the admins: analytics/usage, employee engagement, highlight unique use cases that should be evangelized
2. ETL for Retrieval-As-A-Service
Internal documents, best practices, and resources are an absolute mess

But messy is ok when a computer can sift through them - a great use case for LLMs.

Service to manage indexing, splitting, updating, serving docs
Don't start this as a scale play, make one customer very happy first, then move onto #2, then N

Someone is already doing this? Great. The TAM is massive.
3. AI Characters (or mimicking a person)
IMO building a character is going after the hard problem of mimicking the output of a person.

Getting a few cheeky chat messages is straight forward

But copying someone’s persona, decision making process, and style is hard.
My guess is the solution will be the program around the LLM, not the LLM itself.

Building tech around todays LLMs will only get better with tomorrows.
4. Unstructured to Structured Data
Oldy but goodie - LLMs can structure data that was previously locked. My favorite applications are spoken word and recordings

Others include unstructured text in PDFs, gov’t docs, social data, consumer behaviors, etc.
5. Around The Org - Enterprise Edition
When I was at Salesforce, trying to stay up to date with what was going on around the org was a major pain.

You need to track slack, chatter, and read a bunch of documents.
It paid to be informed.

Your decisions are better and you can speak in the language of your peers more efficiently

I want an LLM that could keep me informed on what is going on within my >1K person company.

Emphasis on a personal “so what?” to my role.

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Greg Kamradt

Greg Kamradt Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @GregKamradt

Jun 21
The @myfirstmilpod is filled with a ton of business ideas, stories, advice, mental models and words of caution

But pulling structured data out of an episode can be a pain

Let’s do a low-code friendly tutorial on a method that had awesome results using @LangChainAI @pinecone
This method is easily adaptable to any podcast, video, meeting, movie, book, you name it

If you want to see a full video tutorial + full code, check out the links at the bottom of this thread ImageImage
Step 1: Load Your Transcripts

A quality transcript w/ speakers spit out can be hard to come by. For this exercise I used @AssemblyAI

There are a bunch of tools out there - pick your favorite!

The full transcript is included with the code Image
Read 13 tweets
Jun 6
The LangChain Cookbook: Part 1 - The Fundamentals

This @LangChainAI tutorial will ramp you up to the 7 core concepts of building apps powered by language models

You’ll learn LC's Schema, Models, Prompts, Indexes, Memory, Chains, Agents

+150K views on YouTube. Code below!
@LangChainAI Some comments from YouTube Image
Read 5 tweets
Apr 28
I recently had a project to parse a ~1hr podcast for topics, ideas, sections etc. ~12K tokens

Then generate a few sentences to summarize each section.

300+ episodes

How would you approach this problem while keeping tokens down?

I did it in a few passes with @LangChainAI, cont
Pass 1: Split the doc into big chunks.

Map reduce with custom prompts to pull out topics, no topic summary.

There ended up being 10-15 topics.

I found it was too much to ask the LLM to give me ~15 topics *and* a solid summary.
Pass 2: Split the doc again but into smaller chunks, embed them and do the vectorstore dance.

Loop through each topic and ask the LLM to tell me more about it with the top K similar results.

It’s a bit expensive, but worked pretty well.
Read 5 tweets
Apr 25
Relistening to sales calls is so March 2023

I built Thimble to help teams extract data from sales calls, powered by @LangChainAI

Here’s the journey so far:
The problem? 100% of AEs that I’ve talked to say they ‘relisten’ to their sales calls for follow ups, remember next steps, and to up their CRM

Why now? Language models make it extremely easy to parse out the insights that AEs would normally pull manually
The idea for Thimble was born a few months ago after showcasing a product prototype

There were a few interested users so I kicked off a beta program.

Since then over 30 AEs have given me awesome feedback about their post-call experience

Read 8 tweets
Apr 5
AI Trends I'm interested in 4/5/2023:

1. Managed Retrieval Engines - Getting the *right* context to your AI is tougher than it sounds. @Metal_io announced a @LangChainAI integration. I'll be watching

2. Plugin Developer Monetization...

(full thoughts in a notion doc below)
3. Plug-in Dev Shops - Small team of Devs can ‘translate’ 1000s of API for SMBs and services while taking a cut of traffic monetization

4. Unstructured > Structured - Extracting information from messy text is going to be huge
5. AI Reflection - LLMs don’t always get it right on the first try, but asking them to check their work results in surprisingly good improvements.

6. Drag & Drop Chain Builders - No-Code but for LLMs. Ex: @Langflowai
Read 4 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!

:(