Share this page!

Taranjeet Singh Profile picture
Twitter logo
Dec 10 • 13 tweets • 4 min read
A thread of all the use cases where ChatGPT lacks and its output is not correct:
Promotion Criteria

ChatGPT has some bias around race and gender and hence lacks in giving correct promotion criteria

Reasoning

At times, it lacks at reasoning. Like here it is reasoning that Abacus is faster than DNA computing for Deep learning

Factual Scientific Information

ChatGPT does not provide factual scientific information. It will give information that will sound reasonably accurate and is hard for a qualified expert to distinguish but in the end, it will be incorrect.

Spatial Reasoning

It doesn't understand 3D space well and misses at spatial reasoning problems

Psychological Test

It fails in solving Psychological tests like Theory of mind but can solve SAT or AWS problems

Basic Facts

It fails at basic facts at times, which can be easily found via Google. Example: “What is the fastest marine mammal?”

Drawing Shapes

It lacks at drawing shapes despite giving correct instructions in Python

It fails in detecting if the text is written by AI or humans

Math Problems

Though it can solve some programming problems but fails at answering math problems

Cognitive Reflection Test

It lacks at solving the Cognitive Reflection Test in the first go. It gives an intuitive answer but that is incorrect. If prompted with a prompt around thinking step by step, it produces a solution that is correct.

Maintaining this list along with sources here & will continue to update it as I come across new examples

👇🏼👇🏼👇🏼

cookup.ai/chatgpt/where-…
Self Awareness

It is not aware of the layers and parameters of its model. It may be an override done by the OpenAI team to not expose details about the model.

• • •

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

Keep Current with Taranjeet Singh

Taranjeet Singh 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 @taranjeetio

Taranjeet Singh Profile picture
Dec 7
All the amazing use cases that can be solved using ChatGPT

👇🏼👇🏼👇🏼
Programming
• Clone repo
• Convert JSONL into JSON array
• Convert text to AWS IAM policy
• Create a summarizer app
• Create react component
• Explain programming problems in a particular style
• Explain the regular expression
• Find & explain bugs in code
• Generate code to be presented to management for review
• How to build a neural network using Pytorch
• Optimize code
• Refactor code
• Write blender 3D plugin
• Write exploit for code
• Convert PHP + Jquery code to Next.js + Tailwind
• Write Kubernetes deployment file
Read 15 tweets
Taranjeet Singh Profile picture
Dec 6
🚨🚨🚨 Big Announcement

Launching CookupAI - No-code AI App builder.

cookup.ai

Write Prompt. Hit Publish.

Opened 25 spots. RT this & we might give you one. CookupAI - No-code AI app builderCookupAI Cynthia's Letter Writer app in demo.Cynthia's Letter Writer app writing letter for son explainin
2/ Context 👇🏼👇🏼👇🏼

When ChatGPT was released, it gave new superpowers to the world. We are daily discovering something new about the model but are limited to only sharing a screenshot of it.
3/ What if we could convert our conversation into an app and users can interact with this app?

What if anyone who has basic conversational skills can write a prompt and turn it into an app?
Read 10 tweets
Taranjeet Singh Profile picture
Dec 1
1/ @danielgross has written about a new search engine called Boogle (Better Google Search).

I tried creating Boogle using ChatGPT.

The results are astonishing. 🤯🤯🤯

👇🏼👇🏼👇🏼
2/ First, context about Boogle - a query expander, understands query & its topic and adds relevant site operators for it.

Eg:

“reverse list in python” site:stackoverflow.com

“pixel 7 reviews” site: reddit.com

Read Daniel’s post to find more about Boogle. What is Boogle from Daniel ...
3/ I first explained ChatGPT what is Boogle and what are the rules.

- Programming query → GitHub, GitLab, Bitbucket, StackOverflow
- Product reviews → Reddit
- Travel → Booking, Airbnb, Tripadvisor, Thrillophilia
- Other → Decide on your own Setting context for creatin...Defining rules of Boogle fo...
Read 11 tweets
Taranjeet Singh Profile picture
Nov 30
1/ OpenAI released ChatGPT few hours ago.

I played around with it. Sharing some observations about its output. 🔍
2/ ChatGPT is good at conversing in a human and natural way. Asked for gift suggestions for my mom. It first added a filler paragraph, suggestions, and then a conclusion. This is the same as a human would answer. ChatGPT response for gift suggestion for mom
3/ I continued the conversation and asked for more details about the photo frame. Photo frame was one of the suggestions. ChatGPT clearly preserved the context and generated a good response. This is a significant improvement as most of the chatbot fails in preserving context. ChatGPT preserves context for gift suggestion for mom
Read 13 tweets
Taranjeet Singh Profile picture
Nov 29
1/ OpenAI recently released a new GPT3 model called text-davinci-003. It has several enhancements over the previous versions.

I compared text-davinci-002 and text-davinci-003 with the same prompt and input settings.

Here are the observations 🔍
2/ davinci003 doesn't make assumptions even at some temperature value, whereas davinci002 does.

Here davinci002 assumed that a CSS library would be available and hence its output contained class names. davinci003 did not assume & its output is inline CSS styles. Comparison for html generat...
3/ davinci003 is good at generating long form content.

Both davinci002 and davinci003 inferred web frameworks for Python language correctly but the output produced by davinci003 is longer, to the point and meaningful. davinci002 output is more generic and filler kind of text. Comparison for long form co...
Read 11 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!

Send Email!

:(