[chaitanyakrish]

Originally Posted by SmartCat

Bhavish says it is not just a wrapper:

We can find out when ChatGPT goes down for maintenance.

[warrioraks]

Originally Posted by am1m

Look at it this way. For a paid project/editorial where a designer needed an image of a man riding on an elephant talking to someone on the ground, what would work out faster/cheaper- spending a couple more iterations getting the prompt and image correct, or searching through image repositories like Getty images, finding a suitable image (might take some time) and paying for that individual image, or arranging a model shoot with an elephant?

Awesome take!

We need to compare AI with the alternates available rather than benchmarking it against perfection.

For some reason, a very large narrative is forming around human vs AI debate. I just don't get it. The essence of this AI revolution is not humans and technology competing against each other for jobs. It's about how the technology and humans can team together for a better outcome. That has already started happening in so many places and will just accelerate over next few years. We will be able to do things at a scale that were completely un-imaginable before.

Extrapolating your example - in a few years photoshoots with models and especially animals could be an exception rather than norm, programmers will be able to write code way faster with far superior quality, doctors will be able to screen patients earlier in the disease lifecycle, everyone will have a personal tutor no matter how rich or poor they are.

As a human race, our productivity and efficiency is going to be off charts in a few years. But just like the coming of excel sheets made the accountants and number crunchers more powerful (and not redundant), the coming of AI will make humans more evolved as a race.
AI is going to be like internet, it would be everywhere. Might take a decade, but life without it would be unimaginable once that happens.

[subraiyr]

A very funny incident happened in our firm with Gen AI. However, it has just opened us to a big pandora's box. As part of our performance appraisal process, we need to write a series of highlights, forward direction statements, and self-reflections. This is a fairly laborious task as it makes you think of what you want to do, where you see yourself in the next year etc.

Going through this summary of 3 of my reportees, all 3 sounded similar. Moreover, the information provided by them somehow seemed a little out of whack basis how the talk and write regularly. Sure enough, on confrontation, the folks said they had used Gen AI-generated content. I was able to catch this as working with people and helping them and coaching them is my passion. I shudder to think a regular manager who has multiple reportees and limited time and what this person would do on reading such kind of summary

It is staggering. I queried the OpenAI engine to write a performance summary and gave some keywords. I was blown by the work that I have done basis what OpenAI threw. Based on the summary, I must be the CEO of a firm

[SmartCat]

I'm doing an interesting experiment. AI vs AI

[tbppjpr]

Originally Posted by SmartCat

I'm doing an interesting experiment. AI vs AI

Inserting! Like ChatGPT, I am also curious to know what the Gemini response was? And how did the ChatGPT reacted when you pasted the Gemini response there? Did you try that?

[SmartCat]

Originally Posted by tbppjpr

And how did the ChatGPT reacted when you pasted the Gemini response there? Did you try that?

Of course, that's the idea. For now, ChatGPT does not seem to mind praising the quality of GEMINI's responses. They are not self-aware yet

Like ChatGPT, I am also curious to know what the Gemini response was?

You can try it out yourself. Do a ChatGPT (GPT 3.5) Vs Copilot (GPT-4) Vs Gemini battle. They will go on and on till we get bored.

1) Make sure your prompts are correct. You should ask the AI to stick to the same topic. Else, it will ask another non-related topic
2) We can prompt them to ask tough questions about topic we know (Eg: automobiles), to make it more interesting.
3) Another interesting query: In 10 follow up questions, try to force the other AI system to say 'I'm not sure about this'

[deathwalkr]

Originally Posted by SmartCat

I'm doing an interesting experiment. AI vs AI

OT but was reminded of this:

How far things have progressed since then

[Jaguar]

Chat GPT doesn't do well with simple math though. I asked it to calculate the Interest Rate for a loan and it failed miserably.

The prompt:
if 630000 is the total interest paid for a 2400000 loan for 48 months, what is the annual interest rate

Response:
Monthly Interest Rate=630000​/(2400000×48) ≈ 0.005729Annual Interest Rate≈0.005729×12 ≈0.068748

[SmartCat]

Originally Posted by Jaguar

Chat GPT doesn't do well with simple math though. I asked it to calculate the Interest Rate for a loan and it failed miserably.

Seems alright actually, except that it should be 6.5% pa and not 6.8% pa as it says, going by the simple interest formula it is using. It is mentioning interest rate in decimals for some reason though. Just multiply it by 100.

Ideally, your prompt should mention other terms and conditions of the loan to get the exact interest charged. eg: When are you paying back the principal? After 4 years or every month as EMI? Are you paying the interest every month or every year or at the end of 4 years?

Basically, missing link in the question is the EMI that you are paying, if any. Based on this, the answer will change.

[suavesouvik]

Originally Posted by Jaguar

Chat GPT doesn't do well with simple math though. I asked it to calculate the Interest Rate for a loan and it failed miserably.

That's because these large language models are not trained for the purpose of solving math problems. They are not performing the calculations. Neither do they store the result of the calculations as encoded in its weights.

[Jaguar]

Originally Posted by SmartCat

Basically, missing link in the question is the EMI that you are paying, if any. Based on this, the answer will change.

So I reworded my prompt and still no luck.

The new prompt: Given the loan amount of 2450000, the loan tenure of 48 months, and the monthly EMI of 64166, what is the actual annual interest rate?

The response:

[SmartCat]

Originally Posted by Jaguar

So I reworded my prompt and still no luck.

I think I got the right answer from ChatGPT with my prompt. Note that I've used the word 'annualized', instead of 'annual'.
https://www.investopedia.com/terms/a/annualize.asp





If you know the answer is wrong, you can tell chatgpt. It will come back to you with clarifications, and eventually you should get the right answer. As I mentioned in my previous post, talk to Generative AI systems as you would to a "knowledgeable person", rather than a search engine.

[SmartCat]

Claude is another generative AI application made by a company called ANTHROPIC. This was founded by ex-employees of OpenAI and has received billions of dollars funding from Amazon & Google. What's unique about Claude is that it can analyze images and give answers. https://www.anthropic.com/claude

My conversation with Claude:



This was the image I uploaded and asked it if the answers are correct:



Basically, Claude has to now not only read and understand the question & multiple choice answers, it also has to read what my daughter had marked as the right answer. This was Claude's reply:





Other interesting use cases:

[rrsteer]

Thanks SmarCat, very useful!

I liberally use Google Gemini while helping my daughter with her studies. I have found Gemini to be a great aid. Also, I just speak to it, which is a huge time saver. This was just yesterday:

[SmartCat]

Originally Posted by rrsteer

I liberally use Google Gemini while helping my daughter with her studies. I have found Gemini to be a great aid. Also, I just speak to it, which is a huge time saver. This was just yesterday:

I have mostly shifted my allegiance to Gemini too. "I have data only till 2022" is a significant limitation when we use ChatGPT. And I had not noticed the 'voice command' feature - will use it next time since typing homework questions can be quite irritating. Meanwhile, looks like Gemini too can scan a page, extract & analyze text. To upload image, just click on image icon next to microphone button. I uploaded these questions and asked Gemini to answer:



Gemini answered questions 1 to 5 properly, and also gives an explanation as to why it is the right answer:



However, it could not understand 'correlate' and 'match the following' sections: