Show HN: WhyBot, making GPT-4 question itself(whybot-khaki.vercel.app)
whybot-khaki.vercel.app
Show HN: WhyBot, making GPT-4 question itself
https://whybot-khaki.vercel.app/
33 comments
Asking WhyBot, "What is today's date?" caused a chain of thought provoking ideas about modern society's dependence on technology to tell the time. Cool!
I think the UI is cool, but I don't really find it that amazing honestly.
It just seem to keep prompting itself and never really going anywhere?
It just seem to keep prompting itself and never really going anywhere?
I have not seen a better demonstration on visualizing unlimited/ open ended reasoning like this.
I really love this.
I'm very curious (and I thought about this before which shows the Zeitgeist) if this can change or enhance the general reasoning skills of people.
Our generation already google things constantly, so much that an older couple told me once that their kids don't ask them stuff but Google it.
But the reasoning aspect was missing.
Now you can't just Google facts but also ask reasoning questions without worries.
Ignoring potential miss guidence in these models and hopping for an emerging reasoning truth through wisdom of the masses.
Self fulfilling prophecy perhaps?
I really love this.
I'm very curious (and I thought about this before which shows the Zeitgeist) if this can change or enhance the general reasoning skills of people.
Our generation already google things constantly, so much that an older couple told me once that their kids don't ask them stuff but Google it.
But the reasoning aspect was missing.
Now you can't just Google facts but also ask reasoning questions without worries.
Ignoring potential miss guidence in these models and hopping for an emerging reasoning truth through wisdom of the masses.
Self fulfilling prophecy perhaps?
This is incredibly!
I've been benchmarking several LLMs to identify a possible bug in a snippet of code given no context on what is expected from the code. Neither chatgpt, vacuna, open assistant or mpt where able to identify the bug. When I fed the same prompt into this app, the initial reply said there were no bugs, but the first branch immenditly questioned itself with the expected line of thought to identify the bug!
I might subscribe to chatgpt just to use this app, great work!
I'd love it if this got open sourced.
Here's the prompt if you want to test it out:
Can you find any bugs in the following snippet?
```ts
const queuePositions = [
1,
2,
3
];
const pairs: number[][] = [];
for (let i = 0; i < queuePositions.length; i += 2)
pairs.push(queuePositions.slice(i, i + 2));
```
I've been benchmarking several LLMs to identify a possible bug in a snippet of code given no context on what is expected from the code. Neither chatgpt, vacuna, open assistant or mpt where able to identify the bug. When I fed the same prompt into this app, the initial reply said there were no bugs, but the first branch immenditly questioned itself with the expected line of thought to identify the bug!
I might subscribe to chatgpt just to use this app, great work!
I'd love it if this got open sourced.
Here's the prompt if you want to test it out:
Can you find any bugs in the following snippet?
```ts
const queuePositions = [
1,
2,
3
];
const pairs: number[][] = [];
for (let i = 0; i < queuePositions.length; i += 2)
pairs.push(queuePositions.slice(i, i + 2));
```
No need for complicated multi-step apps, GPT 4 can find the bug straight-up:
The code snippet you provided is syntactically correct TypeScript code. However, there might be a logical issue depending on the intended use case. The code is set up to iterate over the queuePositions array and slice it into pairs, but if the queuePositions array contains an odd number of elements, the final pair will contain only a single element.
The code snippet you provided is syntactically correct TypeScript code. However, there might be a logical issue depending on the intended use case. The code is set up to iterate over the queuePositions array and slice it into pairs, but if the queuePositions array contains an odd number of elements, the final pair will contain only a single element.
ChatGPT (3.5-turbo) also provides a similar response:
"As an AI language model, I cannot guarantee the correctness of the code, but I could not find any obvious bugs in the given code snippet. However, here are a few observations:
- The queuePositions array must contain an even number of elements for this code to work correctly. - pairs is initialized as an empty array of arrays, where each inner array will contain exactly two numbers. - The for loop iterates through queuePositions array in steps of 2, taking two consecutive elements from the array each time starting from index i. - The slice method extracts a portion of the original queuePositions array starting at index i and ending at i + 2, but not including i + 2. - The extracted portion is pushed to the pairs array. - The resulting pairs array will contain pairs of numbers from queuePositions."
"As an AI language model, I cannot guarantee the correctness of the code, but I could not find any obvious bugs in the given code snippet. However, here are a few observations:
- The queuePositions array must contain an even number of elements for this code to work correctly. - pairs is initialized as an empty array of arrays, where each inner array will contain exactly two numbers. - The for loop iterates through queuePositions array in steps of 2, taking two consecutive elements from the array each time starting from index i. - The slice method extracts a portion of the original queuePositions array starting at index i and ending at i + 2, but not including i + 2. - The extracted portion is pushed to the pairs array. - The resulting pairs array will contain pairs of numbers from queuePositions."
Every single time chatgpt 3.5 told me there were no issues. So even if you got a good response out of it, it might have been just luck.
And I just got a good result off of gpt 3.5 and 4.
3.5 turbo:
The code snippet seems to be correct and there are no obvious bugs. It initializes an array of queue positions, and then creates pairs of adjacent positions using a for loop and slice method. The resulting pairs are pushed into a 2D array. However, it is worth noting that the last pair may only contain one element if the length of the array is odd.
4:
The given TypeScript code snippet seems to be correct and has no syntax errors. It initializes an array queuePositions with three elements (1, 2, and 3) and an empty array pairs. The for loop iterates through queuePositions with a step of 2, slices the array from the current index to the next index, and pushes the sliced subarray into the pairs array. After running the code, the pairs array will be [[1, 2], [3]].
However, it's important to note that the last element of queuePositions will be in an array by itself in pairs array since the loop has a step of 2. If this is the intended behavior, then the code is correct. If you want all elements in pairs to have the same length, you may need to handle the case where the length of queuePositions is odd.
Maybe the tool proposed by the OP just improves the probability of getting a good result. Which is quite good IMHO.
3.5 turbo:
The code snippet seems to be correct and there are no obvious bugs. It initializes an array of queue positions, and then creates pairs of adjacent positions using a for loop and slice method. The resulting pairs are pushed into a 2D array. However, it is worth noting that the last pair may only contain one element if the length of the array is odd.
4:
The given TypeScript code snippet seems to be correct and has no syntax errors. It initializes an array queuePositions with three elements (1, 2, and 3) and an empty array pairs. The for loop iterates through queuePositions with a step of 2, slices the array from the current index to the next index, and pushes the sliced subarray into the pairs array. After running the code, the pairs array will be [[1, 2], [3]].
However, it's important to note that the last element of queuePositions will be in an array by itself in pairs array since the loop has a step of 2. If this is the intended behavior, then the code is correct. If you want all elements in pairs to have the same length, you may need to handle the case where the length of queuePositions is odd.
Maybe the tool proposed by the OP just improves the probability of getting a good result. Which is quite good IMHO.
Shame it's not publicly available.
It is, it just costs money.
Paid access is public access, it's just not free access.
Paid access is public access, it's just not free access.
Sorry, I was mistaking this with api access.
Interesting, but I've found the questions it asks are all pretty conventional and run-of-the-mill, and I wish it would stay on topic for follow-up questions, rather than veering off on barely related tangents.
I think adding some sort of ability to alter the way it forms follow up questions would 10x the outputs.
I love this, I found that I was mostly curious about one branch and it took time for it to fill out all the other branches before the one I wanted was filled.
Is there any way for me to pause all the other branches and have it only go deeper on one of the branches and then maybe restart other branches selectively?
Is there any way for me to pause all the other branches and have it only go deeper on one of the branches and then maybe restart other branches selectively?
Way more interesting than you'd think given the simplicity and slick ui. ability to choose branches would be great.
You should tell ChatGPT to translate your question to client language. I tried in Vietnamese but then the next questions and answers is still in English.
Pretty awesome idea btw. I would like to highlight a branch for deeper questions. Just to speed things up.
Pretty awesome idea btw. I would like to highlight a branch for deeper questions. Just to speed things up.
Fixed--made it generate questions in the same language as the answer! Do you find that GPT 3.5 can write in fluent Vietnamese?
Branch highlighting is on the top of our to do list, should be able to get it out soon!
Branch highlighting is on the top of our to do list, should be able to get it out soon!
Like I mentioned on my other comment, this is amazing!
Are there any plans to integrate this with other LLMs providers like open assistant or hugging faces chat?
Are there any plans to integrate this with other LLMs providers like open assistant or hugging faces chat?
This is crazy cool. This would be epic as an extension for text-generation-webui, so that we can run it with local LLMs!
Does it always stop after going 10 layers deep?
I asked it why 1+1 equal 3. It’s beautiful.
Please add an export.
Please add an export.
Really great work here, it has inspired me and I think you are on to something.
This is very clever. Thanks for sharing.
Amy plans to open source?
This is great for technical exploration! Are there plans to let users create their own personas?
We originally built this for the AngelList Agent Hackathon (https://twitter.com/AqeelMeetsWorld/status/16502799744050421...) and got a lot of interest from folks asking to play around with it. So we thought it’d be fun to brush it up and release it as a web app! It’s a work in progress and we plan on adding more features, such as saving, sharing, focusing on one branch and potentially executing code.
We hope you enjoy playing around with it and would love to hear any of your feedback or thoughts.