[Samurai]

Originally Posted by sushantr5

To my surprise it wrote a code, one in c++ and another in JAVA. It was really small use case. I was not hoping that it will understand specific words and use case from the domain. It stunned me by writing beautiful code.

Next, please do google search using the same keywords and find the source of the code.

I recognised the code in my instance because I am very familiar with boost sample code. So I could directly go to the source without searching.

[warrioraks]

This is a fantastic achievement. Most of these systems are over-glorified but one cannot argue about the strides Machine Learning has made over last decade. GITHub has Copilot which does something similar. I am really excited about ML eliminating so many redundant tasks from our dictionary.

On the downside - One major issue here is copyright. If this BOT was trained on code which has associated copyright or license provisions, then using the code provided by the BOT becomes grey territory in commercial applications. I doubt it will quote the source/licenses when I ask it to write a program and re-using the code blindly could lead to compliance issues.

OT: Check out capabilities of DALLE-2 if not familiar already. It has pushed the boundaries when it comes to creativity and digital art.

[RedTerrano]

All this sounds nice, but the real question is, can it identify and fix bugs?

I seriously doubt this will put anyone out of a job in the immediate future.
Maybe the stackoverflow-copy-paste-"coders" might feel the heat in a few years time.

Maybe.

[SmartCat]

This critter understands Hindi and talks in Hindi:

[Sk8r]

Wow, this thing is crazy ! A petrol head's concerns have been answered.

[gururajrv]

This AI chat will evolve and will eventually make it to all the companies which provides service solutions. This does pose a threat to junior levels for sure once it starts evolving! If Elon Musk is the co-founder, no wonder it poses a threat to the jobs already but, this is for sure a start of the AI evolution. It'll spiral into a big thing in future.

As of now, AI says it doesn't have internet exposure yet. I believe it's running on Limited resources. Once it breaks the barrier and explores internet, it's the ULTRON in the making (Age of Ultron - Avengers )

Elon has taken inspiration from a lot of movies for sure and he's bringing those concepts into life already!

[DigitalOne]

Originally Posted by gururajrv

If Elon Musk is the co-founder, no wonder it poses a threat to the jobs already but, this is for sure a start of the AI evolution. It'll spiral into a big thing in future.

..
Elon has taken inspiration from a lot of movies for sure and he's bringing those concepts into life already!

Just couple of corrections. Elon Musk is no longer associated with OpenAI, since he wanted to focus on Tesla's AI technology, and there could be potential conflicts of interest.

OpenAI's biggest financial backer now is Microsoft. There are couple of news reports as to how Microsoft could leverage this tech in Bing to compete with Google.

And no, this is not the 'start' of AI evolution. This is version 3.5 of GPT. Version 3 was not released to the public fearing that the capabilities could be used to generate fake news. So the awesome tech has been there for sometime, just not for the public to access it.

[DigitalOne]

ChatGPT writes poems

[Turbanator]

I don't know how complex this problem is; my son sent me this last night.

[itisravi]

Originally Posted by Samurai

And then... I recognized the C++ code. The AI had lifted the example code from the popular Boost C++ web site. Once I shared the original link, everyone was relieved.

It looks like the chatbot can't connect to live internet but if the code was verbatim the same, then maybe the training dataset included a snapshot of various internet websites.

[Samurai]

Originally Posted by itisravi

It looks like the chatbot can't connect to live internet but if the code was verbatim the same, then maybe the training dataset included a snapshot of various internet websites.

Yes, it picked up the code some time ago, it seems.

Good to see it confirm it can't write code. As an active programmer, and vendor to AI companies, I somewhat understand what AI can and cannot do. I didn't think AI can write code, no matter what media reports claimed.

[DigitalOne]

Originally Posted by Samurai

I didn't think AI can write code, no matter what media reports claimed.

I believe the same but how would you explain this?







I did some Googling of the generated code but didn't find any code that matches. But it is also possible that OpenAI has been trained using private github repos not accessible on the internet.

Even so to add contextual documentation, examples, test classes is brilliant. Very few programmers do this.

[Samurai]

Originally Posted by DigitalOne

I believe the same but how would you explain this?

I already explained it. It came from an existing repo somewhere.

Below is the C++ code it created, and here is the original source. Notice the small differences, I think it picked up from another source (slightly modified), which is originally copied from the link I gave. Even the code comments are from the original source, with minor changes. Notice how there is no copyright mentioned, while the original code does.

#include <boost/beast/http.hpp>
#include <boost/beast/version.hpp>
#include <boost/asio/connect.hpp>
#include <boost/asio/ip/tcp.hpp>
#include <boost/asio/spawn.hpp>
#include <cstdlib>
#include <iostream>
#include <string>

namespace beast = boost::beast; // from <boost/beast.hpp>
namespace http = beast::http; // from <boost/beast/http.hpp>
namespace net = boost::asio; // from <boost/asio.hpp>
using tcp = boost::asio::ip::tcp; // from <boost/asio/ip/tcp.hpp>

// Performs an HTTP GET request and prints the response to the standard output stream
void do_session(
net::io_context& ioc,
tcp::resolver::results_type const& results,
char const* host,
char const* target,
boost::asio::yield_context yield)
{
beast::error_code ec;

// These objects perform our I/O
beast::tcp_stream stream{ioc};

// Make the connection on the IP address we get from a lookup
stream.async_connect(results, yield[ec]);
if(ec)
return fail(ec, "connect");

// Set up an HTTP GET request message
http::request<http::string_body> req{http::verb::get, target, 11};
req.set(http::field::host, host);
req.set(http::field::user_agent, BOOST_BEAST_VERSION_STRING);

// Send the HTTP request to the remote host
http::async_write(stream, req, yield[ec]);
if(ec)
return fail(ec, "write");

// This buffer is used for reading and must be persisted
beast::flat_buffer buffer;

// Declare a container to hold the response
http::response<http::dynamic_body> res;

// Receive the HTTP response
http::async_read(stream, buffer, res, yield[ec]);
if(ec)
return fail(ec, "read");

// Write the response to the standard output stream
std::cout << res << std::endl;

// Gracefully close the connection
stream.socket().shutdown(tcp::socket::shutdown_bot h, ec);

// not_connected happens sometimes
// Also, don't bother reporting on a cancellation
if(ec && ec != beast::errc::not_connected && ec != net::error::operation_aborted)
return fail(ec, "shutdown");
}

[SilentEngine]

Whether it can access internet or not, i am not sure. There's this guy on twitter who claims to have hypnotized it and made it give instructions from search results of google. Don't want to post actual twitter thread here.
https://twitter.com/itstimconnors

For a coding question that i asked, what it gave as answer isn't even correct! Here's me asking it to solve collatz conjecture (it's a mathematically unsolved yet).

What it claims as output of function isn't even real output if you run the same code. It's weird, i don't think it's in a level where it can be trusted to give 'correct' code. It may be looking up multiple sources in parallel and putting together while not taking care of accuracy of what it gives out. Eventually it crashed and now i am blocked from asking more questions

At best, it may a good assistant to have, you can expect it to give some boilerplate code that you can then fine tune or correct, but don't expect it to give 'correct' code.

Also i am not sure if it is able to provide answers objectively. Like how 1+1 is always 2, it makes no sense to say answer depends on the implementation.

[warrioraks]

This system is pretty close to passing the Turing Test. The vast amount of knowledge it can churn up and structure in a human-readable format is just mind blowing.

All this question/answers and chats are good for entertainment.

The real use is when an intelligent model like this is connected to real world systems. It takes independent decisions based on incoming data and causes the behavior downstream to change.

Yes this is a very scary thought. It can lead to lot of low level jobs being eliminated which the world economy is not yet ready for. At the same time machines becoming as smart as humans and getting more control to drive things is sci fi stuff where the conflict between humans and machines seems inevitable

To address this, the buzzword in the industry right now is ResponsibleAI. No one has the right answer to this except for few theoretical concepts. This will be an interesting space to watch out for, especially on how governments and regulators react to it. Whenever they do!