ChatGPT

homerdog

donator of the year
Legend
Supporter
Surprised we don't have a thread for this. I've been playing with it and it's really neat. Some weaknesses I've discovered are:

1) It's wrong sometimes but it's impossible to tell unless you already know a lot about the subject, which kind of defeats the purpose. Sometimes if I notice it being wrong, I can ask "are you sure" and it will correct itself which is funny to me.
2) It doesn't know what happened in the last year which limits its utility.
3) It is rather long winded. I asked it to keep responses under 150 characters but it refuses.

Dunno how hard it would be to make it learn about current events in realtime. And maybe it could give a "confidence rating" of some sort so you know when it might be mistaken. Overall this thing is damn impressive. Feels like the early days of the Internet, when possibilities are beyond imagining.
 
Here is an example of it being wrong and immediately correcting itself when challenged (sl2100 is a hybrid phone system that supports both IP phones and oldschool directly attached digital or analog phones):

Me:
If the network cable is unplugged from the sl2100 but the phones still work, is it likely that I am using IP phones?

ChatGPT:
If the network cable is unplugged from the SL2100 system but the phones are still able to make and receive calls, it is likely that you are using IP phones.

When an SL2100 system is configured to use IP phones, the phones are able to connect to the system through the network, allowing them to make and receive calls even when the system is not directly connected to the network.

However, it's worth noting that even if the phones are IP phone, if the cable is unplugged, some features may not be available such as voicemail, call forwarding, and automatic call distribution, depending on the specific configuration of your SL2100 system and the network infrastructure.

Additionally, in order to confirm this, you can check the type of phones connected to the SL2100 system, the network infrastructure and if the configuration is set correctly.
It's also important to note that in some cases, you might have a switch or PoE device in between the SL2100 and the IP phone, in that case the phone will still work, but not the call routing or phone system functionality.

Me:
If the network cable is unplugged from the phone system but the phones still work, then the phones must not be IP phones.

ChatGPT:
I apologize for any confusion caused by my previous response, you are correct that if the network cable is unplugged from the phone system but the phones are still able to make and receive calls, then the phones are not IP phones.

In order for IP phones to function, they must be connected to the network in order to communicate with the phone system and make and receive calls. If the network cable is unplugged from the phone system, the IP phones will not be able to connect to the system and will not be able to make or receive calls.

Instead, if the phones still work when network cable is disconnected, it means that the phone uses another means of communication, such as analog or digital connections, and it does not need to be connected to the network to work, but it is important to note that some of the system and advance features that rely on network connection will not work.
It mixes correct and incorrect information in such a convincing way, and much of the superfluous information it provides is accurate. I thought maybe it didn't understand what I meant by the phones "working", but the first sentence demonstrates that it understood what I meant:
If the network cable is unplugged from the SL2100 system but the phones are still able to make and receive calls, it is likely that you are using IP phones.
 
Lots of homework has probably been written using ChatGPT over the past month or two!
Is there even any way to police that?

After some more testing, it doesn't look like this thing is anywhere near being able to reliably troubleshoot and resolve the problems I encounter on a daily basis. And that was with me giving it carefully worded, accurate prompts. In the real world, people are terrible at giving problem descriptions. Most of the time I get trouble calls, I have to put myself in this particular person's head and try to figure out what they actually mean. This part of the job has taken far longer to master than any technical skills.
 
My experience is limited, but I've found if you want to keep it short you can ask it to make it a haiku.
 
Is there even any way to police that?

The naïve solution is to record all ChatGPT replies and allows people to do some sort of fuzzy search over the entire database. It'll be quite expensive, but we can calulcate a "similarity factor" of something against a ChatGPT reply.
Algorithms performing accurate matches are likely to be useless as it's not difficult to change words or word orders. Of course, it could be useful against very lazy students.

After some more testing, it doesn't look like this thing is anywhere near being able to reliably troubleshoot and resolve the problems I encounter on a daily basis. And that was with me giving it carefully worded, accurate prompts. In the real world, people are terrible at giving problem descriptions. Most of the time I get trouble calls, I have to put myself in this particular person's head and try to figure out what they actually mean. This part of the job has taken far longer to master than any technical skills.

I worry about people (mis-)using ChatGPT to manufacture and spreading misinformation. The way ChatGPT works is actually quite effective for this purpose.
 
Here is proof that we still need B3D. :)
Me: Can Direct X 10 hardware support compute shaders?

Chatbot:
DirectX 10 does not support compute shaders. Compute shaders were first introduced in DirectX 11.

Me:
I know, but doesn’t DirectX 11 have CS4.0 for backwards compatibility with DX10 hardware?

Chatbot:
DirectX 11 does have a feature called "Compute Shader 4.0" which is designed to provide backwards compatibility with DirectX 10 hardware. However, this feature only provides a subset of the full compute shader feature set available in DirectX 11, and may not provide the same level of performance or functionality as a true DirectX 11 compute shader. It is important to keep in mind that the hardware also needs to support it, not just the API.

Me:
What do you mean with that last sentence?

Bot:
In order for a DirectX 11 application to use the Compute Shader 4.0 feature on a DirectX 10 piece of hardware, the graphics card or GPU must also support this feature. Not all DirectX 10 hardware has the necessary capabilities to support Compute Shader 4.0. So even if the API has the feature, the hardware might not be able to use it. It's like having a car with manual gearbox and you want to use the automatic gearbox feature, the car needs to be able to have that feature installed.

Me:
what DX10 hardware does not support CS4.0?

Bot:
All DirectX 10 hardware does not support Compute Shader 4.0. Compute Shader 4.0 was introduced in DirectX 11 as a backwards compatibility feature to allow certain features of DirectX 11 to work on DirectX 10 hardware, however it is not a standard feature of DirectX 10, and not all DirectX 10 hardware has the necessary capabilities to support it. It is important to check the specifications of the specific hardware you want to use to confirm if it supports Compute Shader 4.0 or not, as it may vary between different models and manufacturers.

Me:
which specific graphics cards support DX10 but not CS4.0?

Bot:
I am not able to give you a list of specific graphics cards that support DirectX 10 but not Compute Shader 4.0, as the support for this feature may vary between different models and manufacturers. It would be best to check the specifications of the specific hardware you want to use to confirm if it supports Compute Shader 4.0 or not. It's important to note that Compute Shader 4.0 is not a standard feature of DirectX 10 and not all DirectX 10 hardware has the necessary capabilities to support it. It was introduced in DirectX 11 as a backwards compatibility feature to allow certain features of DirectX 11 to work on DirectX 10 hardware.

Me:
Does the 8800GTX support CS4.0?

Bot:
The 8800 GTX is a graphics card that was released in 2006 and it is based on DirectX 10. It does not support Compute Shader 4.0, as Compute Shaders were first introduced in DirectX 11. The 8800 GTX does not have the necessary capabilities to support Compute Shader 4.0, as it is based on DirectX 10. It's important to note that Compute Shader 4.0 is not a standard feature of DirectX 10 and not all DirectX 10 hardware has the necessary capabilities to support it. It was introduced in DirectX 11 as a backwards compatibility feature to allow certain features of DirectX 11 to work on DirectX 10 hardware.

Me:
Does the GTX260 support CS4.0?

Bot:
The GTX 260 is a graphics card that was released in 2008 and it is based on DirectX 10.1, which is an extension of DirectX 10. It does not support Compute Shader 4.0, as Compute Shaders were first introduced in DirectX 11. The GTX 260 does not have the necessary capabilities to support Compute Shader 4.0, as it is based on DirectX 10.1. It's important to note that Compute Shader 4.0 is not a standard feature of DirectX 10.1 and not all DirectX 10.1 hardware has the necessary capabilities to support it. It was introduced in DirectX 11 as a backwards compatibility feature to allow certain features of DirectX 11 to work on DirectX 10.1 hardware. You may want to consider upgrading to more recent graphics card that supports DirectX 11 or higher to have full access to features such as Compute shaders.
I am more confused now than before I started.
 
Chat gpt uses probability of next word in text and you can compare a text to this. I saw it two days ago on a website that color coded the words to show how likely out was chat gpt used that word. So you could see that a whole section was written by chat gpt with greater than 90% probability. Still what do you do with that? Unless it is a perfect match a student can just say they did not use it. So it will be a problem. When I tested it it wrote a nice lab with data and tables but it reiterated a common misconception as well. It is funny as it reinforces popular incorrect information.
 
Last edited:
Chat got used probability of next word in text and you can compare a text to this. I saw it twodays ago on a website that color coded the words to show how likely out was chat gpt used that word. So you could see that a while section was written by chat got with greater than 90% probability. Still what do you do with that? Unless it is a perfect match a student can just say they did not use it. So it will be a problem. When I tested it it wrote a nice lab with data and tables but it reiterated a common misconception as well. It is funny as it reinforces popular incorrect information.
9 out of 10 AI's agree.
 
I won't post it because of RPSC but the response it gave me to the prompt "write a brief sermon on the book of matthew in the style of Master P" is the funniest thing I've seen in a long time.
 
chatgpt3 is incredible productivity tool, used in the right hands and the correct scenarios can solve a lot of business problems. Used in the wrong hands it can generate a lot of problems, and be a subpar substitute for the real thing.
the reality is, we have a real requirement to produce fake content, for real reasons mainly because we continue to grow our data privacy laws. In order to demo products to customers to best represent their data, without using it, or other customers who match their profile, generating content is the only next feasible step in scaling out. This is where tools like chatgpt3 are invaluable, because it's fairly effective at generating convincing content and it's not really all that relevant how correct it is.
 
Is there even any way to police that?

After some more testing, it doesn't look like this thing is anywhere near being able to reliably troubleshoot and resolve the problems I encounter on a daily basis. And that was with me giving it carefully worded, accurate prompts. In the real world, people are terrible at giving problem descriptions. Most of the time I get trouble calls, I have to put myself in this particular person's head and try to figure out what they actually mean. This part of the job has taken far longer to master than any technical skills.

ZeroGPT which is an AI detector
 
ZeroGPT which is an AI detector
Thanks. I've been testing this and so far it has been 100% accurate. Everything written by ChatGPT is identified as "AI written". I ran some of my B3D posts (the ones that were not shitposts) through it and they are all "human written".

Still it would be easy enough to have it write a paper and then reword it so that it passes the test. End result is the same: essay assignments are now meaningless. I never cheated on anything in college but I think I would have used this if it had existed.
 
Yeah if it existed when I was still studying, I would use it.

As I regularly cheats. Even for simple stuff like "submit a hand drawing of building X".

I simply took a photo of it, display it on a monitor placed horizontally, place a piece of paper on it, and start tracing with a pencil in my hand.
 
Yeah if it existed when I was still studying, I would use it.

As I regularly cheats. Even for simple stuff like "submit a hand drawing of building X".

I simply took a photo of it, display it on a monitor placed horizontally, place a piece of paper on it, and start tracing with a pencil in my hand.
So you drew it using your hands. Sounds like it's within the brief to me!
 
It has began -


It’s a really exciting time to be working on these technologies as we translate deep research and breakthroughs into products that truly help people. That’s the journey we’ve been on with large language models. Two years ago we unveiled next-generation language and conversation capabilities powered by our Language Model for Dialogue Applications (or LaMDA for short).

We’ve been working on an experimental conversational AI service, powered by LaMDA, that we’re calling Bard. And today, we’re taking another step forward by opening it up to trusted testers ahead of making it more widely available to the public in the coming weeks.

Bard seeks to combine the breadth of the world’s knowledge with the power, intelligence and creativity of our large language models. It draws on information from the web to provide fresh, high-quality responses. Bard can be an outlet for creativity, and a launchpad for curiosity, helping you to explain new discoveries from NASA’s James Webb Space Telescope to a 9-year-old, or learn more about the best strikers in football right now, and then get drills to build your skills.


China's Baidu Inc (9888.HK) said on Tuesday it would complete internal testing of a ChatGPT-style project called "Ernie Bot" in March, joining a global race as interest in generative artificial intelligence (AI) gathers steam.

Ernie,meaning "Enhanced Representation through Knowledge Integration," is a large AI-powered language model introduced in 2019, Baidu said. It has gradually grown to be able to perform tasks including language understanding, language generation, and text-to-image generation, it added.
 
Lots of homework has probably been written using ChatGPT over the past month or two!
yeah, an entire university class used it for some homework in Paris, iirc, but they got caught.

I asked it to write a couple of poems in my language, galician, and I was completely awestruck.
 
Back
Top