ChatGPT without limits?

[Nerd_1991]

I've just used ChatGPT online (chat.openai.com). It's limited what kind of questions it's allowed to answer in my country, Denmark. Is there a PC version of ChatGPT that will answer any question you throw at it?

[Triple G]

I guess any version is limited in how able it is to answer. What it does good is self-advertising itself.

"I´m just a mere AI, with all the knowledge You can think of, to deliver the most accurate answers there are, but in some cases it might read like the horoscope. This is a user error. Please ask a different question."

[kilésengati]

Dolly is a FOSS implementation of a large language model comparable to Chat-GPT.

https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html

[sleeps]

kilésengati eredeti hozzászólása:

Dolly is a FOSS implementation of a large language model comparable to Chat-GPT.

https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html

interesting. maybe i should use it

[Nerd_1991]

kilésengati eredeti hozzászólása:

Dolly is a FOSS implementation of a large language model comparable to Chat-GPT.

https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html

Interesting and thanks. I'm gonna try it out but.. Dolly was trained on a much smaller language model of only six billion parameters versus 175 billion for GPT-3. So I won't get my hopes too high

[sleeps]

Nerd-1991 eredeti hozzászólása:

kilésengati eredeti hozzászólása:

Dolly is a FOSS implementation of a large language model comparable to Chat-GPT.

https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html

Interesting and thanks. I'm gonna try it out but.. Dolly was trained on a much smaller language model of only six billion parameters versus 175 billion for GPT-3. So I won't get my hopes too high

i think it's better to just wait.

[Devsman]

Eh, AI Chatbots aren't that great. I've tried chatgpt and Bard, and while chatgpt is marginally better, they both have the same problem: they don't actually know what they're talking about.

All it really knows is how to read and form a sentence. I was trying to debug some code and asked it a question, and it gave me a wrong answer, and when I explained why it was wrong, it gave me another wrong answer. After a third wrong answer, it gave me the first wrong answer again.

I tried to tell it a joke and it basically said "I've heard this one before" and explained why it's funny.

I asked it why they make Louis walk in Suikoden III (he's a little kid who is some kind of aide to a group of knights, and even though he travels cross country with them, he's the only one who doesn't get to ride a horse) and while it did successfully name another character from the game, it gave me a completely different story about a character (named Louis) who apparently learned to walk, describing the scene as "an emotional part of the story."

I tried to play 20 questions with it, and after insisting that the word it was thinking of starts with B, but doesn't start with a letter that comes before C alphabetically, it ended up asking me questions about MY word that I hadn't come up with yet until I told it it won.

I'm sure most people just go there and shoot the breeze with it, and for those purposes, I'm sure it's more than adequate, because there are plenty of examples they can scrape from the Interwebz of people talking to each other about the weather.

But as a tool, it has a long way to go. I think what they'll have to do is give it some sort of model of the things they're actually describing. Because it's clear that right now all they're doing is forming connections between words and spitting them out with no clue of what they mean. Kind of like those pattern matching puzzles for kids with the colored shapes. Now, it's incredibly complex pattern matching, and is remarkably impressive to be sure. But it breaks down completely when you ask it something even a little unexpected.

I have no idea if that's even possible, honestly--to model real objects and concepts. But we'll see.

[sleeps]

wow. that's a lot. i read all of those and i understand that. but talking with the chatbot is so much fun. yeah there's an issue and i'm aware of that. they don't know what they're talking and what worse is in my opinion they are very confident with what they said.

personally it's a fun thing to talk to but not the most reliable way either and it's better to take it at a grain of salt and not blindly follow what the chatbot said

[Schindler's Lifts]

Devsman's Comet eredeti hozzászólása:

Eh, AI Chatbots aren't that great. I've tried chatgpt and Bard, and while chatgpt is marginally better, they both have the same problem: they don't actually know what they're talking about.

All it really knows is how to read and form a sentence. I was trying to debug some code and asked it a question, and it gave me a wrong answer, and when I explained why it was wrong, it gave me another wrong answer. After a third wrong answer, it gave me the first wrong answer again.

I tried to tell it a joke and it basically said "I've heard this one before" and explained why it's funny.

I asked it why they make Louis walk in Suikoden III (he's a little kid who is some kind of aide to a group of knights, and even though he travels cross country with them, he's the only one who doesn't get to ride a horse) and while it did successfully name another character from the game, it gave me a completely different story about a character (named Louis) who apparently learned to walk, describing the scene as "an emotional part of the story."

I tried to play 20 questions with it, and after insisting that the word it was thinking of starts with B, but doesn't start with a letter that comes before C alphabetically, it ended up asking me questions about MY word that I hadn't come up with yet until I told it it won.

I'm sure most people just go there and shoot the breeze with it, and for those purposes, I'm sure it's more than adequate, because there are plenty of examples they can scrape from the Interwebz of people talking to each other about the weather.

But as a tool, it has a long way to go. I think what they'll have to do is give it some sort of model of the things they're actually describing. Because it's clear that right now all they're doing is forming connections between words and spitting them out with no clue of what they mean. Kind of like those pattern matching puzzles for kids with the colored shapes. Now, it's incredibly complex pattern matching, and is remarkably impressive to be sure. But it breaks down completely when you ask it something even a little unexpected.

I have no idea if that's even possible, honestly--to model real objects and concepts. But we'll see.

A few days ago, I tried to play chess with chatgpt. I'd tell chatgpt my moves and it'd respond with it's own moves. It worked for a bit, but after 5-6 turns, it already started making illegal moves, illogical moves, or forget about moves I already made several turns ago. I could probably play a full match if I constantly course correct it, but I just don't have that kind of time at the moment.

Right now, it's just a fun curiosity (or a useful shortcut for students writing essays, depending on who you ask). But I imagine it'll get a lot better several years from now.

[PsychoPsyops]

As long as it’s owned by a company, it will probably be restricted and censored. These companies also took political correctness based restrictions too far on them, restricting humor, natural opinions and objective reasoning. It is quite sad really. No one seems to see where this will inevitably go. We are doomed.

[Devsman]

Schindler's Lifts eredeti hozzászólása:

(or a useful shortcut for students writing essays, depending on who you ask).

Now, that's pretty convenient use case. I asked it for a list of papers on a pretty specific topic, and it gave me what looked like a pretty good list. I haven't read any (lol) but from the titles and synopses chatgpt gave, it seemed like exactly what I was looking for.

Trying to Google the same thing was a nightmare.

[sleeps]

ya know, i actually said to chatgpt about chatbots as a source and why it's not the most reliable way to talk to. only to get bombed by a wrong answer on why and how shodan took over citadel