GPT-4o is surprisingly good at Somali

[Webb]

It used to produce gibberish and now it's suddenly at abwaan tier

 

[Ximan iyo Xadeed]

GPT 4 0 replacing Somali linguists

 

[XuseenS]

 

[Adeer_Boomer]

Imagine what this thing will do once it's allowed to crawl the web because so far it isn't using any search engines lol.
cabsi wayne

 

[attash]

Imagine what this thing will do once it's allowed to crawl the web because so far it isn't using any search engines lol.
cabsi wayne

What do you mean by that? GPT-4o already uses most of the data available in mainstream search engines like Google.

 

[daljirkadahsoon]

What do you mean by that? GPT-4o already uses most of the data available in mainstream search engines like Google.

Nah they just increased its data set- it's not drawing from the net in real time. The cost would be prohibitive and it would be ridiculously powerful.

 

[Adeer_Boomer]

What do you mean by that? GPT-4o already uses most of the data available in mainstream search engines like Google.

If I am not mistaken, I remember reading about this when it first came out.
I think it said it's still not allowed to search the web for information due to legal issues.
I don't know if that changed with chatgpt4 tho.

 

[DivineMartial]

Yoo it even perfectly translates Somali vocab to English serving as an accurate dictionary. This is revolutionary

 

[attash]

If I am not mistaken, I remember reading about this when it first came out.
I think it said it's still not allowed to search the web for information due to legal issues.
I don't know if that changed with chatgpt4 tho.

No these LLMs have full access to most of the data on the internet, this is how they are able to produce accurate results quickly. They are actually running into a problem where they are running out of data to use because the internet simply isn’t large enough. This is predicted to severely impact AI’s development.

https://lifehacker.com/tech/ai-is-running-out-of-internet

Nah they just increased its data set- it's not drawing from the net in real time. The cost would be prohibitive and it would be ridiculously powerful.

You right, they don’t search the web in real time since the data is already fed into the model. I don’t see how this makes too much of a difference though, since they already have access to most of the internet’s data anyway.

 

[Adeer_Boomer]

No these LLMs have full access to most of the data on the internet, this is how they are able to produce accurate results quickly. They are actually running into a problem where they are running out of data to use because the internet simply isn’t large enough. This is predicted to severely impact AI’s development.

https://lifehacker.com/tech/ai-is-running-out-of-internet

You right, they don’t search the web in real time since the data is already fed into the model. I don’t see how this makes too much of a difference though, since they already have access to most of the internet’s data anyway.

It does make a difference because this means they have all the data up to a certain date. So let's say they're fed up to today, this means all the information that pours on the web from tomorrow is unavailable to it. That's billions of new information about everything that it's lacking. A problem that's unsolvable today could be solvable tomorrow so if it can search the web in real time, it's basically a beast!

 

[Kamaal]

There is hope for Hooyo mataalo folks after all.

 

[CABDULWALI XASAN.]

there was a somali guy talking to the voice one in somali And it was pretty good but it spoke in Hooyo mataalo accent with a california zing