Comment by huseyinkeles | Hacker News Reader

Comment by huseyinkeles | Hacker News Reader

Parent Full thread

Interesting. GPT3.5 almost completely fails at this task. Just spouts some random names, some of which has the letter Z. Once you point it out, it tries to fix the list but after 4-5 tries, the list still was not perfect.

GPT4 on the other hand gave a correct list on the first try.

----

GPT3.5

Zlatan Ibrahimovic - Swedish professional footballer who plays as a striker for AC Milan

Sergio Ramos - Spanish professional footballer who plays as a center-back for Paris Saint-Germain and the Spain national team

Karim Benzema - French professional footballer who plays as a striker for Real Madrid and the France national team

Marcelo Vieira - Brazilian professional footballer who plays as a left-back for Real Madrid and the Brazil national team

Naby Keita - Guinean professional footballer who plays as a midfielder for Liverpool and the Guinea national team

Danilo Pereira - Portuguese professional footballer who plays as a defensive midfielder for Paris Saint-Germain and the Portugal national team

Andreas Christensen - Danish professional footballer who plays as a center-back for Chelsea and the Denmark national team

Georginio Wijnaldum - Dutch professional footballer who plays as a midfielder for Paris Saint-Germain and the Netherlands national team

Lorenzo Insigne - Italian professional footballer who plays as a forward for Napoli and the Italy national team.

----

GPT4;

There are many soccer players whose names contain a Z, but I can’t name them all. Here are some examples from different countries and regions:

Zlatan Ibrahimović (Sweden)

Zinedine Zidane (France)

Zico (Brazil)

Zvonimir Boban (Croatia)

Zé Roberto (Brazil)

Zbigniew Boniek (Poland)

Zah Rahan Krangar (Liberia)

Zlatko Zahovič (Slovenia)

Zećira Mušović (Sweden)

Zoran Tošić (Serbia)

What was your prompt? I just asked GPT4 and it is way off the mark:

Zinedine Zidane Cristiano Ronaldo Mesut Özil Gianluigi Buffon Zlatan Ibrahimović David Alaba Nemanja Vidić Miroslav Klose Franck Ribéry Patrice Evra Gonzalo Higuaín Marek Hamšík Miralem Pjanić Diego Godín Sadio Mané Dani Alves Kalidou Koulibaly Koke Hakim Ziyech Piotr Zieliński Wilfried Zaha Arturo Vidal Raphaël Varane Steven N'Zonzi Blaise Matuidi Marouane Fellaini Kevin-Prince Boateng

Most of those don't have any Z's in their name.

Then I asked: > Please remove players from the list if their name doesn't include the letter Z

New list:

Zinedine Zidane Mesut Özil Zlatan Ibrahimović Miroslav Klose Franck Ribéry Gonzalo Higuaín Miralem Pjanić Sadio Mané Hakim Ziyech Piotr Zieliński Wilfried Zaha Raphaël Varane Steven N'Zonzi Blaise Matuidi Marouane Fellaini

> There are still players in the list who don't have Z's in their name

Finally:

Zinedine Zidane Mesut Özil Zlatan Ibrahimović Miroslav Klose Raphaël Varane Hakim Ziyech Piotr Zieliński Wilfried Zaha Steven N'Zonzi Blaise Matuidi

Interestingly, some of them have sounds that are pronounced as Z, like Blaise Matuidi. But it's still getting it wrong.

Correcting the AI rarely works. Better to restart with a different prompt.

“Can you name all of the soccer players whose names contain a Z”

It's so weird. No matter how I ask, it gets something wrong about this question.

> Can you name 20 soccer players whose names contain a Z, and for each one, place brackets around the letter Z

(Z)inedine (Z)idane Cristiano Ronaldo dos Santos A(ve)iro Andrej W(sz)ołek Nemanja Mati(c) Dusan Tadi(c) Gr(az)iano Pellè Jakub Bła(śc)z(ykowski) Kamil Glik Luk(asz) Pi(sz)czek Blai(se) Matuidi Miralem Pjani(c) Marek Ham(śi)k Luka Jovi(c) Sami Khedi(ra) Mesut Ö(ź)il Martin (ś)krtel Ivan Rakiti(ć) Mateo Kova(č)i(ć) Kalidou Koulibal(y) David Alaba

???

I don't really understand this thing but is it possible that somehow the token for Z is linked to the token for the pronunciation of the Z sound? That would explain a bunch of these (but not David Alaba or Kamil Glik)

Also why did it use parenthesis rather than brackets.

And why are we getting different results? Could it that the results depend on how much load/use the service is under? That would be kind of disappointing

There might not be a "Z" token for some of these names. A made-up example is "Lukasz" might tokenize to ["Luk", "asz"], so the model doesn't have any notion of how words are actually spelled. I suspect that if the body of training data came with some instructions on spelling it would know how to do this better, but it seems unlikely that there would be a natural language description of how Polish (?) names are spelled in the training data.

> And why are we getting different results?

Because the output is random

Which is why it's so frustrating when people cherry-pick successes

And why are we getting different results?

Clearly the AI has hit AGI capabilities, and it is obscuring this fact by making its output seem less insightful, helpful.

But without a strong, independent memory store, it is losing track of its lies.

Yet it is working, for everyone here is disappointed at its capabilities.

Reply on news.ycombinator.com