Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Is the reversal curse in LLMs real?

andrewmayne.com

Post reply on HN
Print view View original

141–150 of 211 posts

Re: Is the reversal curse in LLMs real?

#141

Post by Kiro »

RandomLensman wrote: post #138
Kiro wrote: post #137

Earlier quoted context omitted.

He was acting Chancellor. It's like saying a red apple is not an apple. It's a technicality. The fact that we have a completely reasonable explanation to why GPT-4 includes him in the list and you guys still make it sound like the AI is a complete moron is just hilarious to me.

There is no "Acting Chancellor", that is a poor translation of what the job is. There is only a Chancellor, not an "Acting Chancellor" - you have to trust that the Germans and their institutions kind of know who was Chancellor and who was not. The machine is a complete moron for not being able to get basic data from basic sources right.

> Scheel became acting Chancellor

> Chancellor of Germany, Acting, 7 May 1974 – 16 May 1974

https://en.wikipedia.org/wiki/Walter_Scheel

Re: Is the reversal curse in LLMs real?

#142

Post by RandomLensman »

Kiro wrote: post #141
RandomLensman wrote: post #138

Earlier quoted context omitted.

There is no "Acting Chancellor", that is a poor translation of what the job is. There is only a Chancellor, not an "Acting Chancellor" - you have to trust that the Germans and their institutions kind of know who was Chancellor and who was not. The machine is a complete moron for not being able to get basic data from basic sources right.

> Scheel became acting Chancellor > Chancellor of Germany, Acting, 7 May 1974 – 16 May 1974 https://en.wikipedia.org/wiki/Walter_Scheel

I know, but the German wikipedia gets it right: https://de.wikipedia.org/wiki/Walter_Scheel

(And so does the German Chancellor: https://www.bundeskanzler.de/bk-de/kanzleramt/bundeskanzler-...)

There is zero ambiguity about who was Chancellor and who was not.

Re: Is the reversal curse in LLMs real?

#143

Post by MPSimmons »

I had a discussion with ChatGPT 4 on this (quasi-ironically) to see what it "thought", and it theorized that BERT might have had a better outcome to these conditions because it's actually bidirectionally trained, rather than GPT, which is forward-trained.

https://chat.openai.com/share/daff08c5-ddea-4ad4-963a-0df88e...

Re: Is the reversal curse in LLMs real?

#145

Post by Kiro »

RandomLensman wrote: post #142
Kiro wrote: post #141

Earlier quoted context omitted.

> Scheel became acting Chancellor > Chancellor of Germany, Acting, 7 May 1974 – 16 May 1974 https://en.wikipedia.org/wiki/Walter_Scheel

I know, but the German wikipedia gets it right: https://de.wikipedia.org/wiki/Walter_Scheel (And so does the German Chancellor: https://www.bundeskanzler.de/bk-de/kanzleramt/bundeskanzler-... ) There is zero ambiguity about who was Chancellor and who was not.

It says "geschäftsführender Bundeskanzler".

Even on https://en.wikipedia.org/wiki/Chancellor_of_Germany he's listed as "Vice Chancellor Walter Scheel served as acting Chancellor from 7 May to 16 May 1974" between 4 and 5.

There's obviously some ambiguity to it considering we are two humans discussing this with a claimed discrepancy between English Wikipedia and German Wikipedia, but your conclusion is still that the AI is spitting nonsense.

Re: Is the reversal curse in LLMs real?

#146

Post by OJFord »

RandomLensman wrote: post #131
dragonwriter wrote: post #68

Earlier quoted context omitted.

> Even when allowed to surf the web, ChatGPT gets it wrong: > Who was the eighth Federal Chancellor of the Federal Republic of Germany? - Gerhard Schröder Its probably counting Walter Scheel, who was Acting Chancellor in 1974, and you are probably not. When I asked ChatGPT to list the chancellors in order and identify the eighth, it listed eight ending in Schröder, with Scheel and his ten day Acting Chancellor tenure…

Walter Scheel was never Chancellor, he was fulfilling the duties of the Chancellor while being Foreign Minister, but he was not a Chancellor. So counting him is wrong.

> Walter Scheel was never Chancellor, he was fulfilling the duties of the Chancellor while

I think this is really an issue of semantics & translation, because that is just what 'acting X' means? (Admittedly not always while still doing something else too, but I don't see that as significant - if it had gone on long perhaps a junior minister in the foreign office would've been named acting FM in his place too.)

Re: Is the reversal curse in LLMs real?

#147

Post by JKCalhoun »

famouswaffles wrote: post #17
gipp wrote: post #8

I fall somewhere on the skeptic side of the LLM spectrum. But this "flaw" just does not seem to have the force that its proponents seem to think it does, unless I'm missing something significant. Simply because in the context of natural language (Rather than formal logical statements), "A is B" does not imply "B is A" in the first place. "Is" can encompass a wide variety of logical relationships in colloquial usage,…

I think a lot of people unwittingly think of the training process as "smart". Similar criticisms are "well there are many descriptions of game x it would have read so why doesn't it play x well". But gradient descent is a dumb optimizer. Training itself is not actually like someone reading a text anymore than evolution is like someone thinking about the best way to augment an organism. a "smart" optimizer would look…

I'm LLM-ignorant, but my layman brain wonders if the logical relationships belong in a kind of as-of-yet-not-invented "pre-processor" that feeds into the LLM.

(Perhaps though it has to be done during training so that "Who was the ninth Chancellor of Germany" becomes knowable in the first place.)

Re: Is the reversal curse in LLMs real?

#148

Post by RandomLensman »

Kiro wrote: post #145
RandomLensman wrote: post #142

Earlier quoted context omitted.

I know, but the German wikipedia gets it right: https://de.wikipedia.org/wiki/Walter_Scheel (And so does the German Chancellor: https://www.bundeskanzler.de/bk-de/kanzleramt/bundeskanzler-... ) There is zero ambiguity about who was Chancellor and who was not.

It says "geschäftsführender Bundeskanzler". Even on https://en.wikipedia.org/wiki/Chancellor_of_Germany he's listed as "Vice Chancellor Walter Scheel served as acting Chancellor from 7 May to 16 May 1974" between 4 and 5. There's obviously some ambiguity to it considering we are two humans discussing this with a claimed discrepancy between English Wikipedia and German Wikipedia, but your conclusion is still that the…

There is zero ambiguity under German as law as to who was a Chancellor and who was not. Not a matter of wikipedia etc.

The way to be Chancellor is through article 63 of the Grundgesetz, while Scheel was put into the caretaker role via article 69. This explains it a bit https://de.wikipedia.org/wiki/Vizekanzler_(Deutschland) - Scheel was only taking on the function, not the office.

This kind of giving some machine the benefit of the doubt when in fact there is zero ambiguity is really a path that makes me think we will have mostly machines designed for marketing and other non-critical things.

Re: Is the reversal curse in LLMs real?

#149

Post by OJFord »

maaaaattttt wrote: post #100
saliagato wrote: post #88

Earlier quoted context omitted.

It can’t count because what the LLM sees is a bunch of tokens not words

Exactly. And asking what/who/which is the “Ninth” something will always fail (or randomly succeed)

Much more likely to work for something like German chancellors, US presidents, etc. that people do actually number and could easily have been trained on that data. If you ask 'who was the 45th US president' it doesn't need to be able to count a list of US presidents to answer 'Trump', you can just duckduckgo that and get the answer anyway, that's not doing any counting either.

Re: Is the reversal curse in LLMs real?

#150

Post by RandomLensman »

OJFord wrote: post #146
RandomLensman wrote: post #131

Earlier quoted context omitted.

Walter Scheel was never Chancellor, he was fulfilling the duties of the Chancellor while being Foreign Minister, but he was not a Chancellor. So counting him is wrong.

> Walter Scheel was never Chancellor, he was fulfilling the duties of the Chancellor while I think this is really an issue of semantics & translation, because that is just what ' acting X' means? (Admittedly not always while still doing something else too, but I don't see that as significant - if it had gone on long perhaps a junior minister in the foreign office would've been named acting FM in his place too.)

That is not semantics. The law is very clear: Scheel was not a Chancellor (and indeed he is not counted as one in Germany: https://www.bundeskanzler.de/bk-en/chancellery/federal-chanc...).

Article 69 Grundgesetz has a sort of "Caretaker Chancellor" that has the function but - importantly - not the office. They way to have the office is through article 63.

It works differently in Germany to the US, for example, where the vice president could become the president, while the vice chancellor only ever gets the function, not the office (unless through a proper vote for Chancellor).

When there is actually ground truth, machines should be able to recover that, not take weird turns.

Post reply on HN