Pascal is dead?

Forum > Other

Pascal is dead?

<< < (7/7)

dbannon:

--- Quote from: Bogen85 on January 15, 2023, 02:15:59 pm ---I conducted a mock job interview of it... It surpassed humans I've interviewed almost across the board in every category. (including programming, design, and troubleshooting techniques, and even demonstrated advanced skill in going through example scenarios).

--- End quote ---
However, how many humans have you interviewed who insist on googling for an answer during an interview ? The AI does not know the answers, it just knows how to find them ....

Davo

Bogen85:

--- Quote from: dbannon on January 15, 2023, 11:54:48 pm ---
--- Quote from: Bogen85 on January 15, 2023, 02:15:59 pm ---I conducted a mock job interview of it... It surpassed humans I've interviewed almost across the board in every category. (including programming, design, and troubleshooting techniques, and even demonstrated advanced skill in going through example scenarios).

--- End quote ---
However, how many humans have you interviewed who insist on googling for an answer during an interview ? The AI does not know the answers, it just knows how to find them ....

Davo

--- End quote ---

Have you playing around with this respondent? Google "answers" are the last thing it's responses read like. (At least, how a human's answers who are googling for them would read like.)

It did not do so "during the interview". It's model of collection stopped in 2021.

Yes, you can point it updated references for things. But that is often fruitless and pailful, and if often just falls back to its 2021 and prior assimilation of the data, even though you have provided it with updated material.

Also, no interviewee would be able to come up with every answer begin so perfect had they googled it.

However, as I said, this interviewee is "too perfect", and that is a major problem. It also does not validate anything it says, unless you explicitly ask it to validate everything it said.

A human googling the answers is not always going to get every answer right. The first answer on Google would often would not have been the first correct one for many of my questions.

Also, a human googling who do not know how to troubleshoot would be able to walk through detailed troubleshooting scenarios the way he did.

Also, I could have easily asked leading questions, on purpose, in order to illicit an invalid answer. I was not thinking about that when I conducted the interview.

Too "perfect"... combined with if you test it for mental competence, you could easily get it to fail.

See my leading questions earlier when I asked why Pascal overtook C++ and Rust in popularity. Those are not things you could just google for and get those kinds of answers immediately in a conversation, as iis most of what you ask it.

--- Quote from: dbannon on January 15, 2023, 11:54:48 pm ---However, how many humans have you interviewed who insist on googling for an answer during an interview ? The AI does not know the answers, it just knows how to find them ....
Davo

--- End quote ---

Knows how to find them, from a data model it has already assimilated and digested. Similar to how humans do it...
I say "Similar" because it lacks real knowledge about the what the answers really mean, and if they are valid (based on any real experiences it has had).

And it is fairly simple to try to get it to use what it knows in a way that it shows it clearly does not know anything it is talking about, if you force it make logical connections that it is clearly incapable of making.

If anything this is going to change how I do human interviews.

Not to prove if they are an AI or not, but to get a better idea of their thought processes.

When I said I would have never thought to have questioned a human who answered the questions in that way, is because for a human to answer to that way, they would have to know what they are talking about.
And AI model does not have to actually know what it is talking about.

And yes, I've done pre-screen technical interviews where it became obvious they were googling for answers. They were of course not considered. And even on the answers they "sort of got" right because the "googled for it", and too many of their answers been like that, they obviously would have been rejected.

The answers the AI give don't read like something that would would expect if the answer were just now googled for. They read, at face value, like it knows what it is writing about. And if you spend anytime with that, you see that it just an illusion.

Yes, with the correct prompts, it can illicit some very powerful completions. And when enough qualifiers are in the prompt, the completions tend to be correct, and often are.

But you have to realize, as I said before, and anyone that has played with it will attest to, it can and will just make up answers, answers that are not found on any web site.

What web site says Pascal utilizes garbage collection? (as was discussed earlier in this thread). And even if some Pascal garbage collection library/mechanism existed, it certainly would not have been in any of the top sites returned when searching for the strengths of Pascal.

dbannon:
Honestly, Bogen85, I really appreciate you extensive answers but I will respond to only the trailing issues.....

--- Quote ---
--- Quote ---Goto statements were removed from Pascal because they were considered to be too dangerous and could lead to code that was difficult to read and maintain. Goto statements can also lead to code that is difficult to debug and can cause unexpected behavior.
--- End quote ---

Please tell me who how an human interviewee would have come up with the above answer in an interview, had the interviewee googled for the answer....
--- End quote ---

Thats actually a pretty reasonable answer. By accepting (incorrectly) that goto has been removed from Pascal, the AI just needs to repeat some of the down sides of Goto. Googling for "why goto is bad", not a hard task. A superficially informed human would respond pretty much the same way but using existing background knowledge. What is clever is how the AI reformats that content to really intelligible text. I have had lots of job interviewee's who cannot put their knowledge in coherent words (mind you, I employed a few of those and never regretted it).

--- Quote ---What web site says Pascal utilizes garbage collection? (as was discussed earlier in this thread). And even if some Pascal garbage collection library/mechanism existed, it certainly would not have been in any of the top sites returned when searching for the strengths of Pascal.
--- End quote ---

Again, lots and lots of web site mention both the keywords "Pascal" and "garbage collection". - so, a system scanning for a list of keywords (that list built during training) will find a lot of hits, the AI cannot really parse the text content but assumes its appropriate to mention garbage collection and pascal in the same sentence.

If you had asked "why does Pascal not use garbage collection", it would conclude from your question the reverse. Paraphrase the same web content but make it negative.

I am not suggesting the AI is not a clever product, but try giving it a code block that won't compile because of a type error, then, maybe, some subtle error in the code's logic (I can supply lots of them from my code) ?

Honestly, I am not too excited about it programming Pascal because I don't think it will be really useful for a few years yet. At some point in time it will be, perhaps we'll write a application by describing it, method by method and feed it through an AI who will fill in the methods and sends it to the compiler. Then, the author needs go back and refine his/her descriptions or debug. If I was worried about that, I'd still be writing in assembly ...

Davo

AndI do not fear AI, I'll be dead before its rules the world. And

MarkMLl:

--- Quote from: dbannon on January 16, 2023, 09:04:57 am ---Googling for "why goto is bad", not a hard task.

--- End quote ---

Interestingly, though, Googling for "why was goto removed from pascal?" (with quotes as shown) yields not a single result.

So it looks as though the "AI" has built up some sort of inference chain from the various "goto is bad" and "should goto be removed?" and "goto removed from Java" discussions and synthesised a nonsensical result... with no attempt to check a definitive source.

MarkMLl

Navigation

[0] Message Index

[*] Previous page