Connectionists: Statistics versus “Understanding” in Generative AI.

[Brad Wyble]

The fact that prompting tricks sometimes manage to improve performance is
not really evidence of understanding though.  On the contrary, this is
exactly how you would expect a massive memory bank with unknown contents
and unknown indexing to function. You throw phrases at it like incantations
and some of them happen to trigger the kind of output that you wanted.

I agree that the fact that LLMs do what they do is amazing and impressive
and they are even useful for some cases.  But the understanding is an
illusion.  Our minds always want to believe that artifacts are sentient
because it plays into our metacognition about agency and theory of mind.
LLMs hit the sweet spot for this belief.


On Mon, Feb 19, 2024 at 7:35 PM Iam Palatnik <iam.palat at gmail.com> wrote:

> The fact that answers are so dependent on word choice
>
> Just to clarify, I was not referring to the prompt of the task itself.
> That is, whether we tell chatgpt to 'add 2 and 2' or 'give me 2+2'.
> I'm referring to the fact that adding something like 'think step by step
> <https://arxiv.org/pdf/2205.11916.pdf>' onto the task prompt has a
> noticeable effect on the performance for various tasks.
> And it doesn't have to literally be 'think step by step' exactly.
>
> An example that happened just the other day: a friend was trying to get
> ChatGPT to write a 30 paragraph long text, and was having no luck. The
> model always wrote the wrong number of paragraphs. I told him to try
> including 'number the start of each paragraph to help yourself' to the
> prompt and bingo. I had never seen someone trying to do this exact task,
> nor this numbering trick. I came up with it on the spot, and it worked, but
> why? Why should we expect the model to perform better with this, at all?
>
> Do examples like this entail understanding or lack thereof? I'm not sure
> everyone would agree on the answer.
> But regardless of the specific wording we would use to describe what
> happened there, it would be a problem if we are 'absolutely sure' ChatGPT
> can't do X because it can't understand anything, while it turns out it can
> do X if you just change the prompt or tool access.
>
> On Mon, Feb 19, 2024 at 6:13 PM Brad Wyble <bwyble at gmail.com> wrote:
>
>> The question is not whether we need to rely on external tools to verify
>> the  line lengths (we do), but whether we can understand that this illusion
>> affects us, and can understand the basic idea of such an illusion. And we
>> can do both of these. You can explain to someone that this illusion exists
>> and they are able to understand that there is a difference between what
>> they think their eyes are telling them about line length and what is
>> actually true in the world.
>>
>> As far as we can determine from GPT experiments, LLMs are not able to
>> reason about their own limitations in this way.  That is the fundamental
>> distinction I'm referring to.
>>
>>
>>>
>>> Because the performance of the LLMs on some of these tests seem to
>>> depend so much on how the questions are formulated and what tools they are
>>> given to respond with, I still tend to think that they understand
>>> something. I'm OK with the idea that their understanding has space to be
>>> much deeper, still, too.
>>>
>>>
>>>
>> This is the opposite of the argument that is typically used though.  It
>> is specifically the face that their answers depend so much on phrase (e.g.
>> that paper by Melanie Mitchell that recently went by) that we argue they do
>> NOT understand.  The fact that answers are so dependent on word choice is
>> an indication that they are parroting remembered examples rather than
>> understanding in a human sense.
>>
>>
>>
>>
>>
>>>
>>>
>>>
>>>
>>>
>>> On Mon, Feb 19, 2024 at 1:50 PM Brad Wyble <bwyble at gmail.com> wrote:
>>>
>>>> Iam, the difference is that while you may need an external source to
>>>> remember all 50 states, for the ones that you have remembered/looked up,
>>>> you are able to verify that they do or do not contain specific letters
>>>> without reference to a resource, or writing some code to verify it.  It is
>>>> even worse that if you push them on their mistakes, they are still unable
>>>> to correct.
>>>>
>>>> A better counterargument to the example Dave provides is that perhaps
>>>> LLMs just cannot ever break things down at the letter level because of
>>>> their reliance on tokens.  Humans can do this of course, but a good analogy
>>>> for us might be the Muller Lyer illusion, which is essentially
>>>> impenetrable to our cognitive faculties.  I.e. we are unable to force
>>>> ourselves to see the lines as their true lengths on the page because the
>>>> basis of our representations does not permit it.  This is perhaps similar
>>>> to the way that LLM representations preclude them from accessing the letter
>>>> level.
>>>>
>>>> However, I think a good counterpoint to this is that while people are
>>>> unable to un-see the Muller Lyer illusion, it is not that difficult to
>>>> teach someone about this blindspot and get them to reason around it, with
>>>> no external tools, just their own reasoning faculties.  LLMs seem unable to
>>>> achieve this level of self-knowledge no matter how patiently things are
>>>> explained.  They do not have the metacognitive faculty that allows them to
>>>> even understand their blindspot about letters.
>>>>
>>>>
>>>>
>>>>
>>>> On Mon, Feb 19, 2024 at 10:06 AM Gary Marcus <gary.marcus at nyu.edu>
>>>> wrote:
>>>>
>>>>> Correct; also tool integration has actually been less successful than
>>>>> some people believe:
>>>>>
>>>>>
>>>>> https://open.substack.com/pub/garymarcus/p/getting-gpt-to-work-with-external
>>>>> <https://open.substack.com/pub/garymarcus/p/getting-gpt-to-work-with-external?r=8tdk6&utm_campaign=post&utm_medium=web>
>>>>>
>>>>> On Feb 19, 2024, at 5:49 AM, Thomas Trappenberg <tt at cs.dal.ca> wrote:
>>>>>
>>>>> 
>>>>> Good point, but Dave's point stands as the models he is referring to
>>>>> did not even comprehend that they made mistakes.
>>>>>
>>>>> Cheers, Thomas
>>>>>
>>>>> On Mon, Feb 19, 2024, 4:43 a.m. <wuxundong at gmail.com> wrote:
>>>>>
>>>>>> That can be attributed to the models' underlying text encoding and
>>>>>> processing mechanisms, specifically tokenization that removes the spelling
>>>>>> information from those words. If you use GPT-4 instead, it can process it
>>>>>> properly by resorting to external tools.
>>>>>>
>>>>>> On Mon, Feb 19, 2024 at 3:45 PM Dave Touretzky <dst at cs.cmu.edu>
>>>>>> wrote:
>>>>>>
>>>>>>> My favorite way to show that LLMs don't know what they're talking
>>>>>>> about
>>>>>>> is this simple prompt:
>>>>>>>
>>>>>>>    List all the US states whose names don't contain the letter "a".
>>>>>>>
>>>>>>> ChatGPT, Bing, and Gemini all make a mess of this, e.g., putting
>>>>>>> "Texas"
>>>>>>> or "Alaska" on the list and leaving out states like "Wyoming" and
>>>>>>> "Tennessee".  And you can have a lengthy conversation with them about
>>>>>>> this, pointing out their errors one at a time, and they still can't
>>>>>>> manage to get it right.  Gemini insisted that all 50 US states have
>>>>>>> an
>>>>>>> "a" in their name.  It also claimed "New Jersey" has two a's.
>>>>>>>
>>>>>>> -- Dave Touretzky
>>>>>>>
>>>>>>
>>>>
>>>> --
>>>> Brad Wyble
>>>> Professor of Psychology
>>>> Penn State University
>>>>
>>>>
>>
>> --
>> Brad Wyble (he/him)
>>
>>

-- 
Brad Wyble (he/him)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20240219/1161689d/attachment.html>