[Tutor] LLM comparison

[dn]

On 19/01/24 02:55, Alan Gauld via Tutor wrote:
> On 15/01/2024 19:12, dn via Tutor wrote:
> 
>> Both of the 'Expert solutions' contained rather 'frilly bits' - perhaps
>> an attempt to convince us of Mastery rather than a true demonstration
>> thereof?
> 
> But remember that none of the current set of AI bots are designed
> to prioritize finding "best" answers, nor even "correct" answers,
> only relevant ones and presenting these in a naturalist way that
> makes it look like a machine didn't do it.

Indeed, often expressed with a great deal of confidence.
(conversely, the apologies when errors are pointed-out are often amusing)


> Judging AI bots on the technical quality of their answers is
> really to judge the internet web sites that they scavenged
> the answers from.

+1

Many of us will have code-repos of incomplete projects - including code 
which won't execute (in its current state of development). Yet, GitHub 
was used as source-material.

I like the description "stochastic parrot". An LLM can only repeat words 
(tokens) it has been taught, using advanced statistics to associate 
these in a combination that could?should be meaningful.


The relevance 'here', is that too many learners are using these tools 
without being aware of such limitations, traps, and snares - often 
seduced by the confident air, and by successful early experience (at the 
intro-level).

Like the previous tactic of copying from StackOverflow, the value of an 
LLM's particular contribution can only be assessed if one first 
possesses sufficient knowledge 'between the ears'!

Will a trainee gain such competence, using a GAI?

-- 
Regards,
=dn