Frontiers in Psychology,
Год журнала:
2024,
Номер
15
Опубликована: Окт. 8, 2024
Introduction
The
recent
advances
of
large
language
models
(LLMs)
have
opened
a
wide
range
opportunities,
but
at
the
same
time,
they
pose
numerous
challenges
and
questions
that
research
needs
to
answer.
One
main
are
quality
correctness
output
LLMs
as
well
overreliance
students
on
without
critically
reflecting
it.
This
poses
question
in
educational
tasks
what
teachers
need
consider
when
using
for
creating
items.
In
this
work,
we
focus
characteristics
conceptual
items
developed
ChatGPT
user-generated
improvements.
Methods
For
purpose,
optimized
prompts
created
30
kinematics,
which
is
standard
topic
high-school
level
physics.
were
rated
by
two
independent
experts.
Those
15
received
highest
rating
included
survey.
dimensions
designed
align
with
ones
most
commonly
used
concept
inventory,
Force
Concept
Inventory
(FCI).
We
administered
together
FCI
172
first-year
university
students.
results
show
medium
difficulty
discriminatory
index
overall
exhibit
slightly
lower
average
values
FCI.
Moreover,
confirmatory
factor
analysis
confirmed
three
model
closely
aligned
previously
suggested
expert
model.
Results
discussion
way,
after
careful
prompt
engineering,
thorough
selection
fully
automatically
generated
ChatGPT,
able
create
had
only
than
carefully
human-generated
procedures
select
such
high-quality
set
require
efforts
point
towards
cognitive
demands
demonstrate
human
oversight
or
student
interviews
necessary
one-dimensional
assessments
distractors
students'
difficulties.
Smart Learning Environments,
Год журнала:
2024,
Номер
11(1)
Опубликована: Дек. 19, 2024
Abstract
Effectively
supporting
students
in
mastering
all
facets
of
self-regulated
learning
is
a
central
aim
teachers
and
educational
researchers.
Prior
research
could
demonstrate
that
formative
feedback
an
effective
way
to
support
during
learning.
In
this
light,
we
propose
the
application
Large
Language
Models
(LLMs)
guide
towards
problem-solving
through
feedback.
We
present
LEAP,
novel
platform
utilizes
advanced
LLMs,
such
as
GPT-4o.
LEAP
empowers
with
ability
effectively
pre-prompt
assign
tasks
LLM,
resulting
stimulates
students’
cognitive
metacognitive
processes,
thereby
enhancing
systematic
prompt
design
can
provide
wide
range
types
scaffolds
students.
These
scaffolds,
which
are
rooted
research,
include
sense-making,
elaboration,
self-explanation,
partial
task-solution
well
motivational
scaffolds.
Through
approach,
emphasize
critical
importance
synchronizing
technological
advances
empirical
theoretical
frameworks.
This
alignment
potentially
ensures
positive
LLMs
landscape.
Many
machine
learning
models
used
in
academia
and
industry
that
identify
organic
compounds
typically
lack
the
ability
to
converse
over
prompts
results,
also
require
expertise
across
a
number
of
steps
obtain
answers.
The
purpose
this
study
was
primarily
gain
insight
into
advantages
current
unmodified
state
art
Large
Multimodal
Models
(LMMs)
several
containing
multiple
spectra
varying
difficulty
evaluate
impact
training
data,
reasoning,
speed.
These
readily
available
easy
use
software
for
identification
an
compound
based
on
molecular
formula
were
found
be
reproducible
three
similar
LMMs.
To
author's
best
knowledge,
marks
first
time
GPT
variants
each
able
correctly
quinoline
using
variety
different
spectroscopic
images.
results
obtained
2-step
process
consisting
a)
Uploading
high
resolution
spectral
images,
b)
Submitting
text
prompt
with
images
requested
determination.
main
findings
1)
Four
LMMs
provided
rationale
step-by-step
interpretations
1H-NMR,
13C-NMR,
3
DEPT-NMR
from
Prompt
A,
2)
Three
these
LMMs,
led
by
GPT-5
preview
model,
combined
correct
chemical
structure
3)
Two
achieved
top
score
5/5
generating
sequential
explanations
reflecting
order
along
most
explanations.
