Circle and Berkeley Utilize AI for Blockchain Transactions with TXT2TXN

Circle
and
Blockchain
at
Berkeley
have
unveiled
TXT2TXN,
an
innovative
open-source
web
application
that
leverages
Large
Language
Models
(LLMs)
to
streamline
blockchain
transactions
by
interpreting
user
intents
from
natural
language
inputs,
according
to

circle.com.

Introduction
to
TXT2TXN

TXT2TXN,
which
stands
for
text-to-transaction,
aims
to
simplify
the
user
experience
in
crypto
applications
by
using
AI
to
translate
freeform
English
text
into
actionable
blockchain
transactions.
This
tool
allows
users
to
specify
their
desired
actions
in
natural
language,
which
the
LLM
then
interprets
and
converts
into
signed
intents
for
on-chain
execution.

Understanding
User
Intents

The
concept
of
intents
refers
to
a
user’s
expression
of
their
desired
outcome
without
detailing
the
specific
steps
needed
to
achieve
it.
Most
existing
blockchain
applications
operate
on
a
transaction-based
architecture,
where
users
must
specify
each
step
of
the
transaction
process.
In
contrast,
an
intent-based
architecture
allows
users
to
define
what
they
want
to
achieve,
leaving
the
application
to
determine
how
to
accomplish
it.

Circle
and
Berkeley’s
research
highlights
that
intents
can
simplify
complex
interactions,
making
blockchain
technology
more
accessible.
For
instance,
instead
of
manually
setting
up
a
token
swap,
a
user
can
simply
state
their
intention,
and
the
system
will
handle
the
intricate
details.

LLMs
in
Action

Large
Language
Models
play
a
crucial
role
in
this
new
architecture.
By
interpreting
natural
language
inputs,
LLMs
can
classify
user
intents
and
convert
them
into
executable
transactions.
This
reduces
the
complexity
of
user
interfaces,
allowing
users
to
focus
on
their
desired
outcomes
rather
than
the
technical
steps
required.

For
example,
a
user
might
input
a
command
like
“send
1
USDC
on
Ethereum
to
kaili.eth,”
and
the
LLM
will
classify
this
as
a
transfer
intent.
The
application
then
processes
this
intent
and
generates
a
transaction
payload
to
be
executed
on
the
blockchain.

The
Prototype
and
Its
Functionality

The
TXT2TXN
prototype
uses
OpenAI’s
GPT-3.5
Turbo
to
interpret
user
intents
in
two
stages.
First,
it
classifies
the
input
into
a
specific
intent
type,
such
as
transfer
or
swap.
Second,
it
fills
in
the
necessary
details
to
create
a
transaction
payload
or
a
signed
order,
which
the
user
can
then
execute.

Circle
and
Berkeley
have
implemented
a
simple
frontend
where
users
can
input
their
desired
actions
in
a
textbox.
The
backend
processes
these
inputs,
leveraging
the
LLM
to
determine
the
appropriate
intent
and
generate
the
necessary
transaction
details.

Accuracy
and
Future
Work

Accuracy
is
a
critical
concern
when
using
LLMs
for
blockchain
transactions,
given
the
irreversible
nature
of
most
blockchain
operations.
Preliminary
tests
with
a
small
array
of
prompts
have
shown
promising
results,
but
further
research
is
needed
to
ensure
reliability
and
minimize
errors.

Future
developments
for
TXT2TXN
include
expanding
the
range
of
supported
intent
types,
enhancing
accuracy
through
advanced
learning
techniques,
and
integrating
more
stateful
features
like
personal
address
books.
These
improvements
aim
to
make
blockchain
technology
even
more
accessible
and
user-friendly.

Conclusion

TXT2TXN
represents
a
significant
step
forward
in
the
integration
of
AI
and
blockchain
technology.
By
simplifying
user
interactions
and
leveraging
the
power
of
LLMs,
Circle
and
Berkeley
are
paving
the
way
for
more
intuitive
and
efficient
crypto
applications.
As
this
technology
evolves,
it
promises
to
make
blockchain
transactions
more
accessible
to
a
broader
audience.

Acknowledgements

This
project
was
a
collaborative
effort
between
Blockchain
at
Berkeley
and
Circle
Research.
The
team
included
Niall
Mandal,
Teo
Honda-Scully,
Daniel
Gushchyan,
Naman
Kapasi,
Tanay
Appannagari,
Adrian
Kwan
from
Berkeley,
and
Alex
Kroeger
and
Kaili
Wang
from
Circle.

Image
source:
Shutterstock