pick your AI programmer friend π€:
'gpt-engineer' - @antonosika
'smol-dev' - @swyx
'AutoGPT' - @SigGravitas
'Metamon' - @yoheinakajima
they commonly "work" this way:
βΈ You give a first set of instructions
βΈ AI asks clarifying questions, generates spec, writes code, π
π
'gpt-engineer' - @antonosika
'smol-dev' - @swyx
'AutoGPT' - @SigGravitas
'Metamon' - @yoheinakajima
they commonly "work" this way:
βΈ You give a first set of instructions
βΈ AI asks clarifying questions, generates spec, writes code, π
π
but do they actually work?
Yes. No. Maybe so.
how can we know?
we can define a set of challenges, use them as a benchmark, and even practice challenge-driven-development
some of the makers of these AI programming friends are doing this
e.g. @Auto_GPT:
Yes. No. Maybe so.
how can we know?
we can define a set of challenges, use them as a benchmark, and even practice challenge-driven-development
some of the makers of these AI programming friends are doing this
e.g. @Auto_GPT:
gpt-engineer made a couple of simple programming tasks and uses them as the benchmark
attached is a snapshot (2023-June-19)
e.g.
gpt-engineer can quickly develop a currency_converter for you,
but can't properly code a pomodoro_timer (yet!)
attached is a snapshot (2023-June-19)
e.g.
gpt-engineer can quickly develop a currency_converter for you,
but can't properly code a pomodoro_timer (yet!)
gpt-engineer reached 15KβοΈ in github in less than a week!
A breakdown of how it works by @tfglaser :
linkedin.com
A breakdown of how it works by @tfglaser :
linkedin.com
@yoheinakajima seems to be on a roll, and made several agent-like projects
most famous is the @babyAGI_
and just recently, Metamon
I didn't see how we can try out with Metamon -- @yoheinakajima??
most famous is the @babyAGI_
and just recently, Metamon
I didn't see how we can try out with Metamon -- @yoheinakajima??
If you read threads on twitter/discord/github,
you see developers trying these projects and generally sharing this feedback:
βΈ The potential is very exciting
βΈ Sometimes work for [trivial] projects, and saves you a lot of time (at least creates a lot of boilerplate for you)
1/
you see developers trying these projects and generally sharing this feedback:
βΈ The potential is very exciting
βΈ Sometimes work for [trivial] projects, and saves you a lot of time (at least creates a lot of boilerplate for you)
1/
2/
βΈ For most projects you still need to know what you are doing
βΈ Requires a continuous feedback loop between the developer and AI
βΈ Mostly suited to start new projects --> but I assume that new projects will aim to address this pain and work better with an existing codebase
βΈ For most projects you still need to know what you are doing
βΈ Requires a continuous feedback loop between the developer and AI
βΈ Mostly suited to start new projects --> but I assume that new projects will aim to address this pain and work better with an existing codebase
Read smol-dev instructions or watch the demo video (by @swyx) to get a pretty realistic view of what can be achieved these days
Loading suggestions...