I'm pleased to present, for free, the first hour of my new course Understanding AI-Assisted Software Development on YouTube!
This course is designed to be agnostic to any one AI tool, and instead provide a mental model and best practices for you and your team. 👇
Devs: in the age of AI you need to differentiate with good user experience, know how to talk to your users, and build product taste and sense...to do that you should be getting familiar with Nielsen Norman Group (if you aren't already).
Seriously: https://t.co/HmQRGJB7TA
Zero LLM is an effort to take some time without an LLM.
1) Agree to some time without AI. Do things yourself.
2) Start a session to track your time.
3) When you feel the addictive itch to ask AI for help, go to the site instead.
4) End your session and share with #0llm.
I had this idea. We are using LLMs too much. We need a recalibration sometimes. A bit of time where you agree with yourself to do things manually, keep your skill and your voice.
Check out 0llm. If you use it, share it with the hashtag #0llm.
A problem with AI output is what we’ll call “effort bias”. Because an output (text, image, video, etc.) looks like things that have historically taken a lot of human effort and thinking, we assume a degree of correctness.
When someone posts an AI take, and that post is obviously AI generated, that means that person isn’t very good at prompting their AI.
So you probably shouldn’t worry about their take.
Words matter. Let's stop calling human evaluation of AI output a "bottleneck." It's human review *tempering* AI output.
This isn't holding back a wave of progress, it's stacking sandbags as a storm surge hits.
@mattpocockuk I built a JS course that 10s of thousands of devs have taken for over a decade. If you ask AI to learn JS, it uses a lot of my stuff. It is what it is, but “financial incentive” feels weird to me here. At least in part, AI is other people’s courses.
If we define software engineering as “building reliable software by avoiding unnecessary complexity” instead of “writing code”…suddenly AI replacing engineers feels a lot less likely.
@jherr And this answers the question "do you need to know how to code". If your compiler occasionally produced incorrect assembly, you would need to know assembly to debug it.
That's LLMs. For coding they're the ultimate leaky abstraction.
It's interesting to watch devs who couldn't be bothered to stop making div soup and write accessible HTML suddenly call themselves "product engineers" who worry about the user.
I'm incredibly excited to announce I've joined Nielsen Norman Group (@NNgroup) as a Senior Experience Specialist. I'll be helping shepherd curriculum and content for their new self-paced courses.
This role sits at the intersection of my entire career and I couldn't be happier.
@kentcdodds Ironically, in some ways you're saying that devs should care about the things that UXers have been fighting and struggling to get their dev teams to care about for decades.
@mattpocockuk I've been iterating on my own spec approach that is similar to this. I find it also gives a place to teach newer devs systems thinking.
I've been calling it "declarative modeling": spec the system but not the implementation details. https://t.co/ibB3jwxYl5