cross-posted from: https://piefed.social/c/technology/p/1903357/should-handymen-use-ai-to-fix-your-electric-outlets-new-mozilla-project-says-yes
In an obscure post, a Mozilla MLE announced a ClawBot ripoff specifically targeted at at tradespeople (electricians, plumbers, roofers, handymen) for help with administrative tasks like scheduling, explaining invoices, etc.
What is their project, ClawBolt? According to their website, it's a compilation of
- A flaky AI agent you run on your computer
- A Telegram chatbot to remotely control it
- A connection to OpenAI to parse all of it
This sounds like a dangerous project to target at tradesmen, doesn't it? AI marketing tells people that AI is smart and safe and powerful when it is anything but. If Facebook's top AI safety genius nearly destroyed her own email inbox, how will a plumber fare?
One of the features of ClawBolt is "memory," something that will store and possibly corrupt questions and answers like "What's Mrs. Johnson's address?" and "My hourly rate is $95".
But it gets worse: ClawBolt documentation says it can scan images to "get help identifying fixtures"
Remember the memes making fun of AI's inability to make wiring diagrams?
I can only imagine how bad this could make things for tradespeople - in addition to violating customer privacy, getting gaslit by a non-deterministic "memory" machine, and potentially breaking their work computers.
AI is generally garbage, but I refuse to judge AI output until I see the actual prompt that generated it. Who knows the they actually asked for. The prompt could have been "generate an image showing how to wire a plug so that it will kill the next person that touches it".
So you'll never judge AI, because people normally don't share their prompts?
I've seen enough output from my own prompts to know that it usually produces garbage.
Judge AI cause it has generated a potentially dangerous scenario.