HACKER Q&A
📣 hsikka

Building a Better Clippy?


I'm fascinated with using ML/DL to build useful features that augment work and productivity, and for my master's thesis at Georgia Tech I'm trying to build a virtual assistant for student learners.

Do you think Clippy, the microsoft office assistant, could have been much more useful if it had learned behaviors rather than rule based decision making?

Do you think this is an interesting opportunity? Is there anything out there like this?


  👤 kleer001 Accepted Answer ✓
I would look at IDEs, specifically feature requests that are the same across platforms, the same in every forum from Eclypse to Amethyst 2. Specifically features that don't tend to be implemented because they're just a bit too complex to be rule based.

IMHO an agent is the wrong way and discovery (like in Emacs) or a status/hint line is the way to go. Keeping it unobtrusive enough not to be annoying, but still present enough to be helpful when needed. AKA be more like Alfred from Batman and less like Ace Ventura.


👤 probinso
Amiga Learning claims they use Deep Learning to leverage AV processing to measure stress of students to inform restructuring first-readers curriculum of children.

I think Charlse River Analytics is doing dynamic dialog systems as well, for guiding through experimental design

I think this is also the strategy that Knewton uses for their product


👤 muzani
I think Google Assistant is awesome. But it seems that its advantage over Siri is rule based.

Word doesn't seem great for digital assistants, but I think what really needs help is documentation. There's often a good deal of bad documentation, and it's something an assistant could learn to navigate.


👤 notahacker
You might find this interesting: https://medium.com/@saranormous/clippy-s-revenge-39f7387f9aa...

TLDR Clippy was probably ahead of its time. Since then the tech to understand behaviours and NLP inputs has got better, but more importantly the median users have got increasingly used to typing queries rather than browsing or reading manuals to get fast results, and spends a lot more time reading and responding to instant messages. I don't know about student learning specifically, but conversational AI assistants for commercial use are approaching the peak of the hype cycle.