stochastictrebuchet

ML engineer here. My intuition says you won’t get better accuracy than with sentence template matching, provided your matching rules are free of contradictions. Of course, the downside is you need to remember (and teach others) the precise phrasing to trigger a certain intent. Refining your matching rules is probably a good task for a coding agent.

Back in the pre-LLM days, we used simpler statistical models for intent classification. These were way smaller and could easily run on CPU. Check out random forests or SVMs that take bags of words as input. You need enough examples though to train them on.

With an LLM you can reframe the problem as getting the model to generate the right ‘tool’ call. Most intents are a form of relation extraction: there’s an ‘action’ (verb) and one or more participants (subject, object, etc.). You could imagine a single tool definition (call it ‘SpeakerIntent’) that outputs the intent type (from an enum) as well as the arguments involved. Then you can link that to the final intent with some post-processing. There’s a 100M version of gemma3 that’s apparently not bad at tool calling.

Comic Code gang represent

Gotta say, it’s kind of a bummer to be downvoted for sharing my own experience. Are those ‘disagree’ or ‘doesn’t contribute to discussion’ votes?

AdGuard does more than DNS blocking. It strips ads from the response content.

Haven’t seen a single YT ad

I’ve really been enjoying Vivaldi. It’s also Chromium-based. It’s easy to customize and it has really good tab management. You can group tabs into workspaces, open split panes, and – this one I really appreciate – you can stack tabs by domain. Added bonus is that the company behind it, Vivaldi Technologies, is Norwegian, which ticks the ‘shop European’ box for me.

As for ad blocking, the shittiness of manifest v3 made me look at options outside the browser rather than rely on extensions. These days I pass all my traffic through adguard, which filters out ads from the request responses. All in all this has been a positive step, because now I can play around with any browser without ever seeing ads.

Well-deserved win! Watched this in the cinema a few weeks back. What immediately struck me about the beautiful art style is that it felt more like what you’d expect from a labor-of-love indie game than from a dreamworks/pixar studio – and it was incredibly refreshing! Also, for a movie where water plays a big role, the fluid rendering was absolutely breathtaking. I could almost smell the warm plastic air of a GPU giving its all.

Task failed successfully

One of the points the article makes is that people boost such content despite knowing it’s fake because it confirms what they’re ’feeling’. Want to feel outrage? Here’s an image that will let you and others feel that. Truth? Irrelevant.

In short: it’s the ‘facts don’t care about your feelings’ crowd doing what they do best: recasting reality as a jumble of vague feelings.

Wait, what happened to LinkedIn?