tylermarques

I've been using Opus 5.5 to write openscad models for me for the past week, with great success. I wanted to expand the project Open Source Watch Winder [0] to have 8 watches, with gears instead of pulleys, and in little back and forth I have a version I've started printing. [1]

[0] https://github.com/mwood77/osww [1] https://imgur.com/a/bWent0K

Yes they definitely do - not claiming it's a perfect solution, but as a V1 product it shows a lot of promise.

We had early access and found it to be pretty useful. Having a second form of verification, where you can ask multiple questions (in the form of Nouls) raised our confidence in the outputs of other models. [0] IMHO This type of model works incredibly well in concert with LLMs, not as a replacement.

[0] https://goodstartlabs.com/research/verification-is-the-bottl...

I've had a lot of success combating this by adding "All summaries need to adhere to ASD-STE100 Simplified Technical English standards" [0] which I discovered from another HN thread [1]

[0] https://www.asd-ste100.org/ [1] https://news.ycombinator.com/item?id=49114639

There is some evidence.[1] The best reviewer is a different model with fresh context, worst is same model with same context.

1. https://arxiv.org/pdf/2603.04582

In the same vein, we recently released a version v0.1 of our humor benchmark. [1] We use human answers from a cards against humanity style game call Bad Cards [2] as ground truth for what is funny. The models get to choose a card from a hand of 3-6 cards, so not quite de novo joke creation.

[1] https://goodstartlabs.com/leaderboards/lol-arena

[2] https://bad.cards/

Sorry about that! We've toned down the music a bit, trying to put more emphasis on the narrator. Thanks for the feedback :)

We're hoping to create more like this! :)