Profile avatar
ianbicking.org
Software developer in Minneapolis. Working with applications of LLMs. Previously: Mozilla, Meta, Brilliant.org
212 posts 449 followers 757 following
Regular Contributor
Active Commenter
Posts 12 Comments 38

Liberal democracy won't stay alive very long if it doesn't start co-opting technology and business, rather than alienating them. They're the two most important vectors of power today. They are actually synergistic -- the rule of law is vital for business. Public education is vital for tech.

submitted 5 days ago • 10 comments

I couldn't come up with any Ians off the top of my head, but after searching Ian Miles Cheong seems like a solid choice

submitted 17 days ago • 2 comments

Whoever designed the new Bosch dishwasher controls should feel very badly about themselves. But I guess with some pride... the controls are indecipherable, unresponsive, and fight with the user in new and unique ways

submitted 17 days ago • 1 comment

The article has a really good discussion of process-based evaluation approaches. It only talks a little about AI, but steps back and thinks more about the goals of evaluation; something we need to do a lot more of when thinking about how to adapt to AI.

submitted 19 days ago • 1 comment

Watching my kid do some pygame tutorials, and encountering sprite sheets... and this seems like such a terrible data format? The optimization of combining images seems very old-fashioned. No consistent metadata. Animation frames and distinct sprites get smashed together. Just seems weird.

submitted 22 days ago • 0 comments

I've been using GPT o1 personally, but haven't found it compelling for use in pipelines and APIs. But now o3-mini is actually cheaper than 4o and I guess I should reconsider... It's not exactly drop-in though, and I don't have an intuition if it's actually a good idea...

submitted 23 days ago • 0 comments

ChatGPT Advanced Voice Mode often interrupts with an error message that it "cannot respond due to guidelines" at bizarre times. These must be applied at a different layer, as ChatGPT itself is largely unaware of the error message and certainly has no concept of what the violation might be...

submitted 24 days ago • 1 comment

I'm designing some tools for use by an LLM. Tools aren't as fancy as people make them out to be: it's really just a description of a structured response the LLM can give. Then the system that calls the LLM can interpret that however it wants, perhaps but not necessarily "executing" the tool. 🧵

submitted 24 days ago • 1 comment

I sometimes see people quote another post that's blocked, but I don't have anyone blocked. I might subscribe to a blocklist (is that a thing?) that I've forgotten about. But I get no info on who the person blocked is. Or has the person blocked me? Who would even care enough to block me?

submitted 24 days ago • 2 comments

At first I disabled my Twitter account, but that felt incredibly passive. So I deleted all my posts and replies. Now I have an empty account, but what should I put there?

submitted 26 days ago • 1 comment

I was talking about AI with my daughter, and mentioned it lying about doing tasks, and she was insistent that AI couldn't lie. Which of course it can... I tried doing a test in chatgpt but was admonished that it "can’t provide intentional misinformation". But firing up the playground...

submitted 29 days ago • 1 comment

cicada— a love song 🎶

submitted 32 days ago • 1 comment