Calvin French-Owen

Doing the actual refactor was probably 25% of the time investment here. When I saw the GitHub launch post about the new UI I started having a few ideas about how it could be improved, so I started hacking on it in Photoshop. Usually these kinds of things just end up in a trash folder somewhere, but I was digging where it was going so I kept at it.

As for writing the article, we actually like to encourage each other to write interesting articles on our personal blogs. One thing to keep in mind is that you'll be most productive when you're inspired to write, so if I get inspired to write something mid-day, unless there's a super important Segment.io deadline I'll just take a detour for an hour and write down my thoughts. And we try to remain open to each other doing that, instead of seeing it as "wasted" time. @calvinfo did the same with his recent article on Node's new streams too: http://calv.info/an-introduction-to-nodes-new-streams/ — since they are timely topics we like to get them out as soon as possible.

It's not without personal gain either, there are definitely traffic benefits, although I haven't looked into how much traffic actually gets from my own site to our Segment.io's site, let alone signs up. There are also hiring benefits I assume (we're only just starting to hire Javascript folks) from putting yourself out in the community.

Mostly I just enjoy it, so I do it ;) I usually write these after 8pm or so when I switch from working on Segment.io to doing…

I think this just further demonstrates the truth behind the truly small & scrappy teams culture at OpenAI that an ex-employee recently shared [1].

Even with the way the presenters talk, you can sort of see that OAI prioritizes speed above most other things, and a naive observer might think they are testing things a million different ways before releasing, but actually, they're not.

If we draw up a 2x2 for Danger (High/Low) versus Publicity (High/Low), it seems to me that OpenAI sure has a lot of hits in the Low-Danger High-Publicity quadrant, but probably also a good number in the High-Danger Low-Publicity quadrant -- extrapolating purely from the sheer capability of these models and the continuing ability of researchers like Pliny to crack through it still.

[1] https://calv.info/openai-reflections

> Their attention apears to already be quite diffuse if buying a python package manager is an item anywhere on their agenda.

According to the blog [0], their whole monorepo is in Python, their models are obviously trained using Python, their experiments are written using Python and core and CLI of their Codex is written using Rust. Uv brings both Python and Rust expertise. You’re talking nonsense because of your blind hate of LLMs. Even though I agree that they’re capitalizing on the fear of SWE being redundant.

0 - https://calv.info/openai-reflections