I see why OpenAI is worried, but this needs to be taken into context. A lot of Deepseek's bright claims are likely a result of management accounting instead of an equal horse race between OpenAI and them

Meanwhile, all of OpenAI's capx costs are allocated to building O1. The buildings, all the chips, cooling... they need to buy everything new. For example, cooling systems can cost between $5-$10 mil alone for a data centre, let alone the cost of the building and all the new chips.

The developers of DeepSeek are bitcoin miners who used their spare compute to develop this, so all of their capx costs are already sunk costs, and allocated to bitcoin mining. Their only costs allocated to this project may be R&D and electricity costs. Continued below...

I can also see this being helpful in identifying key demographics to design policy for and market to. Demographics that are grey but have many orange lines connected to them are likely prime targets to be persuaded to get on-side with the NDP. #BCNDP #cdnpoli #bcconservatives

I can zoom in to see the demographics. Other than looking nice, this has helped me understand support levels better. For example, management, counterintuitively, were NDP supporters in another model. Here, I see they connect to many supportive demographics like having master's degrees.

Hey Kim, long time no see!

The size of the bubbles represents their mean proportion of ridings. For example, women are a large bubble because they make up around 50% of ridings.

This is a different method from my previous ones, so while many demographics are in the same spot as my previous tweets, some have moved. For example, real-estate workers have moved from the conservative camp to the NDP camp. Fist Nations have moved from NDP to unaligned.

I want to add that the individual variables aren't statistically significant in this one. I plan to make a more accurate model later, but it will take a lot of effort so I'm holding off.