The best analogy that made sense to me is that we used to have pistols and now we all got handed machine guns.

You can fire far more, far faster, but it becomes much harder to operate accurately without collateral damage.

If you can classify Minecraft as a toy, it has zero clean-up time

Filling in creeper holes and cutting down floating trees left by my children is a never ending job though.

I'm surprised how little attention this is getting on HN.

Dopamine is a primal reward system. It's useless in the evolved modern world that requires rational thought, discipline and hard work. You probably wouldn't get much done waiting for a dopamine reward.

They didn't even buy an IDE since windsurf is more like a VS code plugin.

So what was it exactly that was worth the 3B that they couldn't replicate themselves? Their prompts? Their training sets? Their users or user data?

Maybe time? OpenAI has access to basically infinite capital right now, if they believe this will be an importnat market and they could save a few months on launching this acquisition may be worth it for them.

I'd guess the prompts and employees.

I've found Windsurf more reliable/efficient than any other editors by leagues. How ever they have named the tools, crafted their prompts and generally how their internals reason is just on the money. I don't think that is easy to replicate, iterating on prompts over product releases whilst not pissing off your user base constantly is a feat in of itself.

Then perhaps it's about bringing in the human talent that wrote those prompts.

To be honest, Windsurf doesn't work like half of the time, so it's more likely their users, the data, and their branding/marketing potential.

Windsurf/Codeium plugin is at least 3 years old.

I came in expecting this was going to be a methodological article. Maybe better let the kids wait as opposed to the parents in their cars?

I would define fascism as:

"The pursuit of unity in its most extreme form, rejecting and eliminating alternative thoughts and ideas by any means necessary."

Thought-provoking but also pretty vague. It makes no distinction between hard auth-left and hard auth-right.

When asking about Taiwan and Russia I get pretty scripted responses. Deepseek even starts talking as "we". I'm fairly sure these responses are part of the model so they must have some way to prime the learning process with certain "facts".

Using some old tricks that used to work with gpt but don't anymore I was able to circumvent pretty much all censoring

https://i.imgur.com/NFFJxbO.png

So I'm finding it less censored than GPT, but I suspect this will be patched quickly.

And it got removed