That makes for a tough discussion unfortunately. I see a lot of value lost by having LLMs in email clients, and I dont observe the benefit; LLMs are a net time sink because I have to rewrite its output myself anyway. Proponents seem to not see any value loss, and they do observe an efficiency gain.

I am curious to see how the free market will value LLM communication. Will the lower quality, higher quantity be a net positive for job seekers sending applications or sales teams nursing leads? The way I see it either we end up in a world where eg job matching is almost completely automated, or we find an effective enough AI spam filter and we will be effectively back to square one. I hope it will be the latter, because agents negotiating job positions is bound to create more inequality, with all jobs getting filled by applicants hiring the most expensive agent.

Either way, so much compute and human capital will go wasted.

For 1, crowbars are generally available but knives and guns are heavily regulated in the vast majority of the world, even though both are used for murder as well as legitimate applications.

For 2, things get even more complicated. Eg if my router is hacked and participates in a botnet I am generally not liable, but if I rent out my house and the tenant turns it into a weed farm i am liable.

Liability is placed where it minimises perceived societal cost. Emphasis on perceived.

What is worse for society, limiting information access to millions of people or allowing csam, harrassment and shaming?

The point of giving it a name is in the second part. Its about explicitly acknowledging the limitations of medicine

Because of old age. Being vegan increased your odds threefold to die of old age instead of prematurely from disease.

Apologies for not having a link to the source

The universally shared image of the US is that they loudmouth how great they are compared to the rest of the world. So when news comes out on how the US fails so dramatically on very basic public/social services, it leaves people from outside the US stumped.

I dont know what I expected. It's a 17 year old earning 600 bucks per month on an affiliate marketing tool.

You should be careful with the meaning you ascribe to the word 'universal'. The list of universal approximators is massive, and the sub-list of universal approximators that can be trained with OLS is still substantial. Still these models can differ significantly:

- How efficient are they (in #parameters required for a certain error) for specific tasks? There is a known 'maximum efficiency' for general tasks, but in high dimensions this efficiency is terrible, such that many models will fail terribly on high-dimensional data. Hence, you should pick a model that is exceptionally good for a specific task, although it might be less efficient for other tasks.

- How well can the model cope with noise? If your dependent variable is severely distorted (think financial data) then you need a model that can balance between interpolating datapoints and averaging out the noise.

Just to name my two favorite properties. The first one is _kind of_ related to learnability, since an inefficient model is often pretty much impossible to learn.