Agree. However I feel that there may still be many priors that are useful in practice, like decomposing the vision task into a hierarchy of sub problems, but as a human - I could be biased.
Two things I would like - advances in detectors or generative content that do not do c2pa, and more transparency in what the usage policy means in practice.
I am very uncomfortable with this being released commercially without the requisite defence against misuse being also accessible. If we didn't have a problem with deepfakes, spam, misleading media before, we surely are now. All leading AI organisations are lacking here, benefiting from the tech but not sufficiently attacking the external costs that society will pay.
Intuitively, the popularisation of FaaS feels inevitable as lower level technical challenges get solved and give way to abstractions like this. Who knows though what our stack will look like in 20 years.
I'm not a fan of the personification here, but certainly someone emerging from a cave after 5 years would think Meta and OpenAI's attitude towards open-source should be totally swapped.
I like that model. Seemingly, some companies like Palantir are special on more than one of your dimensions - but those are likely super rare in practice.
I listen, though (or because?) most of the topics are irrelevant to me. I do find it entertaining, and there is at least some insight being shared. I think you need to know what to listen to and what to ignore; be skeptical of pretty much any rationalization David Sacks makes of his ideological views, question some of Chamath's hypocritical moral grand-standing (how does SPAC grubbiness not stick to him?), ignore Jason's insecurities etc. Fundamentally though, they are four well-connected, intelligent people talking about issues that do matter in an entertaining format.
The paper positions these purpose-built models, that explicitly decompose spatial reasoning tasks into sub-tasks, as better than these huge end-to-end models that do everything, at least in terms of interpretability and generalization. I am partial to that argument; my intuition is that the tighter the specification for a task, the better the model can be - because training objectives are clearer, data can be cleaner, models can be smaller, and so on. I feel like that is how my brain works, at least for more complex tasks. However, I do wonder if this is because I naively still want to be able to understand what the model is doing and how is does it, in a symbolic way - when that simply won't lead to the best empirical results.
People who can't afford to buy get stuck with borrowing costs being passed on, anyway, and get trapped in rent spirals. I do agree that some of the criticism of Lowe is undeserved, though.
There are so many threads here. The reserve bank governor is deeply unpopular in the broader public because of 2021 comments he made about there being no interest rate increases needed until 2024 [1]. As mentioned in the article, many companies are posting record profits, including supermarkets [2] - leaving low and middle income people hurting, and working more hours or side jobs. Those same people also face an adversarial rental housing market [3]. Competition, at least in this market, seems to be exerting very little downward pressure on prices for commodity goods. Buy-now-pay-later schemes continue to damage the poorest in our society - even big banks now make it easy to take on bad debt [4].
There are many challenges in Australia, but also opportunities to make things better.