cyphereal

While I agree the autoregressive training is poor at calibrating a model, it's clear that LLMs encode a kind of world model to be able to reason, do mathematics, programming, and logic. They are certainly imperfect, but the emergent internal representations go beyond mere "what text comes next" even though LLMs typically simply use tokens.

By analogy to your 'Measuring confidence' section: a function estimator (neural nets can be good at this) and a bunch of past observations could be pretty calibrated at making weather predictions. An LLM has a wild mix of past observations and random collated opinions from previous text (it might mention months, recency of rain etc.. etc...) so lands somewhere between 'text completion' and 'estimator' with a model of unknown fidelity. It is not 'merely' generating a plausible sentence.

I would never use an LLM for accurate forecasting, and certainly agree they are not calibrated, but I would not dismiss them out of hand.

YouTubeMusic actually has this. With its "You Music Tuner" there are a lot of configurable parameters that control artist variety and music discovery. It doesn't quite nail tuning by curation, but it's a step in the right direction.

I have a Mazda 2020 CX30. The display is not a touch screen and using CarPlay with the physical selector drives me nuts (it's also dangerous).

I like physical controls for everything else, but please make displays for e.g. CarPlay touch screens.

The ideal transport is be a bicycle, or an e-bike for cities with hills. The problem is that bike lanes are still not great. I bike commuted in SF for years. Bikes are ideal "last mile" transport too, for long commutes trains with bike carriages can be added in. Numerous cities in Europe show this works very well. Having said that, it's an ideal which I fear is probably far from reachable for most people given the way cities have been built, and infrastructure priorities in the US do not favour bikes.

I can - telling CodeGPT what I'd like it to do.

Sure he warned of dangers, but also invested heavily in AI for autonomous driving by hiring DL researchers ("a million robot taxis by 2020").

He also invested in OpenAI until he had a falling out about who would be in charge of the company. He recently founded X.AI, which is supposedly also building generative AI.

Everyone sees danger, but many, including Musk, are forging ahead.

Interesting. Has MS in fact turned the corporate culture around? When I was there, the Windows org was full of fiefdoms with, at times, petty alpha nerds running things. The culture at Google, while far from perfect, was more respectful and data-driven.

Most westerners are not calorie starved. You can get cheap calories everywhere (e.g. calorie-bombs of expiring chicken covered in oil and salt, roasted and sold super cheap to get rid of them ASAP). Reductio ad absurdum - just eat palm oil. I would argue it's not a good metric.

It's not the ID, it's that a phone number is relatively difficult to obtain (not hard, but still a limited resource) and unique. A crypto stake / wallet / something of value on the blockchain could easily meet such a criteria if well designed.

The article is very hand-wavy about the core rent seeking argument. There is much discussion about factors such as productivity improvements concentration of market power and mergers and acquisitions.

From TFA: "says Blankenburg, “the data show very clearly that the means used to obtain these profits cannot be reduced to the use of productive technologies.” Other mechanisms, such as lobbying or mergers and acquisitions, the authors find, have played a significant role in enhancing the market power of dominant companies. “You can show quite clearly how surplus profits increase with mergers and acquisitions, or how changes in regulation that favor control over intellectual property rights for large corporations have a pretty-instant impact on the profit performance of those companies,"

That's the sum total of the rent seeking argument. There is nothing in there demonstrating use of IP to extract rent (which, arguably isn't even rent seeking behaviour, but for the sake of argument may be granted here). My intuition is that profit extraction due to IP for the very large companies is most obvious in pharmaceuticals, but that's just my guess?

There are a few interesting points in this article, but it's weakly written.