With high degrees of automation that need not be the case. Consumer electronics is highly automated right now, where-as clothing production is currently not.
Few customers care about proofs. If you can measure how well the method work for the desired task, that is most cases sufficient and in many cases preferred over proofs.
There are many roadblocks to continual learning still. Most current models and training paradigms are very vulnerable to catastrophic forgetting. And are very sample inefficient. And we/the methods are not so good at separating what is "interesting" (should be learned) vs "not".
But this is being researched, for example under the topic of open ended learning, active inference, etc.
Agreed for consumer products. We are generally much better at noise in product engineering than before, solutions are lower cost, the know-how more widespread. Electric motors with electronic control can be very quiet. Engineering plastics with complex geometry accommodates advanced acoustical designs at very low cost. And many consumer products are in rather late stages of refinement, where manufacturers are looking for added qualities to try to keep prices up.
Devices with constant hum getting quieter can actually make noise annoyance worse! The brain is good at filtering out constant noises, so they are usually less bothersome. But their sounds can actually help masking out more annoying sounds (variable/unpredictable in loudness and/or pitch). This can be used as (part of) a mitigation strategy for noise annoyance.
We are doing much better in terms of food availability, lower chance of dying early from diseases, one can communicate with people over the entire world, travel easily to all sides of the globe, and many more things.
That people find ways to be less-than-happy regardless of material conditions. That does not mean the world is fundamentally zero sum. Actually I would argue that choosing to compete for the same things that everyone else wants, like living in the same places, is choosing a zero sum game where it is not needed.
There are some patterns in natural data, and often also some in modified or fabricated data, that one can check against. None of them are perfect of course, but it can raise suspicion at least.
One aspect that influences is how close to the bleeding edge one needs to be. And how niche the model/application is.
ROCm lags by some years. And application/model/framework developers test less on it, which can be problematic in niches.
For doing something very established like say image classification, that does not really matter - 3 year old CNNs will generally do the trick.
But if on wants to drop in model X just put on GitHub/HuggingFace the last year, one would be buying a lot of trouble.
To make what I consider a halfway decent song with these current easiest-to-use services (like Suno and Udiio) takes a few hours in my experience.
To get there one has to work with the text, the song structure, find a decent style, and then do corrections on sections where the models goes off track.
To make something that is closer to "good", I would go and re-record all the lead vocals myself, and then mix this in a DAW.
"low quality substitute" and "effortless" are value judgements on your behalf. Many made similar judgements about DAWs and VSTs. And that is your right. But not everyone sees it in the same way - for some generative models are opening up a new world of possibilities.
I agree that the slot machine pull of current models is tedious and boring. I look forward to models/systems which better facilitate more creative control, directed exploration and iterative refinement.
If you concern is humans being able to have some oversight/insight, then text is useful but in many cases not sufficient. The models are quite hard to understand and chaotic - what is to us trivial or no difference can change outcomes completely. This is well demonstrated via adverserial attacks etc. There is a lot of potential for stenography also with text, where messages can be hidden in plain sight.
Sure, with several hundred hours to spare one can make some songs in a DAW. Now one can make something as good/bad in maybe 1/10x the time. Or, given the same time investment, one can possibly make something better!
I agree with your overall description of creation. But I do not agree that generative models are something else entirely. They are tools, and while their affordances do influence what people do with it, in the end the responsibility is on the creator. You can make "soulless shit" or "thoughtful commentary" or anything else you put your mind to, by using these tools in combination with all the existing ones.
Models that are oriented around one-shot, text-only direction are pretty limiting in creative flow. This will hopefully continue to improve.
And with more affordable and easier-to-learn tools, the creation of music will be similarly made much more accessible?
DAWs and virtual instruments running on regular laptop was one step, generative AI models will be another?
Wow that sounds like a real superpower in that setting. I have met many that have very strong visualization/conceptualization skills in things like 3d, mechanical, color/looks, code and electronics - but for other peoples mental models that one was new to me.
The closest I can remember would be some sales people I met who are also chess players, they could "play" in their heads a range of different scenarios of interactions in a sales conversation - many levels deep. And would do it, especially if/when stakes where high. In the blink of a second. It would take them muuuch longer to just answer me a few questions on why they thought "C" was the best way to open, and then make sure to get in point "E" (or "G") before proposing "K".
This is very minor, just a friendly little reminder that cooperation is a two way street.
The big ones would be to no longer sell Norwegian oil and gas in USD. Or to move all investments of the Government Pension Fund of Norway out of American companies.
USA could guarantee a deployment to European NATO countries large enough for them to be able to move much of their forced into Ukraine.
And just guarantees on being able to buy armament would be useful.
If there had been any ounce of political will something would have been possible.
Many ways to skin a cat. At least of this size (33k items). And at the size given, string up a database would have no advantages. Which I believe is the main point of the post! If you have a simple problem, use a simple solution.
If one had instead 1M items, the situation would be completely different.
Re 2. Parquet can easily be used with chunked/partitioned files. Then appending is just adding another file/chunk.
The case of 1. really depends on the workload. For embeddings etc selecting column subsets is rare. In order cases, where one has a a bunch of separate features, doing column subsetting might be rather common. But yes, it is far from every case.
A (simple) benchmark would be great to figure out where the practical limits of such an approach are. Runtime is expected to grow with O(n*2) which will get painful at some point.
At 33k items in memory is quite fast, 10 ms is very responsive. With 10x/330k items given same hardware the expected time is 1 second. That might be too slow for some applications (but not all). Especially if one just does retrieval of a rather small amount of matches, an index will help a lot for 100k++ datasets.
Facebook in 2025 is not at all the same software as in 2005. Nor FB the business. The market is also forever changed.
Not that this detracts from your main point. The tricky part is not the software.
If there really is no organizations competent to run government application in the Netherlands, then that is even bigger reason to start doing more of that in the country. I mean, computers are not going away! The competence and infrastructure does not magically appear. It requires consistent investment over time. Not being able to maintain computer based infrastructure is like not being able to maintain water supply of a country. Completely unacceptable.
Heck these days maintaining water supply at city scale is difficult without computers and networking...
Why do you say that the cost of throwing out American tech giants has gone up by 100-1000x compared to 15 years ago?
I mean before everything became cloud/SaaS, American software companies were still essential to most European business and governmental operations. It was just on more traditional server/desktop systems?
What is your personal productivity metric by which you have more than 10% increase? More money earned, less money spent, fewer working hours for same income, more leisure time?
It needs to be something in aggregate to mean something related to what Nadella meant. There are many individual task which LLM system can help with. But there is also may ways for those gains to fail to aggregate into large overall gains. Both on personal level and on corporate, and economy wide level.