This isn't a rebuttal of any of your point but to give a precision, the City of Paris is absurdly small, most of the Paris Area is "suburbs". The US equivalent would be if NYC was only Manhattan, and places like Brooklyn were suburbs.
Yep, the last part in my post was suggesting some ways we could determine if "item X was in the training data" (as well as some potential blockers for that from OpenAI's perspective.)
Well you could search for elements similar to the proof / problem in the training data, even if it's de-identified, right? OpenAI can probably do better than Ctrl-f "Navier Stokes".
There are probably thousands of serious academics taking a crack at millennium problems using AI every day. All those attempts are in the training data. And in fact the two researchers benefited from those attempts as well.
I don't understand the one country vs all of Europe - given a fixed population, being united in a single country surely makes everything a bit more efficient than being split in many small countries? India is much bigger than "all of Europe".
I'd wager they're not binary (absent/present) issues but have many dimensions and a continuum of possible values and thus "climate change makes this problem worse or different" is something new. Random example: https://www.nature.com/articles/s41612-024-00583-8
So EU/Eurostat has surveys, Serbia does too (I’m from Serbia).
However most countries have also national way of determining the rate that works how I described, and they may differ quite a lot from Eurostat.
I guess EU introduced surveys with Eurostat so countries don’t juke the stats.
For example
- Denmark computes net unemployment straight from benefit and PES registers — administrative coverage is complete enough to do it without asking anyone. It still runs an LFS for the ILO figure.
- Germany, Austria, Poland, Czechia, Spain publish monthly registered counts that dominate domestic debate, exactly the way NSZ figures circulate in Serbia. Each uses its own national definition, so none is comparable across borders.
- Finland keeps a register series noticeably above its LFS count, similar in shape to the NSZ/ARS gap.
I don't understand the issue with saying that someone who is employed is ... employed.
If the claim is that a single number (the unemployment rate) does not capture all the high dimensional features of society & work, then well of course!
If the claim is that politicians & media focus too much on a single number to judge whether the labour market is doing ok, then sure? It's not a particularly contentious position though I don't see any way forward - even with a _single_ number to discuss, half this thread is making wrong claims about its methodology.
You're applying nuance to a discussion which so often does not lend itself to nuance. Successive governments love to focus on the unemployment figure as though that is the mark of a healthy jobs market. As you've pointed out, it's not. It's not even close. We should be speaking about it far less, and about other, more qualitative metrics far more. We don't, though, and the cynical reason is that the unemployment metric was designed to make things appear better than they are.
The problem is that even as a single number, unemployment just isn't a very useful metric. Unemployed + underemployed is probably closer to what people care about, but it would probably be better as two separate numbers, and even then there is difficulty in defining both unemployed and underemployed, and accurately measuring them.
>If the claim is that politicians & media focus too much on a single number to judge whether the labour market is doing ok, then sure?
Yes, the gaslighting is the biggest problem. If we can't admit there's a problem (and politicians never want to admit the numbers are bad unless [other team] is in charge) then we can't begin to approach a real solution.
>It's not a particularly contentious position though
You'd be surprised. Even colloquially, that single number is how a lot of people try to justify that "job market is good, people are lazy". Because that person is not in a situation of searching for a modern job. This thread getting it right or wrong is irrelevant to that end. This isn't a board full of policy makers nor economists (at least, I don't think it is...)
Also, not all language models are transformers. You can have language models based on diffusion models or state-space models, or any other model that can be used to model sequences (so all of them, as sequences are just functions).
Transformers are just the ones that are most common and successful today.
I'd argue that meaningful sequences of symbols constitute a language. This example doesn't use a human language but it does use a language IMO (at least AFAIU).
I don't understand what you're saying? An LLM is a transformer model trained on a large corpus of natural language, often with some post-training. An image model is a different type of transformer model. What's controversial here?
Your example is a meaningless sequence. So consider a different scenario where the sequence is meaningful but does not map to any human language. What exactly disqualifies it as a non-human language?
When you encounter a human language that you can't personally read presumably you don't proceed to claim that it doesn't constitute language on the basis of your own lack of ability.
To come at it from a slightly different angle - does compiled binary code count as a language? If not, why not? (I'll suggest that it's a language albeit not a natural one.)
Any sequence could be meaningful or meaningless depending on the grammar involved. Open a word document in photoshop and the program has no idea what it’s dealing with.
The reason the language term in LLM is meaningful is how the training, symbol mapping, etc is designed around human languages. The model doesn’t process raw text, instead there’s a critical processing step which allows the magic to happen.
To my mind the argument against this model qualifying as a language model is that while the sequence of tokens may technically qualify as a sort of language it doesn't appear to be generalized by any reasonable interpretation. Further, the model doesn't appear to be able to handle unstructured inputs and outputs in the "language" - everything seems to be highly structured.
My line of reasoning could be approximately summarized as compiled binaries constituting a "real" (though not natural) language versus a sequential listing of chess positions that represent sequential game actions only being language "shaped".
Still, it's interesting to consider that if scaled up I expect the "repeat yourself" experiment would likely apply to the internal representation of the model in the same manner.
The distinction is meaningful because the process described here isn’t bound by the same constraints, resulting in meaningful consequences.
Suppose we flipped the initial stings and fed that into the process. There would still be meaning to extract from the training set but our new Reversed English but it is not English so the preprocessing step can’t be based on that assumption.
I don't understand what point you're trying to make here. At the risk of being tangential (due to not understanding) LLMs have famously been able to accept base64 encoded natural language as input "out of the box" because (it seems) under the hood they already transparently translate all inputs into their own internal "language" system on the fly. Presumably a similar translation could be learned for input of reverse english (or rather merely adjusting attention) although at a glance it seems like output might not be feasible.
reply