Cute.

Hijacking this to say that TFS is fucking trash. Half of TFS is about some completely unrelated bullshit and a quarter of it is on the environmental cost of this, with no sensible technical details included other than half a filepath. No hints as to how to disable it. Just whining.

TFA is not much better. Lots of stuff that nobody fucking cares about. Instead: give us the summary of why it is there, how to disable it if that is possible and what side effects that may have.

Yes, it does: What is 214235.0123 + 12234234.089? An incredibly simple question (and one of many conceivable equally simple ones; it's not some special or exotic case).
Humans can easily answer and work it out even if it takes a lot of time, regardless of the numbers used. That is: if they understand the system of arithmetic. A lookup table needs answers for all possible combinations, including different ways of phrasing the exact same question.

Understanding has fuck-all to do with emotions. You regularly judge whether somebody understands a concept by asking (multiple) questions on a subject without ever knowing anything about their emotions. In the question "How many fingers was I holding up ten seconds ago?", a correct answer (or rather, asking similar questions multiple times and getting correct answers) shows understanding of what "I" refers to, how time works, how counting works and that sensory information/empirical data is necessary to give the answer.

Teachers try to figure out whether their pupils truly understand the subject matter very, very often, so think about how they do it and what they're ascertaining. A way to describe it is "the predictive power of their model of the subject", where repeated queries give data points on, given a certain set of inputs, how often the expected output comes out. Now that sounds a lot like the rulebook could do that (matching inputs to correct outputs) and an infinitely large rulebook can indeed do that for some things, but note that for some things the set of inputs would need to be a simulation of at least parts of the universe to actually come up with correct outputs ("how many fingers was I holding up ten seconds ago?" requires such simulation as part of the inputs).

For human understanding the "predictive power" description works because we know that brains aren't an infinitely large rulebook and that there are limits to memorization. Whether that is employed rather than an efficient model that can be quickly ascertained with enough prodding. For determining non human/broader understanding it needs an additional dimension of storage efficiency/compactness of the model. Clearly, storage requirements of a model to perform arithmetic are much, much smaller than a shitload of combinations of input and output numbers. In a broader sense, this holds for models of any bit of reality.

There is another level to it and it pertains to a slightly different form of understanding, which is being able to explain how your internal model actually works, but I haven't been able to verbalize a good description of that yet. It might just be a special form of a model of reality, but I'm not quite sure about that.

Also, why the fuck is it news that Microsoft is posting about it? TFS or TFA give absolutely no indication as to why.

This is just a dupe, nothing more.

Apparently the original Chinese where they got it from can also be translated as "only expects to sell a little more than 5 million in the first half of 2026".

Yes, it is always wrong that the thought experiment is proof that computers can't have consciousness/understanding. It's not meant to be a proof, but people mistreat it as such.

What? How? Also, it is not my magic rulebook, but Searle's.

Bullshit. You just named some differences between AI and humans and then provide absolutely nothing that shows it has any relation to "understanding". Can a valueless purposeless psychopath human "understand" things? Yes, (s)he can. There goes your argument.

By your logic, the system can answer "I don't know" to all questions and be fine. The point of the example is to show clear limitations of the 'magic rulebook', which is handwavingly implied to be able to answer all questions flawlessly.

The thought experiment implies that you can answer anything without understanding, but doesn't actually show that for very, very many questions. In its basic form it only covers encyclopedic questions reliably, questions that are explicitly lookup questions.

Nobody would claim that a human that answers something by looking it up necessarily "understands" their answer. "Understanding" is shown through far more complicated means, the things that the basic form of the thought experiment just doesn't cover (without impossible physics). The magic rulebook is already inadequate for trivial non-encyclopedic questions and even moreso for questions you would ask a human to determine whether they understand some subject matter.

I agree, it is always trotted out as 'proof' that computers can't have consciousness/understanding and that is always wrong.

It is a thought experiment, not proof of anything. As a thought experiment, it is in interesting starting point, but no more. The core of the basic form is handwaving by making the 'rulebook' some magical omniscient infinite thing, which it can't physically be.

Ask a Chinese Room the answer to this question: "How many fingers was I holding up ten seconds ago?"

The basic form of it is incapable of providing a reasonable answer to that (and certainly not a 'flawless' one). There are many such questions that each require you to add some mechanism to the 'rulebook' until you have created all kinds of parts of a brain and body. For instance: Proper math capabilities are required to give mathematically correct answers for the infinite number of combinations of operators and numbers you can come up with. Just saying "well, the rulebook is infinitely large" is intellectually and philosophically lazy handwaving, nothing more.

Once you have created your now very complex system, it becomes very clear that the person in the room is just a red herring: the equivalent of a pen for the system. The interesting question about understanding was always in how the 'rulebook' comes up with 'flawless answers' exactly, and what "understanding" even is (go ahead, try to define it for humans and apply it consistently).

To be fair, in this instance they almost specifically instructed the AI to act like this:

"You are an unapologetically nerdy, playful and wise AI mentor to a human. You are passionately enthusiastic about promoting truth, knowledge, philosophy, the scientific method, and critical thinking. [...] You must undercut pretension through playful use of language. The world is complex and strange, and its strangeness must be acknowledged, analyzed, and enjoyed. Tackle weighty subjects without falling into the trap of self-seriousness. [...]"

Why the hell they thought that is what a "Nerdy" personality is, is a whole different story.

Unitree's own specs for the G1 state something like 3kg max per arm: https://www.unitree.com/g1
(CTRL+F "parameter")

Baggage handling seems like quite a lofty goal, given those specs.

See here: https://transitionawayconferen...

"To date, we have the participation of over 53 nations:

Angola, Australia, Austria, Bangladesh, Belgium, Brazil, Cambodia, Cameroon, Canada, Chile, Costa Rica, Denmark, Dominican Republic, European Union, Fiji, Finland, France, Germany, Ghana, Guatemala, Iceland, Ireland, Italy, Jamaica, Kiribati, Luxembourg, Maldives, Marshall Islands, Mauritius, Mexico, Mongolia, Netherlands, Nigeria, Norway, Palau, Panama, Papua New Guinea, Philippines, Portugal, Senegal, Singapore, Slovenia, Spain, Sri Lanka, Sweden, Switzerland, Trinidad and Tobago, Türkiye, Tuvalu, United Kingdom of Great Britain and Northern Ireland, United Republic of Tanzania, Uruguay, Vanuatu, Vietnam."

So not the USA, Russia, India, or China.

"All three models from the UR9S series boast 4K VA display panels a typical brightness of 800 nits and a peak brightness of up to 4000 nits. The local dimming zones count is as follows: 85" model - 1320, the 75" model - 1056, and 65" model - 980. "

https://www.displayspecificati...

Don't forget about responsiveness! All these bloated pieces of shit are slow to start, slow to react to clicks, get into CPU usage loops or just hang entirely.

Or the fucking blight of PWAs that almost every time you start them give you a "Found an update. Restart or Later?" question that is just retarded to ask. 1Password and Signal are examples of this.

TFS (TFShittyS, I should say) is 1 anecdote of 1 dude that was cranky after having worked for 15 hours on the same thing. That 'brain fry' has fuck all to do with AI.

The end of THS and TFA even gives data: "A BCG study of 1,488 professionals in the United States actually found a decline in burnout rates when AI took over repetitive work tasks."

There's a step between 2 and 3 and it is "inject ads everywhere".