same argument can be made about excessive athlete salaries and really any sports related business ventures. Athletes go after specific stats to hit contract goals, get their bonuses and live good lives. Gambling industry is just one of the hundred detractors to the sport itself.
logical or well reasoned argument doesn't equal a casual or factual relationship. The well reasoned arguments on many issues change over time, just take the same issue and go back in time 100 or 50 years to find much less consensus and much weaker logical links. Elon shows pretty consistently that truth for him is mostly just what Elon deems truthful or useful.
how much do you know about the signature collection process? I can easily see how a large number of these could've been out of fear of retaliation and some out of greed. Many autocrats claim high support numbers on every election, doesn't mean they are beloved.
In author's defense he doesn't claim to be exceptional and even concludes his essay with labeling this A.I. moment as “the revenge of the so-so programmer.” I actually enjoyed author's nuanced take, different from the common narrative of AI takeover. To the contrary I think LLM-driven coding is not a tool of power users, but rather a tool of disempowered users. They will bring many more people into the engineering profession and change the profession for many in it towards more complex tasks. Quickbooks hasn't destroyed the need for accountants, yet it allowed millions to become the "so-so accountants", when that's all their business needed.
I think the deterministic requirement was what made the semantic web nearly impossible at scale. Aligning on the schema for all the entities can be challenging within a small group, yet alone the world. Doing it continuously is impossible today. Wikidata has tons of abandoned and branches which are no longer relevant or even wrong.
It seems like LLMs and vector representations for words are more practical ways to explore semantic data. Then the specific, reccuring queries may be optimized through graph representations.
in fairness, the society did self regulate as evidenced by Meta's declining engagement numbers. Many people got depressed from reading Nietzsche, should we regulate those too? Was the internet a net good or will admit that it's a glass half-full argument.
There is an extreme conflict of interest in the OpenAI's proposal. I don't see regular people protesting and asking politicians to act, I don't see small business owners writing petitions. I see a small number of highly invested, very rich individuals weaponizing media attention and lobbying to create extremely favorable combinations for their business.
Is Twitter dangerous? Is a steam powered engine dangerous? Is a printing press dangerous?
In retrospect they aren't. In history they were hated, vilified, regulated and banned.
I think the discussion has a thick veil of anxiety and sci-fi movies. It's very strong on hypotheticals and very light on actual evidence of harm committed by AI and actors. "We need to act before there is evidence" is hard to argue with, thankfully human imagination is not regulated yet.
the inspiration for founding replica was the founder's intense pain and trauma from suddenly losing a friend to an accident. Then reflecting on the psychotherapy culture in a country with a rich history of government assigning what constitutes proper behavior patterns.
It's easy to label somebody's psychological state as inadequate, but what's harder is providing them tools to recover. What's nearly impossible today is providing such tools at scale.
Would your general sentiment on replica and it's users apply to using idiotic VR simulators for PTSD survivors? I can make an argument that a reasonable adult should not be tricked into believing they are actually on a roof ledge when they are clearly on a couch. However, that kind of therapy has shown some promising results and is certainly more accessible than traditional care. Granted replica may not be the working tool, but I'd be curious if you have other options its users can go to when traditional human relationships are challenging for them.
Mechanized factories failed to kill humanity two hundreds ago and the Luddite movement against them seems comical today. What makes you think extinction is most likely?
this is not whataboutism, it's actually guilty-by-associationism displayed on both sides. Should we blame children whose elderly parents support the war for taking care of their parents, instead of disowning them? Should we blame employees of Google for allowing employees of Lockheed Martin use Google maps during the Syria invasion? Should we blame employees of Yandex for paying taxes and working for a company which did not cease operations in Russia?
It probably won't change your mind, but here is a pretty transparent poll on a mix of Russians at a train station in Moscow (to capture some from outside the metropolises). The multi hour poll is a continuous video to show lack of selection bias (parts with no speech are fast forwarded)
https://youtu.be/eNwe24PlPTQ
About half the Russians don't support the war. I hope the conflict ends soon.
earlier in the thread someone provided a definition along the lines of: "a pimp is a manager who is allowed to use violence, drug addiction and other forms of non-consensual coercion."
I hope you can explain your point more. Coincidentally you won't find many nannies from the wealthy western countries.
Which doesn't tell me very little about this kind of work. But maybe it does to you.
that's why a lot of responders are calling the approach naive. This function is potentially the last layer in thousands of functions. Thinking that one can predictably change the behavior of a large nonlinear model via one variable is like thinking you can change the US political system by passing one good law in Reno, Nevada.
This approach only works if auditability is mandated to also be a property of the model, which typically reduces model complexity and accuracy too. We do this with credit scoring models semi-successfully.
I feel like authors are thinking about the "algorithm" almost as if it's just a bunch of if-else statements in a tree.
Read only audit is an overly naive way to think about understanding complex algorithms. In an audit they would find an enable of large DNN transformer models with thousands of layers and thousands of features often with their own transformation trees. There are entire CS departments dedicated to researching tools to understand complex nonlinear models. You definitely can't do with a read only code audit. You can _barely_ do it with full access to the model, full access to the input data and being able to retrain and rerun the model on that data.
One can say that Solomon Asch experiments on social conformity are great illustrations of the benefits of free speech. Diverse opinions discourage group think and encourage critical thinking and discourse. That works even if the opinion is bogus, since in that case it may encourage the individual to avoid adopting an overwhelming majority point of view. As you mentioned it is indeed a slower process, but a much more fault tolerant one. The simple analogy is the slower, complicated democratic process vs fast, simple autocratic process.