We are talking about the big cultural/organizational decisions that allocate where creative freedom lives or the lack of it.
To not recognize the decisions, weigh their tradeoffs and their implications, is to make the decisions blindly.
Creativity has to be actively protected where it is wanted or impactful. Or in large companies, it will get crushed as power to make choices flows elsewhere.
The problem is when practices are chosen, that improve things on one dimension, and harm things on another, without recognition of the long-term tradeoff being made.
Simply recognizing the tradeoff exists, is likely to result in wiser decisions with lower tradeoff costs, and greater opportunity expansion.
Intelligence can operate without learning. At a minimum inference and learning don’t need to be co-concurrent.
Not disagreeing with your point, but your terminology muddies your point.
But your point doesn't acknowledge that even with inference, there is a lot of room to tune the calculations. Multiple models, quantization tradeoffs are just the most obvious examples. Every architecture can be adjusted to increase intelligence/watt or other measure, even without further training.
A great number of people here are engaged in civilized debates about whether parenting needs age verification.
That might seem to be the issue, but it is a red herring.
Powerful people are ramming age verification through in a way that will surveil you and give government the ability to not just attest for your age, but surveil what sites you use, and revert their attestation, for anyone, on any site, for whatever reason.
Not just porn sites but any site they convince to use their age verification scheme.
Even Wikipedia is having to fight this with the help of the EFF to not force all its users to submit to this.
Wake up. We all need to wake up.
The purpose of an anonymizing open source alternative is to head off dystopia. Nuanced opinions about parenting are not a defense strategy. Not against a closed internet permissioning system run by governments. Implemented by people who are using parenting as a cover.
Your parenting opinions won’t help you log onto increasing numbers of sites that block you without a government supplied key.
Wake up.
We must not let governments use this issue to lock down the internet. But that is now the default outcome.
But apparently are unaware that this debate is being forced by powerful unreasonable people who are not going to give up if we leave a resolution up to them.
Reasonable and informed people want to avoid that. By stronger means than repeating “I don’t need that” until we realize that didn’t work.
When a site you and i need requires you and I to register with our governments to access it, how is your reasonable opinion going to help us navigate that?
You are imagining that a solution for you will be deemed a solution for the political powers pushing for this. Or that being age-verified is the main danger of having age-verification.
That would be nice!
But if there isn't a safe market driven solution to age-verification, which provides anonymous, unsurveiled, age-attested site access, with no ability for the government to individual monitor, deny or revoke, then that is exactly what is going to get pushed on all of us.
You don't defeat an enemy by not needing the manacles they are very motivated to force on everyone..
Increasingly: We adopt zero knowledge proofs, and other decentralized open-sourced hard-security technologies, and resolve seemingly-small, but not-going-away practical issues like age & porn, or empower and "trust" every weak politician, interest group and stranger on the internet to not use our lack of awareness and defense against us.
Add AI to the mix, and the risk/damage of passivity becomes extreme.
If you need personalized government attestation to visit a site, then the government has the ability to dynamically deny and rescind your individual access to any site that adopts age verification, at any time.
Once adult sites adopt the system, it will creep over to any site wanting to limit their liability. Banks. Business services. Eventually almost everyone.
Liability the government will dramatize and escalate. You won't see the government pass any laws to create age-liability safe harbors.
Wikipedia is already being forced to fight to not implement age verification. Age verification managed by the government = No Wikipedia access without individually tracked, controlled and revokable government permission. [0]
Seldom has a slippery slope been so slippery.
The distance between government controlled per-citizen access to obviously adult sites, and government permissioned/controlled access to any site of substance, does not even involve a technical hurdle. It just becomes a site adoption curve. Every adoption increasing the scope of real-time government surveillance in our minute-to-minute lives, and its real-time at-will ability to deny access to whatever it chooses, whenever it chooses, and for whoever it chooses. In any combination.
Dystopia is here.
In my opinion, this is terrifying.
We need: Third party attestation, providable by anyone/entity meeting basic openly-defined criteria, limited to age attestation only, implemented with Zero Knowledge Proofs, to create a safe anonymous (unsurveiled/no personalized denials) alternative, to take the wind out of the sails of this constant governmental power grab. If it isn't solved by security minded technologists and the marketplace, the freedom destroying version will prevail - and it won't be undone.
I think one of the drivers of arguments is we are trying to understand something, and the argument is subconsciously a way to not just bash their ideas, but test the strength of our own.
In the effort to create a written response we:
1. Find it takes more effort than we imagined to nail down our views, and describe them well.
2. Often have to adapt our own ideas more than we expected to make a clear description possible. Achieving a "better understanding of ourselves" is almost always a reassuring euphemism for "fixing our own surprisingly under-developed/under-coherent ideas".
3. This puts in a more humble mood.
4. Realize the issue is more nuanced than we imagine, and better assess the depth of conversation that even the most constructive engagement would require.
5. Realize we don't have energy left for that.
6. Feel good about having put our ideas to the test and "understanding them better". So a good place to stop.
The practical need to argue (test our ideas in battle) was real, we just didn't need the other person to have the argument. Their imagined pushback forced us to improve our thinking, probably more than we acknowledge.
When preparing for a race, the best thing anyone can do is tell you your opponent is achieving better times than they are. You will achieve more gains by competing with an imaginary competitor, who always sees all your weaknesses, than a real one who likely would not.
The problem with companies people can't trust, is unless they have a long track record of disclosing who shouldn't trust them before violations, they are companies nobody can trust.
Anthropic also has a habit of making major changes, without notifying anyone, then when they are caught apologizing and making that particular thing more clear. Then doing it again.
For a company that emphasizes the importance of alignment, they seem to be habitually ethically incompetent regarding "smaller" things.
Kids have always self-managed themselves with entertainment. But entertainment used to take a lot of effort. Hobbies, reading, making things, getting out and socializing in real life - which also meant inventing things to do, etc.
Entertainment is on tap now. Psychologically, kids are not built to avoid that.
They are supposed to be exploring. And they are. Finding out what is fun. And they are. Doing the hard work, of finding the easiest path. They are doing the latter, but the need for the former had been 24/7 automated way.
It isn't a mystery to me why sustained efforts may be harder for many.
> 2. Current LLMs are powerful but in no way does this indicate that it's just a matter of time until we have AGI - itself a pretty vague marketing term.
Objectively, they are surpassing human performance in more areas every year if not every few months.
So we don't need to wait for indicates, they are very visible.
If there were signs this was not continuing, that would be something to bring up. But there are not. The quality of models keeps going up. Relative to human beings.
And where they have surpassed us, they don't stop improving. On many dimensions are clearly superhuman. It is so easy to complain about their weaknesses that somehow, even people who use them, fail to "notice" they are increasingly operating well beyond human capabilities in many ways.
> US social media platforms already kowtow to domestic and foreign policy
That is an entirely different and less dangerous issue.
The issue is that when you give power of individual attestation to the government, over your ability to access age-gated sites, you have also given them the power to individually withhold or revoke attestation.
Age-gating will spread as sites limit their legal liabilities, and governments saber rattle.
Even child-friendly sites will age-gate to avoid liability issues, by confirming parents set up their child's account.
This isn't theory: Right now, Wikipedia is having to fight to avoid age-gating. [0]
This fine-grained government control over each citizen is an unprecedented expansion of power/control/threat to freedom. Whether used explicitly, or as a tool of fear that operates implicitly no matter who is in power.
Age verification starts the gating of Internet access by governments.
Any site, with any concern about age of user liability, is likely to adopt the practice. Strong laws, sold on their face value safety benefits, will increase that liability.
You won’t see any laws removing or limiting that liability.
The trend will be many more sites becoming government-gated, than we are imagining now.
Beyond surveillance, it’s a real step into government permissioned internet access, on an individual citizen level.
You keep saying LLMs. But models across many modalities are rapidly improving, evolving and merging.
Not sure what you mean by "hand waving" or "fantastical nonsense". Maybe just address points I actually make.
1. Machines with greater cognitive ability than us, potentially much greater, would be an economic challenge for human beings. With no clear answer as to how humans could manage that challenge.
2. Machines are getting more capable, year-to-year, faster than any human can or ever will improve. With no signs of slowing, or any areas where they are failing to improve.
3. None of this is new. Computing capabilities have compounded steadily since the first transistors less than a century ago. Human's biologically driven cognition, in contrast, has not improved.
4. Machine capabilities are now regularly passing us in new areas, and rapidly approaching the general threshold noted above.
Explain your perspective, I am genuinely interested.
For instance, what cognitive capability do you have, that you believe machines won't exceed within 5 years.
Yes, that is what "fi" means. Fiction often introduces and explores ideas about the future. Ideas.
> LLMs are a useful technology. They have no relation to AI terminator bots
Technology that can communicate with human language. Also relevant, other SOTA models operating in other modalities.
The important thing if you reference SciFi, are the major ideas that stand on their own merits. Whereas, specific characters, such as terminators, are usually not predictive.
You are actually underlining my point. The disconnect a lot of people are struggling with is real.
> AI will go no differently than the Industrial Revolution.
So it is different this time is it?
"We" are different?
Unlike the 99.9% of species that went extinct, and continue to go extinct, we can't be supplanted? Obsoleted? Out competed? [0]
A few thousand years at the top of the food chain, and suddenly that few thousand years is "what can never change".
Technological change already drives machine cognitive progress forward faster than any human who has ever lived. That factor alone makes competence/capability obsolescence unavoidable, outside of technological collapse.
This is an unprecedented rate of real-time change, outside of instant disasters like volcanic eruptions or comet impacts, in direct competition with human capabilities.
SciFi has been broadcasting the potential and risk of AI for a couple centuries. This is a spectacular moment in the universe's history, assuming we don't blow ourselves up. Not just another day for humankind.
There must be a word for the inability to see the import of change. It seems to be a common disability.
This seems to be a real effect: The faster things go, the less sensitive we are to the rate of change. Precisely because we can't imagine what the world will be like in a decade, our horizon shrinks, and we treat short intervals as if they were long intervals. Anything that isn't instant feels like it is barely moving.
I am terrified because the majority of discussions are not treating Problems #3 or #4 seriously.
Problem #1: Verify ages for adult-appropriate content.
Problem #2: Do not create a new surveillance system for the government.
Problem #3: Do not create a --dynamic-- per-user internet-access permissioning/control/denial system for the government. This would be madness.
Problem #4: Do not create a --general-- per-user internet-access permissioning/control/denial system for the government.
Problems #3 & 4 are very very serious. Any age-verification system, or generalized control-system, will become a pervasive requirement for a large percentage of internet sites, simply by being convenient for risk-adverse sites to adopt. Which means, this will be (not could be: will be) a pervasive government managed system of per-user internet control/denial.
We need zero knowledge proofs. Or similar. Minimum.
Form of solution:
• Any third-party institution meeting publically-defined government requirements, can be an attestor.
• Attestors provide persistent age-verification tokens per user-device, on user request/verification: Using actual identification and one-off-anonymized device identifiers.
• Sites are given one-off, device and user-identity anonymous, attestation tokens, derived from the device token, on request.*
Maximum anonymization is necessary so:
• Governments can't identify your devices.
• Sites can't identify your devices or you.
• Government and sites cannot collude to remove anonymity (via this system), enabling a new surveillance system.
Attestation tokens need to be persistent so:
• The government is not being given dynamic control of our internet use.
Age-attestation tokens, must be provided independent from any other types of attestation, so:
• We didn't just invite the government to link multiple permissioning systems together, which will inevitably be used in coordination. And so failure to supply age-attestation tokens is a clear documentatable act.
Unrestricted (other than by public standards), third-party attesters:
• So users can choose an attester they trust. So there are multiple points of attestation. So there is individual accountability, natural documentation, and reputation risk for any denial of attestation.
Anything less anonymized, independent, persistent or available than this is a nightmare situation. Really.
I am terrified because the majority of discussions are not treating Problems #3 and #4 seriously.