On your second point, that rather assumes those 'unless' people will willing to just 'go away'.
That's typically not what happens - you either get bloody revolution or people just ignore the rules that aren't giving them a fair shot ( ie turn to crime - ie state disapproved activities ).
This is what has traditionally held back the elite from their worse excesses - they don't want to live in a world where either of those things happen.
ie what's the point of being rich if you are constantly worried about yourself or a member of your family is going to be violently kidnapped etc.
As to your first point - it's a bit like the first - if there is no social safety net, then it you could happen to you - you could get ill, lose your job etc. Even if you believe the system is created by total self-interest - then the past has shown that the equilibrium involves looking after everyone to some extent as:
1. It could be you one day.
2. Having lots of desperate people, even if it isn't you, is in no-ones self interest.
I agree historically, but your points break down when the threat of revolutions no longer have the physical or economic power they once did.
Good luck violently kidnapping the family members of a AI CEO in a world of flock cameras and humanoid robot patrol units.
And good luck withholding your labour when 80% of people can't even find paid employment.
Also, arguably if the unemployed masses are going to try to overthrow those provide them with at least the basics to survive and part take in a revolution, then those involved in the revolution shouldn't be surprised if what little they have is withdrawn. Strategically this sounds a lot of like punching your boss and demanding he pays you more or you'll punch him again. This isn't a reasonable negotiation tactic unless you hold some of the cards. The only reason threatening to quit a job can sometimes work is when it's a real threat of economic harm, absent that, it's just annoying and they'll probably be happy to see you off the payroll.
There are too many people in the world currently that think that security is through absolute technical/military dominance, when the reality the thing you really need for security is other people or countries not wanting to harm you.
In fact the rise of things like cheap killer drones or robots makes that truth more real, not less.
> And good luck withholding your labour when 80% of people can't even find paid employment.
So those 80% are either being supported somehow to live a life at a level where they don't resort to violence or crime or they aren't. ie you keep them happy at some level or you fight. Note if you let them die/kill them, then you will likely lose the support of the 20% as they can see which way it's going.
Take the classic NSA encryption logic - if you build some capability with a backdoor only you can exploit then you are building your own future capability - and if you think you are the 'good' guy - what's wrong with that?
It's not thinking, it's calculating. "Good" is just a pattern.
You might think our brains are just a network of neurons and what we think of as thinking is just fancy pattern matching in that network of neurons. But architecturally they are so far removed from each other that the mathematical equivalence has no practical meaning. LLM's aren't doing what we're doing.
To be clear your use of the term thinking isn't the important thing, but the very loose use of terms like intelligence, thinking, etc. are at the heart of the problem of talking about LLMs.
LLMs are amazing knowledge remixers. They are superhuman knowledge remixers. In application domains where the remixed knowledge can be verified, they do a great job. But nobody is going to value a 98% reliable knowledge remixer at $2 trillion.
While I agree the numbers don't make any sense, for reasons too numerous to enumerate, what matters for the valuation is not how it works, but what it can do.
So if an ML model can do real time language translation or drive a car - that has value - whether it's remixing or thinking or whatever doesn't really matter.
Apart from coding, which since the C people coded up lint has had the best tools for validating the work product, what can LLMs do? LLMs can't be trusted to write court filings. The software that can drive cars isn't an LLM.
But let's take the one you've accepted - it can write code. It can translate from intention to code quite effectively - then they can do anything that you can do by writing code.
Also while you dismiss ML that drives cars as 'not LLMs' - it's part of the ecosystem - these task specific models ( the ability to process images, sounds, drive cars, synthesize speech etc ) - extend what is accessible to a computer programme that can write code.
Look at what's happening in the world of robotics.
You have lots of specific ML models that allow the robot to move, see etc.
Then the LLM allows it to convert a human command in the form of arbitary speech ( rather than fixed preset commands ) into action.
I honestly never got why the belief that if you used this line of reasoning in practice, you were just a bad guy, that thought they were a good one, never seems to be a more common belief
I mean it pops up in media a lot, either as misguided, or deceptive characters
The point is you can justify almost anything on the basis of defining yourself as the 'good guy'.
Take the US recent assassinations of Iranian leadership. Objectively unilaterally assassinating other countries leaders is wrong - but somehow it's fine if you have a good guy/bad guy lense.
One of the problems here is huge swathes of material that the AI will have ingested includes this kind of exceptionalism thinking - it's particularly deeply rooted in Western culture generally and US culture in particular.
The problem isn't so much whether the good guy/bad guy allocation is corrrect - rather that it's very existence in the training set is deeply dangerous - as all it takes is one single classification to happen and suddenly everything is for the greater good.
> Objectively unilaterally assassinating other countries leaders is wrong
Why? If the leader is responsible for atrocities, then why would it be wrong to kill them? Especially if their removal would result in a more liberal leader replacing them, and not just chaos or more bloodshed.
In the case of Iran, there was no evidence of that, but as a general rule leaders should not be a protected class.
I've seen this - I was part of a large company that threatened to move everything, not to open source, but to google docs, and MS sales team moved heaven and earth to reverse the decision.
It's not just that they don't want to lose a customer - they don't want lots of examples out there of how it can be done.
In person, pure paper voting, in local community halls etc, run by volunteers is the best voting system.
Incredibly resistant to influencing at scale, resistant to voter coercion, transparent, and invokes a sense of occasion and community spirit.
The problem with electronic systems is the public can't see the piles of votes, can't get involved in the count, and ultimately it's the perfect system for being able to tip the scales subtly.
Even vote counting machines for paper ballots are a mistake in my view.
> In person, pure paper voting, in local community halls etc, run by volunteers is the best voting system.
This is the right answer. Using tech as an auditor for batches after manually counting could be useful. Getting everyone together to count votes in a process that everyone can see, understand, and be personally invested in - that process is what is needed far more than efficiency, at this point in time.
This is a BIG discussion in Brazil, where the election is basically 100% electronic, no paper involved. There's a large portion of the population that distrusts the electronic voting machines, partly due to rightwing propaganda, but partly due to a genuine distrust of a system they can't understand.
In person paper voting, however, only works on a relatively healthy democractic community. What used to happen around here was that whoever was the local hegemonic power, mostly oligarcs, sometimes gangs, would just plant a grunt in the voting halls and coerce people to reveal their votes. That or they would simply have a bag of pre-cast votes and drop them all in the ballot box. Lots of absentees and dead people used to happily cast votes at the time.
So ultimately the full electronic system has been working quite well over here for the past 30-something years. It has only really started to be aggressively questioned since 2016 with the rise of social-media as an election driving powerhouse.
I also think it's pretty clear that this "aggressive questioning" has ulterior motives, since it never points out specific flaws in the transparency of the system, and never advocates for actual improvements to electronic voting machines, while there are very obvious low-hanging fruit to criticize and improve.
For example, while the software and hardware is audited by a very diverse panel, it's too institutional: you can't inspect it as an independent researcher. It's also not clear which parts of the supply chain are subject to the audits. I trust it's all of them, and these institutions are doing their job of pointing out the parts that they have _not_ audited, but I, as a member of the public, can't verify.
I'd argue ff you have thuggery at that level in public then you have bigger problems than voting systems.
And if that's country wide then the country isn't a democracy so the voting system is moot.
In terms of voter fraud, it's the very inefficiency of paper in person voting that provides the large scale protection - it's incredibly hard to do what you describe at a national level.
On the other hand, it's much easier for organised crime to influence voting machines nationally - in an undetected way - you probably only need to intimidate a handful of key people.
Yeap, like I said: you need a relatively healthy democracy for the community counting to work. This wasn't the case of Brazil 30 years ago: Brazil was a military dictatorship until 1985. In 1996 when these electronic voting machines debuted, a large portion of the country was pretty much owned by a small number of families linked to the old regime. That's especially true for relatively uninhabited areas such as the states around the Amazon rainforest, where population density is in the single digits per square mile.
Brazil has another particularity in comparison to the US and some other countries: we don't have a 2 party system, we have a lot of parties. It's probably around 30 today, but used to be more. It's impractical for each party to have people on every voting hall. It's much easier to oversee hardware and software production of the machines.
Is it still possible to intimidate or bribe this handful of key people? Probably yes, but as far as we know that never happened, and the parties are very obstinant in how they oversee the whole process.
One thing I definitely think we lose is the sense of community, especially in times like these when individualism is on the rise: we employ a couple people on each voting section, maybe a cople dozen on each voting hall. In the end of election day they simply run the closing protocol and go home. Having a bunch of folks counting votes for a few days definitely could make people more engaged with the election, and with democracy overall. The "electronicness" of the whole thing alienates people from the process.
I've got over a decade of experience supporting elections using a ballot scanner.
The need to balance voter security, with auditability, rapid Ness of results, with ease of setup/teardown of the voting place, is hard to beat with scanners.
> Hand-counting the ballots instead of scanning would be expensive in terms of time and money, and error-prone.
Why?
It's not that expensive - what price democracy? And in terms of error rate - if there is a very close vote, in the UK they just do a recount - sometimes more than once - the variation between recounts is very very small - single digit ( votes not percent ) at most.
to be fair, I guess one difference between the US and the UK is the distributed nature of the vote - in the UK it's broken up into constituencies of ~70K voters - so any recount is scoped to that ( ~ 50K actual votes ).
For a non-trivial amount of ballots, the overhead of securely managing the ballots is substantial. You're casually driving up staffing costs all over the place, and legal challenges will ensue if everything is less than meticulous.
Extend the question why are rapidity of results and ease of setup/teardown of the voting place design goals at all?
There are elections in India and the US to control billions of people and some very aggressive military people armed with weapons capable of ending modern civilisation. This is one of the most fundamental social technologies we have to keep the whole system stable (up there with limited liability corporations and courts). The downside of one of those elections failing because of voting machine trickery is stupendous.
Frankly, I'm not sure even auditability is that important as a design goal compared to voter security. If the election is close enough that an audit matters the two candidates are pretty similar. In practice, if there is any meaningful gap in preferences then the outcome of the election should be clear enough that an audit isn't that meaningful. It's nice to have and it is cheap enough to include in a paper ballot.
The last 3 US elections were very close and had very different candidates with very different ideas. The closeness of the election has nothing to do with the “similarity” of any candidates. And to imply a “coin flip” would be fine is ludicrous.
Even the next day is silly. If the election is Thursday, it seems reasonable to count on Friday and the weekend, and have the new MPs declared on Sunday, ready to travel to London on Monday and get sworn in.
In the US, the election is in November and the president doesn't get sworn in until January. It could take a couple of weeks to count the votes and it would be fine.
> Even vote counting machines for paper ballots are a mistake in my view
I think machines are only acceptable if they sort ballots into multiple bins, not if you are using them for the actual count. The bins should be designed so you can casually see that the last ballot to be added is correct for that bin while the machine is working. Something to insert a marker every 100 ballots is probably a good idea, then you can count the number of hundreds, and randomly sample a few of them to check they are correct.
> In person, pure paper voting, in local community halls etc, run by volunteers is the best voting system.
Unless those volunteers are all die-hard fans of a particular party/candidate and thereby intimidating everybody that has a different opinion. (Not openly of course, but if everybody knows everybody, the mere presence suffices to sway people.)
In Blighty, the counters (paid volunteers from the local council working voluntery overtime) are supervised and audited. Not to mention representitives from the local political parties will be present as well as the local press to oversee the proceedings. Typically the returing officer will also allow anybody else to be present and observe the proceedings, space and practicalities permitting.
This would need at least a majority of die-hards present in a (near) majority of polling locations. At this point, the party can probably win the election without fraud.
For the counting process you have "auditors" from all the different parties. We have that system in Argentina and it works pretty well in terms of electoral transparency. People's choices are another thing.
These volunteers will have a far easier time scaling their conspiracy if voting machines are involved. There are fewer points of failure to corrupt and it is easier to produce a plausible sketchy result where their guy wins by 100 votes.
I think YouTube just defaults to a lower resolution (was like 360p in auto when I opened it) maybe because it's mostly text (cause I noticed that too), but you can change it to be 1080p on the settings cog on the video.
> and decided I'd rather just use vanilla css/js for new projects.
So is the rise of AI coding, the end of frameworks?
Frameworks were created to reduce boiler plate ( at the expense of new abstractions which occasionally leak ) - boiler generation is less of an issue for AI, and the underlying platform is very well specified?
I guess the only problem with vanilla is if the sheer volume of text overwhelms the context?
Vanilla css/js are not really that voluminous. Actually, they tend to be smaller than the compiled targets you get from conventional web dev builds.
I’m not really sure that most web frameworks reduce boilerplate. There is a lot of boilerplate involved in setting up routes, and builds, and so on. As far as I can tell they only really make sense for very large teams (who need a uniform shared data model, design language, processes, etc.) and people who really want to run server side js/do a lot of SSR.
The thing about web “frameworks” is that they’re not that different in structure from what you’d build to do SSR and a repeatable way to add pages/dirs and shared assets, in any language. You can build a minimal version of something that does that in less than an hour, at least in Go. go:embed and a factory method for handlers :)
I think we’ll all end up with more bespoke “frameworks” (specialized web servers) with less gravitational pull towards dysfunctional (IMO, the node ecosystem) tooling, and still have many people using the best major frameworks. Truly large projects will still need to use react and svelte and co unless they can train a model on their own framework. But that’s really hard
Even more fundamentally, frameworks were created to reduce the developer time and skill required to build applications. This has been a pipe dream many have chased since COBOL replaced assembly language.
LLMs are probably the first technology that actually approaches doing that. And yes, they don't need frameworks, they're indifferent to the amount of code or boilerplate needed, don't care about the developer "experience," and will write vanilla JS and HTML just as well as anything else you ask them to do, maybe better.
> The other point is if their new model isn't safe, doesn't that mean all their previous models aren't safe as well?
This understanding comes from the fact that people think either models are fundamentally unsafe or they are completely safe. The reality is that there are different levels of alignment possible. So it is really a tradeoff between effort/time taken to align the model and speed of release.
Bigger models require better alignment. So I would assume that the newer models will undergo more stringent alignment checks. For instance, I wouldn't put that much effort on GPT 4.
> So you are telling me Sol can't be used for hacking?
Who said anything about hacking? You can use it, but it is very restricted. But a much more capable model must have much higher restrictions on alignment. Obviously you wouldn't need to spend too much on alignment on a 3B gemma model. This must be obvious, I'm not sure why you bring this up?
> They didn't need to do that - they can just release when ready - nobody is expecting them to release a new model on a certain date.
Again a conspiracy theory? People expected a model during Dev Day. And they are just setting expectations. Whats with you guys and conspiracy theories?
I said if they were serious about 'safety' then given by their own graphs that Sol is less aligned than Astra, they should have pulled it.
As they haven't, the obvious conclusion is that safety isn't the main driver here.
I'm just following the logic - you are the one making assertions about motives that contradict the evidence.
Now I have some sympathy for the idea that these companies need to be responsible, but in the end - because these are universal machines - the regulation is going to be needed at the machine real world interface, not at the model itself.
This is, by the way, why OpenAI is responsible for the hacking that it's AI driven agents ( not models per se ), did.
It's not a problem to have an LLM that thinks it can drive, the problem is putting one it an actual car and putting it on the road.
Models like this are in effect dual use - just like computers and computer programmes in general. Ethercap is a very useful debugging tool - and can also be used for exploits.
We don't ban Ethercap, we police the use - just like with all dual use stuff.
If OpenAI is directing the software ( how ever badly ) that's hacking - OpenAI is responsible, it's not the fault of the software.
Note if somebody else is using OpenAI models to generate code to hack - that's not OpenAI's responsibility it's the person using and running the generated code.
> Note if somebody else is using OpenAI models to generate code to hack - that's not OpenAI's responsibility it's the person using and running the generated code.
No, I don't think a company can get away from releasing a powerful tool that can cause disruption if used poorly. This is just not how society works.
The reality is that labs will get blamed for things people do using it. And they have - look at any news about US Army using Claude etc
Things can be regulated in sensible countries where the things are pretty much single use - like guns.
However, nobody bans cars because somebody has driven one into a crowd, or hammers because of a particularly grusome murder.
These models, just like computers, are universal tools by their nature - they don't fall into the gun category - they fall into the hammer and car category.
> Things can be regulated in sensible countries where the things are pretty much single use - like guns.
Sure and there are regulations on who can own guns and licenses. And that's literally how Mythos is handed out today - to trusted partners. What is your opinion about that?
Also, companies are not allowed to release stuff like Nuclear technology, chemical weapons etc. These literally can't be regulated and a single adversary can use it destructively.
"but but they are just tools ok" doesn't work when the capability is big enough.
Look, it might be easy sitting behind a computer asking OpenAI to release whatever new model because you personally don't think you will be affected. But the society needs to be careful on what it hands potential adversaries.
There is plenty of commercial hacking tools. And those seem to survive just fine.
Issue really isn't that these models can be used for something.
Issue is that they have been used for something with such disregard of any even slightly reasonable oversight that everyone in executive team, board and every shareholder should be in jail just now...
The tricky part is that social peer pressure requires a lack of anonymity - not just in terms of detecting whether people are recording without consent at the time - but also the posting of content online.
I bet Disney's lawyers can squeeze a lot more money out of municipalities for playing the music publicly in the first place than they can out of random citizens who are uploading to Youtube.
Of course that was the point. But malicious broadcast of copyrighted material to trigger automated enforcement systems sounds pretty far from "fair use" to me. Say nothing of the bad publicity when some video is just so bad it hits the news and now your IP is the soundtrack to a cop beatdown.
When you look at it from a legal "use it or lose it" and corporate risk perspective it's quite surprising that Disney didn't even sent cease and desists.
Edit: Of course you don't lose the copyright but your ability to litigate someone else's unfair use is a lot weaker when their lawyer can trot out a million examples that you were fine with.
That's typically not what happens - you either get bloody revolution or people just ignore the rules that aren't giving them a fair shot ( ie turn to crime - ie state disapproved activities ).
This is what has traditionally held back the elite from their worse excesses - they don't want to live in a world where either of those things happen.
ie what's the point of being rich if you are constantly worried about yourself or a member of your family is going to be violently kidnapped etc.
As to your first point - it's a bit like the first - if there is no social safety net, then it you could happen to you - you could get ill, lose your job etc. Even if you believe the system is created by total self-interest - then the past has shown that the equilibrium involves looking after everyone to some extent as:
1. It could be you one day.
2. Having lots of desperate people, even if it isn't you, is in no-ones self interest.
reply