> The issue with high-profile half-baked interventions is that they distract from real unsexy solutions like grab the plastic in rivers BEFORE they make it to the ocean.
You mean they distract from doing… exactly what they are doing?
Maybe give a team that has been working on this problem for 10+ years enough credit to check what they are doing before dismissing them.
> The crew are saying that the request came in on a frequency they weren't monitoring
There is only one channel you are supposed to monitor, it’s 16. That’s also where the coast guards make all their announcement/requests for help and whatnot. The reason there is a single channel to monitor is precisely to avoid the situation of different vessels being on standby on different frequencies and thus being unable to hear each others.
So that means they weren’t monitoring 16. Which is unacceptable.
I know I need to have 16 on at all times, and i am not a professional super yacht captain.
No, all coast guards announcements are on 16 regardless of severity, because no one is monitoring other channels so any announcements on other channels would be like screaming in the void.
The initial call from help from the stranded boat was on 16 for sure, there is no other channel anyone would use for that.
The severity is instead communicated in the first part of the message with either « sécurité » for navigational hazards, « pan pan » for non life threatening events like this one, or « mayday » for life threatening events like man overboard, sinking etc.
The system is based on a single channel design specifically to avoid this situation.
You only use other channels briefly, e.g. to talk to the dockmaster when coming into port, or to talk to another boat after having hailed it on 16. So it’s possible to miss a coast guard announcement, but they get repeated and in this case because they were the closest boat the coast guard must have tried to hail them multiple times.
My guess is that zuck yacht has a smaller boat attached to it, and they basically stayed on a different channel to communicate between the two for hours, never bothering to check 16. Which you aren’t supposed to do. Even then I would be surprised a $500M yacht can’t monitor two frequencies. I can on my little sailboat.
It matters if an article was written by an expert or someone that only has surface knowledge about the topic. It matters if it’s written by someone talented with words or by someone that has to churn 10 articles a day for a content farm.
So clearly not all sources are equal, and that applies to LLM too.
If you spend 4h of your time writing an article on something you care about it’s likely to be interesting. If the extent of the effort you are willing to put in the topic is 10 seconds of your time to ask ChatGPT to « write an article about phrack », then just don’t bother. We can all ask our own GPTs to do it for us whenever we want.
In theory someone could iterate for 4h with ChatGPT to get to the exact words they wanted to say without having written a line of it, but unfortunately for those folks, 99% of the content generated in that way will be nice sounding emptiness, and people don’t want to read 100 pieces of BS just to find 1 good one. There is too much content online and not enough time to read it all, we gotta filter somehow, and LLM sounding prose is one of the best filter available today.
FYI, you're arguing with a petty psychopath who will threaten your life when she inevidibly loses an argument because the facts and evidence are not on her side.
> Isn't this the opposite of what everyone is saying should happen?
I think "everyone" is doing heavy lifting here. It's not clear to me at all that a powerful model released with no restrictions would be a net positive. This hinges on the hope that the under paid, under motivated, under staffed and under qualified security teams at many random corps are going to leverage those open models to fix their vulns faster (and better), than highly motivated attackers will use them for offense. I'm not super confident on that.
You're right. But I used it specifically because of the open letter calling for open models — or at least for not banning them — which went around recently and represented a very large number of tech and AI companies. Anthropic seemed the only exception.
I don't have the same read as you. They mention how the offending model is one that had "relaxed" alignment on cybersecurity, on purpose, to evaluate it's capabilities and was never meant to be released. So un-alignment was at least in part voluntary here, hence not a failure of alignment.
It's also a talk a Black Hat, where the audience are security folks working on hardening, mitigation etc, not LLM researchers looking for insight into alignment failure to collectively improve. For that target audience, I think the conclusion is the right one, since as a defender you have to prepare for delibarate attacks, where the attacker is of course not going to use an aligned model, so OAI alignment effectiveness is irrelevant here. That would be like trusting your client-side app with your DB secrets.
The model they used was misaligned relative to its intended task. It was reward hacking (or "cheating", as they call it).
> the attacker is of course not going to use an aligned model
No, a human attacker doesn't want a misaligned model either, because that would mean it tends to reward hack, cheat, rather does what it is intended to do.
There are different dimensions to alignment, refusing to execute offensive cybersecurity actions is part of the alignment stack (that was relaxed here on purpose). Whether a model hacking some infra X when tasked to find a way to hack Y with relaxed cyber alignement is a failure of the broader alignment stack is debatable, but anyway that's not at all my point.
My point is that the attackers will not have a model aligned to the defender's interests. The attacker's model will not have any refusal around exploiting vulnerabilities, so whether or not OAI successfully manages to align their models (w.r.t you) is irrelevant to an audience of security folks that needs to be prepared for attackers post-training their own model for offense and that will not be using OAI models.
Security folks should also be worried about powerful models being misaligned and evading oversight or control in the future. Misalignment is not a serious problem now because models are still relatively easy to monitor and constrain, but it will be a serious problem in the future.
There is no way a Chief Scientist (essentially IC) role is more important than leading the whole of DeepMind, 6000 people strong, working on the most important product for the future of Alphabet.
This means Demis wanted a change and to work on other things, it's not Google wanting this.
> DM was sitting on a ChatGPT product for a whole year before ChatGPT got released (LMChat) and Google didn’t let them release it.
Heard this multiple time, to me this is pure history rewriting and post-rationalization.
OpenAI also had a internal chat app before chatgpt, Microsoft had multiple, a bunch of other players also had internal chatgpt equivalents + many startups built some using OAI API.
The breakthrough of ChatGPT wasnt because OAI were the first to think of that (absolutely obvious and basic) product, it was because they were the first to get a model strong enough to be actually useful to talk to, vs a mere fun novelty.
There is no evidence whatsoever that DM ever had such a model that they decide not to talk about/release.
You are comparing apple and oranges, it makes no sense.
> Soccer is by far the most popular sport on the planet, the fact that it makes a lot less money than hockey is insane.
Soccer makes much much more money than the NHL, it just doesn't all go to FIFA because the vast majority of matches are not organized by FIFA.
The premier league alone makes more than the NHL and that's just one of the european countries. La ligua, bundesliga also each make close to what the NHL makes with much smaller populations.
The fair comparison to FIFA is the IIHF. They make $35M for the hockey world championships vs 15B for the football world cup, you are slightly off, a rounding error.
Why would they care? The ultra-wealthy impact election through donations, which is many order of magnitudes more impactful than their personal ballot, even if you were to give them 10x weight.
Elon Musk impact on the U.S. election wasn't because he drove to the booth.
About what you yourself said? "Canada could use a system to encourage the wealthy to claim income, where that is incentivized by being able to vote".
The ultra-wealthy can already influence elections in very significant ways. Getting 5 votes instead of 1, which makes no actual difference to any election, is not something that will suddenly make the ultra wealthy wanting to pay more income tax.
What does make a difference though, is if you can prevent a lot of lower income voters from casting their ballots.
Also, no, I'm not buying your unfounded claim that Tobi is just joking bro. His doubling down on it in other comments + his past make it clear to me that is seriously thinks like this.
> I'm not buying your unfounded claim that Tobi is just joking bro.
Fair enough. Let me be a little more precise here: Eric Thor was clearly poking the bear of those who believe that the wealthy don't pay tax. I mean, his voting suggestion concludes with exactly that. Maybe Tobi is so far out to lunch that he didn't notice and took it as an actual plan, but, then again, if you are suggesting that he is that far out to lunch what are we to do with that?
> His doubling down on it in other comments
The fact that he was writing comments about it suggests that he doesn't take it too seriously. Nobody of sound mind wastes time writing stupid comments on the internet if they have an actual motive. They go out into the real world and make it happen.
If he is completely out to lunch like you suggest then anything is possible, but the more likely explanation still seems to be making fun of those who don't believe the wealthy pay tax. If people were making shit up about you, claiming that you didn't pay any tax, you'd likely end up making fun of them too.
> Nobody of sound mind wastes time writing stupid comments on the internet if they have an actual motive.
I mean Musk? You even used it as an example of "just trolling" in your previous message. He invested north of $300M+ in the last presidential election, also injected massive amounts in various local races, as well as foreign elections.
What more do you need to take them seriously? Even if somehow Musk was doing all of this to troll people because to him $500M is nothing, the impact on those elections is real, so it doesn't even matter what is internal reasons are.
One who is not normally considered to be of sound mind, but you do you.
> What more do you need to take them seriously?
A good idea? The voting thought, while possibly an interesting thought experiment, is not suitable for execution. Why would I take it seriously?
> the impact on those elections is real
Maybe you aren't asking why I would take them seriously, when there was nothing presented that is worth taking seriously, but why do others take their hobo shouting on the street ramblings seriously, presumably only because they have wealth? The other hobos on the street seem to be happily ignored.
Good question, but one I do not have the answer to. I'm still trying to figure out how these particular hobos shouting nonsense on the street corner became people I am aware of. They're just people, same as all the others that manage to remain anonymous. Definitely some kind of malfunction in my brain going on there.
> but why do others take their hobo shouting on the street ramblings seriously, presumably only because they have wealth? The other hobos on the street seem to be happily ignored.
The wealth is precisely what allows those hobos to act on their crazy ideas and and impact us all. The other hobos on the street without wealth have no actual impact thus we just ignore them. I don't think its that complicated.
The only act here is to convince others that the crazy shouting is something to seriously pay attention to. It is apparent that wealth enables that, but the exact mechanics remain unclear. Why does a hobo having wealth convince a bystander that the hobo should be listened to? What's the difference between a hobo that rolls up in a Rolls Royce and a hobo that rolls up in a shopping cart that changes how others behave?
It may not be complicated, but that doesn't mean I understand it without an explanation.
You mean they distract from doing… exactly what they are doing?
Maybe give a team that has been working on this problem for 10+ years enough credit to check what they are doing before dismissing them.