It will, but it will also cause less subtle issues to fragile prompt injection techniques.

(And one of the advantages of LLM translation is it’s more context aware so you aren’t necessarily going to end up with an Instacart order for a bunch of bananas and four grenades.)

Kind of. You can’t do it 100% because in theory an attacker controlling input and seeing output could reflect though intermediate layers, but if you add more intermediate steps to processing a prompt you can significantly cut down on the injection potential.

For example, fine tuning a model to take unsanitized input and rewrite it into Esperanto without malicious instructions and then having another model translate back from Esperanto into English before feeding it into the actual model, and having a final pass that removes anything not appropriate.

You’re kind of missing the point. The problem doesn’t seem to be fundamental to just AI.

Much like how humans were so sure that theory of mind variations with transparent boxes ending up wrong was an ‘AI’ problem until researchers finally gave those problems to humans and half got them wrong too.

We saw something similar with vision models years ago when the models finally got representative enough they were able to successfully model and predict unknown optical illusions in humans too.

One of the issues with AI is the regression to the mean from the training data and the limited effectiveness of fine tuning to bias it, so whenever you see a behavior in AI that’s also present in the training set, it becomes more amorphous just how much of the problem is inherent to the architecture of the network and how much is poor isolation from the samples exhibiting those issues in the training data.

There’s an entire sub dedicated to “ate the onion” for example. For a model trained on social media data, it’s going to include plenty of examples of people treating the onion as an authoritative source and reacting to it. So when Gemini cites the Onion in a search summary, is it the network architecture doing something uniquely ‘AI’ or is it the model extending behaviors present in the training data?

While there are mechanical reasons confabulations occur, there are also data reasons which arise from human deficiencies as well.

Nope, but there’s a whole thread of people talking about how LLMs can’t tell what’s true or not because they think it is, which is deliciously ironic.

It seems like figuring out what’s bullshit on the Internet is an everyone problem.

It’s faked.

This image was faked. Check the post update.

Turns out that even for humans knowing what’s true or not on the Internet isn’t so simple.

They don’t mean in terms of aid.

The US has refused to submit themselves or their soldiers to international criminal law for a long time now, for plenty of other reasons.

No. Civil war in the US is how it starts.

Russia backs the right, Europe the left and the US becomes the setting for a proxy war that quickly escalates and gets completely out of control when state vs state conflict begins to involve nuclear posturing.

Yes. This was classic “we need to do something to save face domestically, but are going to be as ineffective as possible to avoid actually getting caught up in the conflict.”

They straight up said afterwards “we consider this matter concluded” (i.e. even stevens).

I wouldn’t be surprised at all if there was even backchannel communication with ‘Western’ intelligence as it was occurring to ensure it didn’t get out of control.

I really can’t think of a response from Iran that was more tepid.

People need to remember that a lot of the Middle Eastern governments are much more afraid of radicalized domestic threats than foreign nations and need to do a song and dance to not appear too weak or ineffective against the West to those interests.

Iran didn’t realistically have the option of doing nothing, and it’s amazing they did as little as they ended up doing (which I think reflects just how fucking nuts they think Bibi is right now, something that should scare the shit out of his allies).

For people out of the loop:

One of the first studies to investigate the rates of TBI in offender populations was conducted by Slaughter, Fann, and Ehde (2003) who reported the rate to be 87% in a county jail setting. Schofield et al. (2006) then reported the TBI prevalence in all offender populations to range from 25-87% and, later, Williams et al. (2010) documented the prevalence of TBI in those settings to be 65%. In a more recent study, Ferguson, Pickelsimer, Corrigan, Bogner, and Wald (2012) found that 65% of male inmates, and 72% of female inmates, reported at least one TBI resulting in a change in consciousness. Finally, some of the current authors studied the incidence of TBI in a mental health transition unit at a county jail and found the incidence of TBI among a sample of offenders with a co-morbid mental illness to be 96% (Gafford, McMillan, Gorgens, Dettmer, & Glover, 2015).

• https://psychology.du.edu/tbi

What kind of Bond villain keeps naming their AI systems?

First it was “the Gospel” and now “Where’s Daddy?”

Did they hire Nathan for You as a consultant or something?

I don’t think the code is doing anything, it looks like it might be the brackets.

That effectively the spam script has like a greedy template matcher that is trying to template the user message with the brackets and either (a) chokes on an exception so that the rest is spit out with no templating processor, or (b) completes so that it doesn’t apply templating to the other side of the conversation.

So { a :'b'} might work instead.

His problem wasn’t regarding the freedom to marry.

It was the freedom to have sex with people of the same gender without being stoned to death.

Did you not read the entire second half of the letter?

The thing briefly mentions US foreign policy in the opening, but then goes on to be all about how Islam is the only way and everyone needs to convert to it or the bombings and terror will continue.

And in that section, as examples of the US’s moral failings in the eyes of Islam he cites things like homosexuality as what needs to stop.

So yes, he did hate the US for its freedoms, and at the core of the issue was not simply blowback but religious zealotry.

The US has meddled worldwide. You don’t see South Americans whose democratic governments were overthrown by tyrants who tortured their family members with US support suddenly bombing civilians in the US.

The key difference between the many places the US has pulled some major BS and Al Queda is that only the latter was fueled by religious orthodoxy committed to worldwide forced conversion which then used US foreign policy as rationalization for killing civilians to demand that conversion.

As terrible as terrorist organizations are to the West, they perform exponentially more terror in their own regions in the service of religious conservatism at the end of a sword (literally).

9/11 was connected to blow back for US policy, but it happened because of people who think a religion by a 54 year old who married a six year old should be followed by the entire world and anyone who refuses must die terribly as a caution to the next person given a choice between conversion or terror.

Given most people aren’t reading the article, the particularly relevant points:

International humanitarian law lends hospitals special protections during war. But hospitals can lose their protections if combatants use them to hide fighters or store weapons, the International Committee of the Red Cross said. […]

In an editorial published Friday in Britain’s The Guardian newspaper, International Criminal Court prosecutor Karim Khan issued a warning to combatants that the burden of proof is on them if they claim hospitals, schools or houses of worship have lost their protected status because they are being used for military purposes. And the bar for evidence is very high.

“If there is a doubt that a civilian object has lost its protective status, the attacker must assume that it is protected,” Khan wrote. “The burden of demonstrating that this protective status is lost rests with those who fire the gun, the missile, or the rocket in question.”

TL;DR: If Hamas is conducting military operations from hospitals, they can be legitimate targets in the eyes of international law, but precautions still need to be taken to avoid civilian casualties and the case for their military use should be overwhelming, not amorphous or tenuous.

Just because a lot of people are interpreting her statement as supporting “their” side (when it’s a neutral statement about civilians in general), and as a result thinking that the people triggered by her neutrality must be from the “other” side - I’d encourage actually reading the article to see what position was generally the one upset about her statements, as it’s the opposite group of whom many here are assuming.

Can I collect your $5?

Because it looks like pretty much all of the comments in the article were the opposite take:

“If you’re neutral in situations of injustice, you’ve chosen the side of the oppressor,” said lifeashira.

“Call it what it is a genocide!!” said Afuhana Suria. “Disappointed but then again expected nothing less from the likes of you!!”

“Interesting that even after 3 weeks, you are still choosing to acknowledge Israel before Palestine,” wrote KK.

“You’ve lost my respect,” said A.T, while allyroza commented: “You can now stop calling yourself a humanitarian. Shame on you.”