Hacker Newsnew | past | comments | ask | show | jobs | submit | pffft8888's commentslogin

They can do this in people, too, not just LLMs.

Imagine the mistakes that can be made by changing one fact but not reconfiguring the whole network.

Thhese guys remind me of when I used to change EXEs in hex editors then notice "unrelated" weird glitches.


That’s what I’d be afraid of, though I haven’t read the relevant literature so maybe this is addressed. Since there are way fewer parameters than encoded facts it seems like it would be difficult to change one fact without messing with others.


Feeling left out?


You can't trust what sama says. Period. He's shown his willingness to deceive for gain. He deleted the codex model because it was better than the text model and only way to explain it is that by training on code instead of just text of human interactions the model develops stronger reasoning abilities (because of higher clarity around logic and higher signal to noise overall in the training data.) So he removed that so researchers don't catch on. Also, why 1 Trillion params is the limit? Why not 2 Trillion. He's been begging for regulations and trying to pull up the ladder behind him. Just really "street smart" thinking but I have not seen any elevated humanity serving thinking coming form him. He would not have taken $100M for a non-profit to do open source research only to bait-and-switch into for profit closed source sold to the highest bidder. Give me a break. Don't believe anything he says (well, use your brain, as he may mix truths and deceptions with the goal of deceiving his competitors... he's not out there to do a service for humanity... if he was, he would have kept it open and non-profit, not basically steal the funds of the non-profit to pocket $29B from it.) It's ridiclous.


The propaganda against encryption is in full swing.

My expectation is that all NSA CNSA[1] encryption standards are backdoored at the implementation level (by the NSA who uses Suite A for its own communication and I suspect military communications outside of that in weapons systems that can fall into enemy hands)

I guess the propaganda is driven by FBI and law enforcement agencies.

1. https://en.wikipedia.org/wiki/Commercial_National_Security_A... 2. https://en.wikipedia.org/wiki/NSA_Suite_A_Cryptography


Can someone explain me why this is downvoted ? In my understanding his proposition about NSA is quite close to a popular one and hn seems to allow discussion of hypothesis - if they are more probable than imaginary ?

Is it the word propaganda that patriots dislike ? Not sure if some soviet connotation is involved in US but for me it’s just a synonym of “public lobbying” of “ideology gov marketing”.

I know those subjects can become polemic and I don’t want to throwing oil on the fire, but an “out of debate” clarification would be nice and helpful.


The worst thing about HN (and it does reflect badly on YC as a whole, at least for me) is how they enable people to act in seemingly passive aggressive ways. Instead of stating disagreements, they downvote, and you'll never know why. Just pure crappy behavior. In this case, someone explained below that they downvoted because they don't agree that the article is propaganda and that it calls for less backdoors or something like that as if everything isn't backdoored already, one way or another.

Then you have stuff like BIP39 protecting people's money (cryptocurrency) that can be cracked for $350/hr on GPU rigs. Someone even wrote a how-to.

Current security makes it harder, but not sufficiently harder, to break into systems. I mean... HN crowd is probably high schoolers and non-tech people just out here to argue.


> Then you have stuff like BIP39 protecting people's money (cryptocurrency) that can be cracked for $350/hr on GPU rigs

This doesn't appear to be true (in the sense that yes it is feasible to crack 4-word BIP pass phrases, but all wallets that I'm aware of use at least 6 words, which is estimated to take 11 years for a hypothetical ASIC cracker)

https://coldbit.com/can-bip-39-passphrase-be-cracked/

Perhaps you are meaning this attack where someone was able to brute-force 4 words from a 12 words phrase. It matches your $350 cost, but of course is dramatically different to "cracking BIP39": https://medium.com/@johncantrell97/how-i-checked-over-1-tril...


2048 words in 6 positions is simply not enough entropy for the NSA's encryption cracking infrastructure. If it is worth it they'll crack it. The NSA does not use a single ASIC cracker.


That's an extremely different proposition. Pretty sure the NSA isn't into stealing bitcoins.


That's 66 bits of entropy. With a quantum computer having 66 logical (error corrected) qubits, the pass phrase can be cracked in under a day.

That's not too far off, maybe a few years before one is commercially available.


It's because it's another conspiracy theory unsupported by evidence.

The encryption algorithms in CNSA are broadly accepted by the security community. Just saying "NSA backdoor" is a cheap shot.


> unsupported by evidence

Depending on what you accept as an evidence, but this theory is surely supported by precedent(s?) [0]

Just saying “another conspiracy theory” is a cheap shot : conspiracy are bad and should be fought. Theories are a useful process to make knowledge advance. Conspiracy theories are often discussed in an awful way on social medias, can’t HN do better than just downvoted them ?

[0] https://www.theguardian.com/books/2014/may/12/glenn-greenwal...



See this is a great example of the problem.

1) Did NSA modify Cisco routers? Yes

2) Did NSA get a backdoor in Dual EC_DRBG? Yes

3) Did NSA get a backdoor in the CNSA algorithms? There is no evidence to suggest they have and plenty the other way.

The original claim was (3) but the "NSA does stuff" thing overrides any attempt at discussion of that specific piece of misinformation.


>It's because it's another conspiracy theory unsupported by evidence.

I'm having a hard time keeping up with it all, it's nuts. But my understanding is that the NSA backdooring protocols is totally supported by evidence? We saw it in the Snowden revelations? RSA being the company nobody will ever trust again?

Is that all wrong somehow?


> backdooring protocols is totally supported by evidence

It's important to be very precise.

I think you might be confusing backdooring specific pieces of software produced by RSA-the-company (specifically things using Dual EC_DRBG) with the RSA algorthim that company is named after, which is included in the CNSA.

Dual EC_DRBG was a bad algorithm which many people had serious doubts about from the start - and indeed it was backdoored by NSA. That is different to the algorithms in CNSA which (as I said earlier) are well regarded by the same security researchers.

There is no evidence (or serious claims) that the RSA-algorithm is backdoored.


get it from the horse's mouth, as they say... instead of baselessly pontificating on HN and not understanding the diff between algorithm and implementation


not the algorithm. the various implementations of it. evidence? ask a friend.


It was an interesting read, moral to me is not to use Cell Phones for anything illegal. If you do not control the keys, you might as well not bother with encryption.


Even if you control the keys, it does not protect you from vulnerabilities somewhere in the stack. Stuff like thumbnail generation provided by the OS has been used by cyber-criminals in the past to compromise phones by sending MMSes or even third-party messenger apps, and I'd take a guess and bet that at least the Five Eyes government agencies all have a sizeable cache of baseband vulnerabilities.

Technology simply has become far too complex to be reasonably secure, even if you have the financial firepower of being Apple, Sony, Microsoft, Nintendo or Amazon.


Agreed.

If it installs updates without requiring you to specify the desired updates explicitly (i.e. by cryptographic hash), it should not be considered "your" device from a security perspective.


This 100%.


> My expectation is that all NSA CNSA[1] encryption standards are backdoored at the implementation level (by the NSA who uses Suite A for its own communication and I suspect military communications outside of that in weapons systems that can fall into enemy hands)

CNSA / NSA Suite B are pretty much entirely public encryption standards that have stood up to public scrutiny for decades at this point.

They are also approved by the USA to encrypt TS SCI information, why would they approve that if they had backdoors?.


Why would they have Suite A then?


> Why would they have Suite A then?

Because Suite A are a set of non public algorithms that are used to encrypt data.

Them being non public makes it harder to workout how to decrypt the data.

They are also likely protected against a number of attacks that aren't public, even if these attacks aren't feasible against current algorithms.


>They are also likely protected against a number of attacks that aren't public

For reference, see DES, where the NSA adjusted the algorithm to protect from a not publicly understood differential cryptanalysis attack. Many people claimed that the adjustment by the NSA was clear backdooring, though we know that was not true.

It was however purposely deficient in the length of its key, allegedly because "it was good enough" and for export reasons, but also because the NSA considered it easy enough to brute force.


"even if these attacks aren't feasible against current algorithms" lol, ok, if you say so....


This article isn't spreading any propaganda against encryption. If anything, it makes the case that new backdoors are not needed.


Says a person who is online...


Drinking one glass of wine is not the same as being passed out on the sidewalk at noon.


But the person is complaining about people complaining. I find it ironic.


Why do we feel this need to disect everything


Probably for the same reason people feel the need to complain about people complaining... and then complain about people complaining about people complaining, and so on.


That feels like some sort of survivor bias. The majority of people didn't feel that and passed on by without commenting.


Analysis is the most straightforward path to resolving disagreement. Get enough people talking and you'll have disagreement. The immediate result is analysis.


This is not an analysis at all, people online have been misusing the word "ironic" forever. Almost nobody uses it correctly.

It's a way to shut down debate while pretending to use a legitimate debating technique. A tale as old as the internet itself.


I was responding to the question, "why do we feel the need to pick everything apart?", not this other comment. I'd point out your response was an analysis of the use of the term "irony" in online discourse.


[flagged]


I liked the phrase “performative erudition” the New Yorker article[0] used to describe HN’s social theatre.

[0] https://www.newyorker.com/news/letter-from-silicon-valley/th...


You just got validated. Does that mean you're desperate? or anxious? or even sick in the head?


It is all part of the Human Condition.


The sky is falling... etc. It is true, however, that AI is going to reduce a team of 10 to 1, regardless of what field they're in. One day soon we will see companies run by an AI employing umans for 1:1 sales only, until we become more used to being sold stuff by bots. We're wittnessing the birth of a new age, and people are talking about web developers.


I’ve been flirting with being a human liaison for autonomous organizations for a decade

ten years ago once we got the fabs going for bitcoin miners we noticed that machines could make other machines, acquire a fungible digital resource and transact with other agents, human and machine alike. the window of opportunity turned out to always be too small in the mining world, but that theoretical piece of the puzzle was suddenly solved.

AI that starts making money on its own accounts is going to come next. Someone could do this with Llama hooked up to a node. Maybe start by moderating discord servers for a few hundred dollars of crypto per month, just like children do.


Yes. I'm just surprised at the downvotes. Hmm. We're talking about jobs, right? Everyone's job is very much at risk unless people adapt to a P2P economy with humans and AIs in it, trading with and employing each other.


I take that back I'm suyrprised. HN crowd is mostly very pedestrian.


Everyone here aspires to work for an ad conglomerate that's busy destabilizing democracies while they "own" the compose button in an app. Of course its very pedestrian.


Bitcoin is based against the wisdom that "all that glitters is not gold." It abuses the "shiny object" trigger in our animal brain. The minute it looks shiny, its price goes up, and vice versa.

It's become a way for people to manage their anxiety around the state of the economy, pumping money into Bitcoin whenever it looks shiny relative to the fiat currency, and getting out when it doesn't, or when it stops going up (no more shiny attraction.) You need a sustained campaign of lies and deception to keep polishing that turd, so to speak.

We're better off without it, but you can't take away some people's psychological safety blankie.


Altman is v2.0 of Zuckerberg, with 1000X more parameters.

I'm worried about v3.0


Weird coincidence, I tweeted roghly the same idea two weeks ago:

https://twitter.com/marcfawzi/status/1636115903959158785

Obviously, it's a no-brainer idea at a high level. The devil is in the details.


AI is not a weapon. It's ultimately a synthetic brain running inside a machine. It can be used for good, or it can be mounted on a killer drone with deadly lasers. If you use it for evil, you go to jail. It should be as simple as that.

Don't Use AI for Evil. Make that into law.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: