4 ms·

I truly believe one of the biggest dangers of AI in our not so distant future will be the reaction to a lot of the LLM spam introducing an internet realID issue

I truly believe one of the biggest dangers of AI in our not so distant future will be the reaction to a lot of the LLM spam introducing an internet realID issued by governments to keep track of Real People and the internet will become divided into a walled/non-anonymous garden and a hellscape of GPT content on the other side. I hope i'm wrong but it's one of those roads to hell that can get paved with good intentions.

SmoothBrain12 4y ago
<Click here if you're over 18>
- paulmd 4y ago
  [are you a real human being?] https://www.youtube.com/watch?v=kNdDROtrPcQ https://www.youtube.com/watch?v=kNdDROtrPcQ
holoduke 4y ago
I think AI will slowely encapsulate us individuals. Just like people wearing glasses to see better we integrate AI to be able to perform or even cope with society. Slowely more and more skills will be aided or replaced by AI. Till in the end we only have some brains left to guide the AI in everything we do in life.
randcraw 4y ago
Yeah, I think a primary challenge with LLMs is to somehow identify them. If we can't, the net will become swamped by fake humans bent on distorting human reality in whatever ways serve their masters' whims. Frankly, I think identification of LLM personas vs humans will be impossible. It's increasingly likely then that the rise of LLMs may introduce the decline of the internet as a way to connect with other humans. Soon nobody will want to take an internet entity seriously unless it can prove it's human. And I don't see any way to do that. So maybe LLM fake entities will be the straw that breaks the Net's back as a social medium and propels us back into the physical world? I hope so.
- Loughla 4y ago
  It is true that the rise of language models has the potential to introduce fake personas that could manipulate and distort human reality. However, I believe that identifying LLM personas versus humans is not impossible. There are techniques that can be used to detect synthetic content, such as detecting patterns in the language used or analyzing the response time. Additionally, researchers and companies that have been discussed on HN itself that are already working on developing tools to combat deepfakes and synthetic content. While it's important to remain vigilant about the potential negative impact of LLMs, it's also important to remember the positive ways in which they can be used. Language models have the potential to revolutionize industries such as healthcare and education by analyzing vast amounts of data and providing insights that can improve people's lives. Furthermore, I don't think that the rise of LLMs will necessarily lead to the decline of the internet as a social medium. Rather, it will likely lead to a greater emphasis on trust and transparency in online interactions. We may see the development of new technologies that enable people to verify the authenticity of online identities, or the rise of platforms that prioritize human-to-human interaction. Overall, I believe that while there are challenges associated with the rise of LLMs, there are also opportunities for positive change. It's important to continue to monitor the development of these technologies and take proactive steps to ensure that they are used in a responsible and ethical manner.
  - meghan_rain 4y ago
    @dang, should boring, ChatGPT-generated comments likes these be banned?
    - Loughla 4y ago
      That's fair (but please don't ban me for that @dang). I genuinely just wanted to see if anyone would figure out it was chatgpt generated. What tipped you off? Because I believe it's going to get harder and harder to figure out what's human and what's not.
      - meghan_rain 4y ago
modeless 4y ago
Yes, this is a good realistic danger too. Is anyone seriously working on ways to authenticate human-generated content online that don't destroy all possibility of anonymity? CAPTCHAs will soon be impossible. Watermarking or otherwise detecting LLM content is not going to be possible long term. Centralized ID services are unpalatable. Maybe some kind of web of trust would be the way to go.
- tzs 4y ago
  There are three properties I think we probably want. 1. I can prove to a site that I am a human without the site getting any information from the proof about which specific human I am. 2. I can visit the site multiple times and prove to them that I'm the same human, without them getting any information from the proofs about which specific human I am. 3. Any third party site that is involved in the proof does not get any information about which site I trying to prove my identity to. One way this could be done is by having ID sites that you are willing to prove your real identity to in a robust and way which certify your humanity to other sites in a way that has the three aforementioned properties. Good candidates for running ID sites would be entities like major banks that most of us have already verified our real identity to in a robust way. One way to do this might be something like this. When a site wants to check that I'm human they give me some random data, and I have an ID site that I have an account with such as my major bank do a blind signature of that data, using a blind signature system the allows the blinder to remove the blinding. When I get back the signed blinded data, I remove the blinding and then send the data and unblinded signature back to the site I'm trying to prove my humanity to. They see that the data is signed by my major bank's ID service, so know that the bank believes I'm human.
avgcorrection 4y ago
Why government-issued? Could just as well become a de-facto corporate monopoly by Google or someone else. Is Gmail the “standard” email or is it a government-backed one? This problem has been coined “Digital Feudalism”. And medieval feudalism (or at least our popular conception of it) wasn’t driven by governments.

SmoothBrain12 4y ago

paulmd 4y ago
[are you a real human being?] https://www.youtube.com/watch?v=kNdDROtrPcQ https://www.youtube.com/watch?v=kNdDROtrPcQ

holoduke 4y ago

I think AI will slowely encapsulate us individuals. Just like people wearing glasses to see better we integrate AI to be able to perform or even cope with society. Slowely more and more skills will be aided or replaced by AI. Till in the end we only have some brains left to guide the AI in everything we do in life.

randcraw 4y ago

Yeah, I think a primary challenge with LLMs is to somehow identify them. If we can't, the net will become swamped by fake humans bent on distorting human reality in whatever ways serve their masters' whims. Frankly, I think identification of LLM personas vs humans will be impossible. It's increasingly likely then that the rise of LLMs may introduce the decline of the internet as a way to connect with other humans. Soon nobody will want to take an internet entity seriously unless it can prove it's human. And I don't see any way to do that. So maybe LLM fake entities will be the straw that breaks the Net's back as a social medium and propels us back into the physical world? I hope so.

Loughla 4y ago
It is true that the rise of language models has the potential to introduce fake personas that could manipulate and distort human reality. However, I believe that identifying LLM personas versus humans is not impossible. There are techniques that can be used to detect synthetic content, such as detecting patterns in the language used or analyzing the response time. Additionally, researchers and companies that have been discussed on HN itself that are already working on developing tools to combat deepfakes and synthetic content. While it's important to remain vigilant about the potential negative impact of LLMs, it's also important to remember the positive ways in which they can be used. Language models have the potential to revolutionize industries such as healthcare and education by analyzing vast amounts of data and providing insights that can improve people's lives. Furthermore, I don't think that the rise of LLMs will necessarily lead to the decline of the internet as a social medium. Rather, it will likely lead to a greater emphasis on trust and transparency in online interactions. We may see the development of new technologies that enable people to verify the authenticity of online identities, or the rise of platforms that prioritize human-to-human interaction. Overall, I believe that while there are challenges associated with the rise of LLMs, there are also opportunities for positive change. It's important to continue to monitor the development of these technologies and take proactive steps to ensure that they are used in a responsible and ethical manner.
- meghan_rain 4y ago
  @dang, should boring, ChatGPT-generated comments likes these be banned?
  - Loughla 4y ago
    That's fair (but please don't ban me for that @dang). I genuinely just wanted to see if anyone would figure out it was chatgpt generated. What tipped you off? Because I believe it's going to get harder and harder to figure out what's human and what's not.
    - meghan_rain 4y ago

modeless 4y ago

Yes, this is a good realistic danger too. Is anyone seriously working on ways to authenticate human-generated content online that don't destroy all possibility of anonymity? CAPTCHAs will soon be impossible. Watermarking or otherwise detecting LLM content is not going to be possible long term. Centralized ID services are unpalatable. Maybe some kind of web of trust would be the way to go.

tzs 4y ago
There are three properties I think we probably want. 1. I can prove to a site that I am a human without the site getting any information from the proof about which specific human I am. 2. I can visit the site multiple times and prove to them that I'm the same human, without them getting any information from the proofs about which specific human I am. 3. Any third party site that is involved in the proof does not get any information about which site I trying to prove my identity to. One way this could be done is by having ID sites that you are willing to prove your real identity to in a robust and way which certify your humanity to other sites in a way that has the three aforementioned properties. Good candidates for running ID sites would be entities like major banks that most of us have already verified our real identity to in a robust way. One way to do this might be something like this. When a site wants to check that I'm human they give me some random data, and I have an ID site that I have an account with such as my major bank do a blind signature of that data, using a blind signature system the allows the blinder to remove the blinding. When I get back the signed blinded data, I remove the blinding and then send the data and unblinded signature back to the site I'm trying to prove my humanity to. They see that the data is signed by my major bank's ID service, so know that the bank believes I'm human.

avgcorrection 4y ago

Why government-issued? Could just as well become a de-facto corporate monopoly by Google or someone else. Is Gmail the “standard” email or is it a government-backed one? This problem has been coined “Digital Feudalism”. And medieval feudalism (or at least our popular conception of it) wasn’t driven by governments.