4 ms·

> - book me a highly rated hotel in Berlin around 100 euro p/n most suitable for a staycation for next week Monday to Friday. Make a reservation near that hotel

by rchaud 4y ago

> - book me a highly rated hotel in Berlin around 100 euro p/n most suitable for a staycation for next week Monday to Friday. Make a reservation near that hotel for the highest rated traditional German cuisine restaurant with a price per plate of 20-30 euro I imagine this will go about as well as if you chose to purchase an apartment purely on the basis of the real estate agency's description.

ralfn 4y ago
This is what I do now on booking.com. If you have a smarter strategy let me know. But whatever the strategy is: it can be automated. I also spent a lot more time on the choice where I live than the hotel in going to stay at for a week. Weird. And although ChatGPT currently doesn't have up to date information about hotels or the ability to book it (a simple booking.com plugin would enable that though), it is smart enough to either pick a hotel with remote working facilities or consider one with coworking spaces nearby.
- rchaud 4y ago
  I don't know if it's a smarter strategy, but you can send a travel agency your list of requirements and they'll send you back a quote. My parents' generation are probably the only people that still use agencies, but it works for them. It takes longer than a search query, but they don't want the hassle of coordinating logistics, and have some peace of mind that they're getting what they asked for.
  - ralfn 4y ago
    So what's the bussiness model of the travel agency? - they take a cut - they are neutral because they want repeat bussiness - they are personalised around your requirements - they take care of the little details It's exactly what I'm describing here, except automated. Because at heart their work is analyst work and thats exactly what you can automate with a LLM
    - ezst 4y ago
      What makes you think those results would be neural, and not sponsored/SEOed the same way everything is these days? What makes you think the machine works for you and not the advertisers? Also, to this day, LLMs/ANNs are not capable of explaining their outcome, and you won't argue/reason with them. That's a pretty big difference if you ask me.
      - ralfn 4y ago
        >What makes you think those results would be neural, and not sponsored/SEOed the same way everything is these days? Because it would be the less profitable bussiness. >Also, to this day, LLMs/ANNs are not capable of explaining their outcome, and you won't argue/reason with them How would I be able to tell that your response is human and not a LLM? I definately can argue within a chat thread. They can also provide a plausible sounding explanation. We know due to the nature of the technology that that explanation (sounding rational, deductive) isn't the truth. We are just not all ready to accept all the same shortcomings also apply to humans.
        nico 4y ago
        Very well said. Thank you for the great thread.
        nwienert 4y ago
        > Because it would be the less profitable bussiness. How is this different from Google?
        ezst 4y ago
        >> What makes you think those results would be neural, and not sponsored/SEOed the same way everything is these days? > Because it would be the less profitable bussiness. How so? Considering that almost everything on the web right now runs on ad revenue, why would they pass on the opportunity to cash on that if they do no worse than the rest? Especially considering how lucrative those referrals might be, with our lack of preparedness countering those with traditional ad blocking techniques and being more susceptible due to inexperience. > We are just not all ready to accept all the same shortcomings also apply to humans. Because they do not. If you hire a human to do a job for you, you are bound by a contract (and protections, and recourses and enforcement). In the end you will be able to know whether you got robbed into a referral scheme and that's bad for business. With this kind of AI, the result is all you get (technically there is no way to explain it), which gives unfair deniability. Regulators might end up ruling whole classes of algorithms as unfit for purpose/deceptive, and I would personally applaud that. We are just at the beginning of all that and we should focus on developing the technology so it keeps working for us reliably, and not against us or under undetermined circumstances.

by rchaud 4y ago

ralfn 4y ago

This is what I do now on booking.com. If you have a smarter strategy let me know. But whatever the strategy is: it can be automated. I also spent a lot more time on the choice where I live than the hotel in going to stay at for a week. Weird. And although ChatGPT currently doesn't have up to date information about hotels or the ability to book it (a simple booking.com plugin would enable that though), it is smart enough to either pick a hotel with remote working facilities or consider one with coworking spaces nearby.

rchaud 4y ago
I don't know if it's a smarter strategy, but you can send a travel agency your list of requirements and they'll send you back a quote. My parents' generation are probably the only people that still use agencies, but it works for them. It takes longer than a search query, but they don't want the hassle of coordinating logistics, and have some peace of mind that they're getting what they asked for.
- ralfn 4y ago
  So what's the bussiness model of the travel agency? - they take a cut - they are neutral because they want repeat bussiness - they are personalised around your requirements - they take care of the little details It's exactly what I'm describing here, except automated. Because at heart their work is analyst work and thats exactly what you can automate with a LLM
  - ezst 4y ago
    What makes you think those results would be neural, and not sponsored/SEOed the same way everything is these days? What makes you think the machine works for you and not the advertisers? Also, to this day, LLMs/ANNs are not capable of explaining their outcome, and you won't argue/reason with them. That's a pretty big difference if you ask me.
    - ralfn 4y ago
      >What makes you think those results would be neural, and not sponsored/SEOed the same way everything is these days? Because it would be the less profitable bussiness. >Also, to this day, LLMs/ANNs are not capable of explaining their outcome, and you won't argue/reason with them How would I be able to tell that your response is human and not a LLM? I definately can argue within a chat thread. They can also provide a plausible sounding explanation. We know due to the nature of the technology that that explanation (sounding rational, deductive) isn't the truth. We are just not all ready to accept all the same shortcomings also apply to humans.
      - nico 4y ago
        Very well said. Thank you for the great thread.
      - nwienert 4y ago
        > Because it would be the less profitable bussiness. How is this different from Google?
      - ezst 4y ago
        >> What makes you think those results would be neural, and not sponsored/SEOed the same way everything is these days? > Because it would be the less profitable bussiness. How so? Considering that almost everything on the web right now runs on ad revenue, why would they pass on the opportunity to cash on that if they do no worse than the rest? Especially considering how lucrative those referrals might be, with our lack of preparedness countering those with traditional ad blocking techniques and being more susceptible due to inexperience. > We are just not all ready to accept all the same shortcomings also apply to humans. Because they do not. If you hire a human to do a job for you, you are bound by a contract (and protections, and recourses and enforcement). In the end you will be able to know whether you got robbed into a referral scheme and that's bad for business. With this kind of AI, the result is all you get (technically there is no way to explain it), which gives unfair deniability. Regulators might end up ruling whole classes of algorithms as unfit for purpose/deceptive, and I would personally applaud that. We are just at the beginning of all that and we should focus on developing the technology so it keeps working for us reliably, and not against us or under undetermined circumstances.