I’ve seen a few examples that breathlessly proclaimed that they literally saved lives or saved hours of time, but when I typed the 3 most obvious keywords from their input into Google, the first result always has the answer right there.
I need either information, a service or a product. Or information about a service or a product.
I never need a search, a website or an app. That's never the problem I'm trying to solve.
Search, websites and apps are the tools I have to use (currently) to solve my problems. In the process I have to:
- deal with the fact that the interest of the makers of the app or website do not align with mine
- deal with privacy invasions/ads/notifications
- deal with things being phrased disingenuously
- deal with apps needing to update, website being broken/bandwidth wasted
And most importantly, unless two vendors implemented an integration together, I have to integrate their services myself!
I never want to leave the chat. I want to do all my transactions from the chat. All of them. On my terms. With my prompts.
Whoever facilitates that can take a 5-30% cut on all my online transactions.
Examples:
- 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
- send every contact ive emailed with from my work email in the last year a Christmas card with a funny but professional pun about the invention of the steam engine. Send it to their work address.
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.
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.
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.
- 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
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.
How is this different from Google?
> 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.
The devil is always in the details. A travel agency builds contacts with airlines, hotels, tour groups etc, so they can have a backup option available if your first preference is not. That requires humans.
An AI system that is simply browsing through the same websites a human would is going to come up with a lot of "no" responses for specific booking requests, or will simply quote higher prices once the booking websites figure out how to detect them.
Perhaps they will try to up the price on this platform compared to others. But only until AirBnB comes to compete on this platform too (and uploads their API integration)
It's a platform game.
There was a time when Google tried to index everything and afterwards came a time when everything tried to be indexed by Google. Once you are the default, they have to be there and they will have to compete with others there.
And as long as the chat protects users from losing control over their intent, that's where the users will be and that's where you will need to be offering your products to be viable. Obviously at that point the regulators come knocking..
Assuming there would only be one player. There won't be, but for now only OpenAI is actually aware of the opportunity and Microsoft doesn't even seem to understand it's a platform game now and not an ad game.
Google on the other hand, is at a point where even Yandex of all search engines is starting to give better results, especially when it comes to image searching and content Google doesn't like as much like piracy. Google used to be amazing back in the day, nowadays I find myself having to specify REDDIT (or a specific forum) on my searches because the top results without it are hardly ever good.
The search was something like: "Polish supermarket with red and white logo and a bird in the logo, that begins with the letter K".
The first google result to wikipedia has it, to be fair, but without this back and forth idk if I would have found it.
P.S. It was not Biedronka! it was Auchan, and I thought Auchan started with K, because of the bird.
> We need to travel by train from Rome to Montpellier, but it is too many stops to do at once. So we would like to stay somewhere for a day in between Rome and Montpellier. What are our options? It needs to be safe, not adding a lot of extra travel time and suitable for a workathon
It returns with four options, approximate travel times and advantages and disadvantages. It understood coworking cafes or bussiness hotels would be a requirement.
> What are all the places an SSL certificate could be cached when calling an API from a Jupiter notebook?
It mentioned some obvious things such OS/browser/network-device, but also that a particular python package exists that is doing this at the application level as well, but only in memory and that a restart of the Jupiter kernel would resolve that.
Now, I'm curious with what kind of keywords I could get meaningful output from Google for these two examples.
What i suspect is happening is that you are micro managing your chat as if it's search. Picking keywords rather than just telling it the problem you are actually trying to solve. Splitting up your problem in sub problems and then trying to find each peace of information manually and combining it yourself to form a conclusion.
This trust in it's response reminds me early days of the internet. That would be the time when the people who created it were just exploring what the technology could do. It was a wonderful time. People posted amazing HOWTO's to Usenet that were chock full of useful information, you only received email from like minded friends, and almost nothing was censored and most of it was useful. It looked like the future was bright.
But then it grew so big you could not recognise most of the sources, and into that void stepped content aggregators like MySpace and Facebook I refuse to go near, and Alta Vista started ranking stuff by how much they were paid. Usenet was overrun by lies from spammers as was my inbox. And now if I don't run an ad-blocker, I find it difficult to differentiate Google's search results from the ads at the top, and near bloody impossible on sites like Amazon, and GMail / Hotmail mines my personal correspondence for ways to manipulate me.
And so it will be with those concise AI responses. Except you have absolutely no way to cross verify - you only get what a machine that will likely be far smarted than any of us wants you to know, in order to maximise it's profit.
Thank you, but I'll pass.
> Florence, Italy: Florence is a beautiful city with a rich history and culture. It is only a 90-minute train ride from Rome, making it a convenient stopover point. Florence also has plenty of cafes, co-working spaces, and other amenities that are conducive to a workathon.
Mistake: The purpose is to split up a long journey. Florence would split it into a 1.5h day and a 10.5h day.
> Marseille, France: Marseille is a bustling port city on the Mediterranean coast of France. It is about a 3.5-hour train ride from Rome and has a vibrant cultural scene and a variety of co-working spaces and cafes. You could spend the day exploring the city's many museums, parks, and markets.
Mistake: Factually wrong about the travel time, per Google maps GPT is off by a factor of 3.
Mistake: The purpose is to split up a long journey. Marseille would split it into a 10.5h day and a 2h day.
Mistake: Forgot about the workathon requirement
GPT3.5 suggested a list of Florence, Nice, Genoa and Marseille. When I asked it for a breakdown of the travel times, it got them pretty close ("Can you give me rough travel times for each of these options?") given the large variations in travel time depending on specific timing and transfers for several of these.
When I then asked it for somewhere closer to the midway point, it didn't come through (it suggested Pisa, which is about as bad as Florence), Turin, which is an alternative to Genoa, but doesn't fix the imbalance, or Lyon which is about as bad as the Pisa option but in the opposite direction (fast train to Montpellier), but the options also aren't in aggregate much worse.
The problem here is that there aren't as far as I can tell any very balanced options. You can try to do e.g. Ventimiglia or Sanremo (which GPT4 gave when I pressed it on splitting it more evenly), which will give you a more evenly distributed travel time, but because the overall travel time will be longer they're not at all obvious options.
You also seems to focus on catching it out rather than getting a result. If you want precision from the first query, then you need to give far more precise questions.
You need to treat it as a conversation, not a brief to be answered with a report. If you insist on treating it as a brief you will not get good results out of it. Your loss, in that case.
For cases where a simple lookup will work, it's a waste. Just use Google. For cases where there is in fact not a single perfect option, and where you need to weigh pro and con, it works well with the caveat that you do indeed need to be careful and check specifics. The same way I'd check specifics if I had a conversation about this with a friend.
With plugins, so it can verify precise details we can expect a significant leap in capability here.
But even for now, the answers I got to this were ones I was happy with, and another reason for me to use it more.
> You also seems to focus on catching it out rather than getting a result.
Nope. I have nothing against GPT. I find Copilot useful enough to pay for. I'm just sick and tired of people promoting it with obviously wrong examples.
The story of Clever Hans shows that humans are very good at convincing themselves something is smart when really they're subconsciously feeding it answers. So I do think validating GPT requires thinking a little adversarially rather than aiming to help it.
> The problem here is that there aren't as far as I can tell any very balanced options
Then that's exactly what GPT should have said here.
> For cases where there is in fact not a single perfect option, and where you need to weigh pro and con, it works well with the caveat that you do indeed need to be careful and check specifics. The same way I'd check specifics if I had a conversation about this with a friend.
If someone had a track record of doing rote work decently but messing up anything requiring critical thinking then I'd entrust them only with work appropriate to their skillset until they proved otherwise. That's exactly what I'm doing with GPT: I use it to do my busywork, but I'm not asking it for anything like travel advice until it gets quite a bit better.
In the example you picked apart, it got it close enough to be useful even though the answers have plenty of issues.
> The story of Clever Hans shows that humans are very good at convincing themselves something is smart when really they're subconsciously feeding it answers. So I do think validating GPT requires thinking a little adversarially rather than aiming to help it.
If the goal was to validate GPT, sure. But the goal above was not to validate GPT. The discussion was over whether it could be useful. That doesn't require "validating it". It just needs a rough understanding that is more right than wrong with respect to which types of queries are productive in producing results that saves us time without doing harm.
Yes, that means there are lots of applications where it's not suitable. That's fine.
> Then that's exactly what GPT should have said here.
I disagree. The question explicitly did not ask for that. It said it was too far to travel in one go, without explaining what the longest number of hours acceptable to travel in one day was. That a human might implicitly interpret it that way based on personal preferences might well be the case. But responding that there were no evenly split options would indicate a failure to carefully read the question. Explaining why the options did not split it evenly would be good (but GPT really would not be up to the job in this case).
Note that GPT still gets this plenty wrong, so I'm not suggesting it's up to scratch in this area.
Adding a constraint of no more than 8 hours per day, GPT3.5turbo (free ChatGPT) messes up (still suggests Florence, and gives the nonsensical suggestion of Avignon). GPT4 (paid ChatGPT only) suggests Genoa and Nice.
Lowering the threshold to 6 hours (which AFAIK is not possible), GPT3.5turbo gives the same broken set of options. GPT4 still suggests Genoa and Nice, wrongly claiming no more than 6 hours per day, so that is definitely a problem. In this case it should have said there's no way of doing that.
Trying to be more explicit about this (" We want to travel no more than 6 hours from Rome to the stopover, and no more than 6 hours from the stopover to Montpellier") does not help, so this is indeed a strong indicator that it struggles with this particular type of constraint and you shouldn't trust it on this subject other than to give ideas.
When thinking of what would make it get this right, it's not surprising: There likely aren't that many travel descriptions containing distances and travel times in its training data, and it can't read maps yet.
> If someone had a track record of doing rote work decently but messing up anything requiring critical thinking then I'd entrust them only with work appropriate to their skillset until they proved otherwise. That's exactly what I'm doing with GPT: I use it to do my busywork, but I'm not asking it for anything like travel advice until it gets quite a bit better.
That's exactly what I'm suggesting. Maybe with the extension that asking it "tell me how to do X" questions often works well, and that sometimes even questions where you know it'll mess up the details will give you enough ideas to go on. E.g. in this case, at least GPT4 gives reasonable options even though the travel times are messed up (when the plugins are opened up, hopefully this will improve significantly). That might not matter for a region you know, but for a region you don't, getting a list of cities to plug into route planners might still be worthwhile as long as they're more right than wrong.
I can't speak to the answer the person I replied to got, because they didn't post it. If they got the answer I got asking the same question the answer they got wouldn't have been useful to them.
It's only counterproductive if I'm wrong. If I'm right that it's not yet useful for this sort of thing I'd only waste my time giving it more chances.
> in this case, at least GPT4 gives reasonable options even though the travel times are messed up
The reasonability of the options fundamentally depended on specific facts in this case. Mixing up a 3 hour train ride and a 12 hour train ride ruined the answer. So the answer I got from ChatGPT was fundamentally broken.
Oh my god, have you tried to search for a recipe before? Pages and pages and pages of bloated garbage. I just want ingredients and steps goddamnit.
Something similar will happen here. Today, recipes on Chatgpt are to the point. Tomorrow, they might start adding affiliate links to the ingredients. A year from now, the experience won't be any better than what you currently have on Google.
Correct me if I've misinterpreted your comment but it appears you're suggesting that rather than use something that is a superior product in more than one factor, you're advocating for using a product that is terrible in many factors just because... The superior product might get worse... Eventually...
Blind loyalty to a corporation no matter how terrible they make my experience and abuse my trust isn't really something I'm desperate to cling on to. There is a thirst for ChatGPT for a reason. Nothing is stopping OpenAI from pulling all the same anti-consumer behaviour that Google pulls today, but then the field is wide open for someone to come along and pull a ChatGPT on them, and the cycle continues.
If anything, I'm advocating for a perspective on technology that goes beyond dunking on incumbents because the newer, sleeker thing has arrived. Look at the economics of this industry and the motives of the market entrants; the new thing is only a donut or two or away from turning into the same bloated mess as those it sought to replace.
We have a brief window where ChatGPT will be clean and useful, before it gets loaded down with ads and bloat. Might as well use it while we can.
https://www.wikihow.com/Cook-Lamb-Steaks
They've got the basics, so I'd say it's best for novice chefs who want to fix a basic meal. There aren't a lot of gourmet variations or brand-name shills there.
I've e.g. had it produce the HTML, CSS and basic web app for a site I was putting together. Custom, with my content and the layout and design I wanted. None of it was stuff I couldn't do myself from scratch without even a Google search, but it wrote out an hour or two's worth of boilerplate in a few minutes of prompting.
Until ChatGPT has the ability to index at the scale of Google it never will be.
Google:
> latex macro behaves differently math mode normal mode
ChatGPT:
> How can I make a LaTeX macro behaves differently in math mode and normal mode?
Google failed, ChatGPT succeeded instantly.
Now that's but a single data point, but I found it cute.