It seems like Kagi has been putting a lot of their eggs in the LLM basket. This makes me nervous about their future. The t-shirt decision is another orange flag in my opinion
It seems like Kagi has been putting a lot of their eggs in the LLM basket. This makes me nervous about their future. The t-shirt decision is another orange flag in my opinion
To be honest, I don’t get that impression at all. Where other companies attempt to shove LLMs down your throat whether you like it or not, Kagi has so far managed to put it out of the way. You...
It seems like Kagi has been putting a lot of their eggs in the LLM basket. This makes me nervous about their future.
To be honest, I don’t get that impression at all. Where other companies attempt to shove LLMs down your throat whether you like it or not, Kagi has so far managed to put it out of the way. You only experience integrations if you actively ask for them, but never (as far as I could tell) unprompted for. Websites remain JavaScript-less, only activating that once you go out of your way to enable LLM features, etc.
Their founder has previously stated that offering LLM features costs them a fraction of the actual web search core business, and if it makes a subset of users happy to get more out of their subscription and makes the search engine seem more “modern” at the same time, while not obstructing other users who don’t care, why wouldn’t they add this?
To be clear, with just about any other company, I’d share your concerns. But with everything I’ve seen coming out of Kagi Inc. – now a Public Benefit Corporation – so far, I remain optimistic.
The t-shirt decision is another orange flag in my opinion
May I ask why? They announced this a couple of months back when they hit the 20,000 paying subscriber milestone, and this freebie campaign is their way of introducing the store and opening up future merch revenue for content customers (and rewarding “earlier birds”). What’s the issue?
That they are obviously for profit, silly. But in all seriousness, I've never understood why people get upset when companies announce merch. Merch has to be the least anti-consumer way to earn...
What’s the issue?
That they are obviously for profit, silly.
But in all seriousness, I've never understood why people get upset when companies announce merch. Merch has to be the least anti-consumer way to earn more revenue. It's way better than locking more features behind paywalls etc (Kagi already does this, but it's mostly related to features that actually cost Kagi more to operate and they are very transparent about how much each search costs them etc).
So, I actually pay for Kagi's unlimited searches plan. I like their product. That said, I don't want to be a "fan" of a company to the point of buying merch and being a free walking billboard....
But in all seriousness, I've never understood why people get upset when companies announce merch.
So, I actually pay for Kagi's unlimited searches plan. I like their product.
That said, I don't want to be a "fan" of a company to the point of buying merch and being a free walking billboard. Frankly, I don't think it's a healthy mindset for anyone to be in.
Give me a good product for a fair price, and if not I want to walk away without feeling like I was complicit in helping something grow that I now no longer trust.
That's totally fair. A company isn't entitled to you advertising their product in any way. My point is that merch existing doesn't hurt you if you don't want it, but if you do want their merch...
That's totally fair. A company isn't entitled to you advertising their product in any way. My point is that merch existing doesn't hurt you if you don't want it, but if you do want their merch then it's a win-win (assuming the quality is good and it's reasonably priced).
In at least 90% of cases, “merch” is fairly low quality goods with branding on them that make them completely undesirable for the secondhand market most of the time even if they did last. They are...
In at least 90% of cases, “merch” is fairly low quality goods with branding on them that make them completely undesirable for the secondhand market most of the time even if they did last. They are also made in sweatshops internationally that abuse their workers in ways that would be completely illegal in the US. The people who sell them are not in the business of manufacturing them and rarely ever have any actual ways of proving that slave labor is being used to produce those goods.
And even if the shirts are being produced ethically and with good quality, the fashion industry in general is one of the planets most polluting market segments, so literally any unneeded additional production is bad for the planet.
This isn't specific to merch but modern fashion overall. Almost anywhere you go your clothes will be low quality and involve worker abuse. It's just the world we live in. You can get high quality...
This isn't specific to merch but modern fashion overall. Almost anywhere you go your clothes will be low quality and involve worker abuse. It's just the world we live in. You can get high quality and more ethical clothes but they're hard to find. I just wouldn't put that as a negative towards merch specifically.
Not OP but, the issue to me was doing a round of fundraising, and spending ~1/3 of that on the freebie campaign. As a paying customer, as much as I appreciate the free t-shirt, would rather that...
Not OP but, the issue to me was doing a round of fundraising, and spending ~1/3 of that on the freebie campaign.
As a paying customer, as much as I appreciate the free t-shirt, would rather that money getting spent productively on improving the product or selling it to more people. Obviously not as dumb as spending all the funds on a domain name, but the roi feels like it's in the same ballpark.
Wow, I was not aware of that. Last I heard they were profitable (https://blog.kagi.com/what-is-next-for-kagi), and were only doing financing rounds to boost growth, not because they needed them to...
and spending ~1/3 of that on the freebie campaign.
As a paying customer, […] would rather that money getting spent productively on improving the product or selling it to more people.
Wow, I was not aware of that.
Last I heard they were profitable (https://blog.kagi.com/what-is-next-for-kagi), and were only doing financing rounds to boost growth, not because they needed them to survive (unlike a lot of startups especially in tech…).
But as @vord’s commented already pointed out, I’m not sure what better alternatives there might be to spend it outside of directly going for more full-time hires.
I can order 20,000 custom-printed T-shirts from Alibaba for $15,600. If Kagi could only raise less than $50,000, I'd be worried about them regardless of what they spent that money on.
I can order 20,000 custom-printed T-shirts from Alibaba for $15,600. If Kagi could only raise less than $50,000, I'd be worried about them regardless of what they spent that money on.
They actually started as an AI company, and most of their announcements recently seem to be expanding their LLMs or adding them more places. I agree that it's not shoved down user's throats, and I...
They actually started as an AI company, and most of their announcements recently seem to be expanding their LLMs or adding them more places. I agree that it's not shoved down user's throats, and I hope it stays that way. As you said, it's only used by a minority of users and that should be reflected in their resource allocation. Instead, it seems like a major focus and marketing ploy.
As others have covered, the shirts were directly subsidized by a new injection of investor money. I generally support businesses diversifying their revenue sources through avenues like merch (if ethically sourced, which is often difficult), but this seems like a poor use of these funds specifically. At the end of the day it's more important for Kagi to be a useful tool instead of a strong brand.
These are just my opinions. I've read other blogs & posts online with similar sentiments over the last year, but unfortunately I didn't bookmark them.
As someone who pays for Perplexity (search engine with AI) and uses Google at the same time, here's my two cents about search engines with AI VS traditional search engines. Usually my searches fit...
As someone who pays for Perplexity (search engine with AI) and uses Google at the same time, here's my two cents about search engines with AI VS traditional search engines.
Usually my searches fit into these four criteria, it's been like this for years now:
find the website that I want to access (basically I'm lazy to type in the url)
find solutions to coding problems (python, SQL, etc)
find opinions about products, games, etc (usually I check links from reddit)
research miscellaneous things that I read online (fact check what some guy said, get more info about an article, etc)
I don't have a habit of doing any complex queries, at best I use "site:reddit.com" for example, but nothing more.
Eventually I tried out perplexity, and it changed how I do searches. It has both good and bad sides, which is why I still rely on Google.
Starting out with the bad:
- their response quality are highly dependent on the quality of the search results.
I know, "no shit sherlock", but let me elaborate. AI, no matter what model or what company, has trouble understanding things like jokes, sarcasm and lies. Some deal with it better than others, but it's a common problem.
I should mention, Google's Gemini is particularly terrible at this, Perplexity (which uses Sonar, Claude and GPT) in my experience doesn't have as many problems as Google's Gemini have.
So, if among those search results pops up a reddit thread, and there's a comment that says some joke, the AI may take it at face value and insert it in your response. Depending on the rest of the response, you may be inclined to believe it (if the rest looks ok, why would you mistrust one random part in the response?).
It's why, depending on what I'm looking for, I filter by results that I know that will be relevant. If I'm searching for opinions, I filter by reddit. If I'm searching for more factual results, I filter by academic links. For news, news websites, etc.
- they tend to treat the search results as the gospel of truth
This goes hand in hand with the previous point. The AI is instructed to read the sources and pass a summary of it to the user. And the AI will do just that.
But also, it depends on the model and topic. Claude Sonnet 3.5, the model that I mainly use in Perplexity, will every once in a while clarify nuances about what it said, without referring to any source, especially in more controversial topics. A few days ago I searched about if Pitbulls actually eat their litter mates (obviously not, but I was fact checking a photo), I had Sonnet clarify several times that there is a syndrome that could explain it, but that it's not unique to pitbulls.
But, in more niche topics, or that aren't controversial, the AI will generally trust what it read. Unfortunately I cannot remember what it was, but several weeks ago I searched about a topic, and the AI read a random comment on reddit and passed it on as if it was true. I didn't believe it and asked it fact check that point, and sure enough, it couldn't, there were no sources that could confirm what that redditor said. (I do remember the AI saying something along the lines "it could still be true", as if it was a prideful kid unwilling to admit they may have said something wrong lol)
The good:
- it's great at introducing you to a topic, and summarising it
Perplexity does not substitute proper deep dives, I still will read the articles if I need more details, but it's great at helping you get started. If you know nothing about a topic, it will give you a general idea of what it is. I found myself getting to know better several topics, which previously I wouldn't have bothered because I would have to deep dive a bit into some pages.
Some time ago, I found out that the Japanese in WW2 had plans to invade the USSR. Normally, finding out about this would involve searching google, learn about the plan's names, search about the plan, click the links, read about it. With perplexity it was like "Did the japanese plan to invade USSR?", and it answered yes, elaborated on it and provided the sources. Just like that I had everything I needed in one spot after one interaction.
- it's useful for comparisons, especially in type of products that I'm not familiar with
At least in the case of perplexity, there's a feature called "Pro Search", which basically turns your prompt into several different search queries and analyses all of them.
So for example, you want to compare product A with product B, the pro feature will search first about product A, then about product B, and then a final search about people comparing product A and product B, and gives you a final response with all the sources combined.
And when it's about products I'm not familiar with, I can ask about little details to understand why they matter and what's the difference in practice. E.g. why it matters if headphones are open or closed.
- because it reads and quotes sources, there's little to no hallucinations
Hallucinations is when an AI makes stuff up. They tend to happen after long interactions with them, and/or when you start going deep into niche subjects.
But in this case, it doesn't happen. Or at least, in my experience it never happened.
Because the AI will be always reading the sources and essentially repeating what they said, it never gets the chance to make stuff up. Obviously the new problem is that the search results have to be good, but assuming that you do a decent job at filtering by sources that are relevant, you will get decent responses.
- it's good at giving you references in documentations
I work with looker and lookml. Without going too much into it, it's a dashboard platform, and you need to code to make dashboards. For one reason or another, there isn't a lot of resource material readily available on the internet about lookml. In python you are almost guaranteed to always find someone with a similar problem to yours, in lookml it's rare.
Because of this, there's the expected consequence: none of the models I tried know a lot about lookml. Back in early 2023, I remember the first iterations of GPT 4 knowing absolutely nothing about lookml. Edit: hallucinations were constant. LLM's don't know that they don't know, so chatGPT would always make up functions that didn't exist
Thus, there's only the documentation. Thankfully, perplexity saves me a lot of time. Whenever I have a doubt or don't know how to do something, perplexity has been very useful at referencing the docs and telling me where to look at.
Where I am going with this
Idk if I'm doing a good job or not - probably not - but what I'm trying to say is that at the end of the day, AI in search engines are still imperfect and they don't substitute the traditional search engines. But they do introduce a new dynamic in searching.
I consider them a different tool, not a replacement, at the moment at least.
Whenever I want to find out something really quick, I use Google. For example, stock value, weather, translation of a word, meaning of a word, etc. Google will in all these cases show the information right at the top.
For a bit bigger/complex topics, I use perplexity. What's the function in lookml that does this, give me more information about this function in python pandas, what's a library that let's me do X, etc.
This new dynamic has been useful for me, both in my day to day and work, enough that I pay 20 euros a month. If Kagi offers something like this, a traditional search engine and an AI on the side, it's honestly the best of both worlds (that said, for transparency, I never used Kagi, I will research more after posting this comment)
Kagi's main LLM feature I use is Quick Answer which works much like Perplexity and summarises the top few search results. I've used Perplexity's free version as well, and I find it's usually...
If Kagi offers something like this, a traditional search engine and an AI on the side, it's honestly the best of both worlds
Kagi's main LLM feature I use is Quick Answer which works much like Perplexity and summarises the top few search results. I've used Perplexity's free version as well, and I find it's usually comparable to Kagi's quick answer. The only exception is Perplexity is much better at digging up scientific studies for complicated scientific questions.
Kagi has blown Google out of the water for me. I was skeptical, but the ability to rank websites and set up custom bangs really lets you dial in exactly what you want out of a search engine.
you can get custom bangs completely for free in your browser by making bookmarklets with keywords in firefox for example, make a bookmark for https://en.wikipedia.org/wiki/Special:Search?search=%s...
you can get custom bangs completely for free in your browser by making bookmarklets with keywords in firefox
for example, make a bookmark for https://en.wikipedia.org/wiki/Special:Search?search=%s and give it they keyword w
then type ctrl+L (focus url bar) then w bananas + enter and u are at the wikipedia page (or search results) for "bananas"
bangs are just a worse version of this built-in functionality
TIL. I guess the main advantage of website bangs in Kagi is that it's easy to add bangs that have already been configured by the community. I also use Kagi bangs to go straight to the image...
TIL. I guess the main advantage of website bangs in Kagi is that it's easy to add bangs that have already been configured by the community.
I also use Kagi bangs to go straight to the image results (!i ripe heritage tomatoes) or get international results (!es, !jp, etc).
you can do the same thing with bookmarks, for example this is my google images search (keyword i): https://www.google.com/images?oe=UTF-8&gfns=1&q=%s for an arbitrary country code, bookmark the...
you can do the same thing with bookmarks, for example this is my google images search (keyword i): https://www.google.com/images?oe=UTF-8&gfns=1&q=%s
for an arbitrary country code, bookmark the following url with the keyword kint (or your choice). usage is like: kint es como adoptar un gato en espana and the first word after kint will be used as the country code, the rest is the search term, so itll take you to https://kagi.com/search?r=es&q=como+adoptar+un+gato+en+espana (note i swapped the region & search term from kagi's default but query params like this are order agnostic)
more info in a blog post I wrote a few years ago, I copied from my bookmarklet for searching on an arbitrary wiki.gg wiki (sadly RIP gamepedia) and just changed the url to be kagi (reordering the params means I didnt have to edit any code)
Letting you edit questions in the middle of a chat is something a lot of AI chats do. It gets more interesting if you can also edit the AI’s answers. You could correct its mistakes, or add some...
Letting you edit questions in the middle of a chat is something a lot of AI chats do. It gets more interesting if you can also edit the AI’s answers. You could correct its mistakes, or add some new ones :)
I've been paying for a chatgpt subscription for about a year now, I'm curious to hear what people think about switching that over to kagi now (I'd rather not pay for both)
I've been paying for a chatgpt subscription for about a year now, I'm curious to hear what people think about switching that over to kagi now (I'd rather not pay for both)
The Kagi assistant is currently not very polished, and it also doesn't yet support image recognition. I'm only paying for Kagi Ultimate and it works fine for using Claude 3.5 and GPT4o, thought a...
The Kagi assistant is currently not very polished, and it also doesn't yet support image recognition. I'm only paying for Kagi Ultimate and it works fine for using Claude 3.5 and GPT4o, thought a bit buggy.
It seems like Kagi has been putting a lot of their eggs in the LLM basket. This makes me nervous about their future. The t-shirt decision is another orange flag in my opinion
To be honest, I don’t get that impression at all. Where other companies attempt to shove LLMs down your throat whether you like it or not, Kagi has so far managed to put it out of the way. You only experience integrations if you actively ask for them, but never (as far as I could tell) unprompted for. Websites remain JavaScript-less, only activating that once you go out of your way to enable LLM features, etc.
Their founder has previously stated that offering LLM features costs them a fraction of the actual web search core business, and if it makes a subset of users happy to get more out of their subscription and makes the search engine seem more “modern” at the same time, while not obstructing other users who don’t care, why wouldn’t they add this?
To be clear, with just about any other company, I’d share your concerns. But with everything I’ve seen coming out of Kagi Inc. – now a Public Benefit Corporation – so far, I remain optimistic.
May I ask why? They announced this a couple of months back when they hit the 20,000 paying subscriber milestone, and this freebie campaign is their way of introducing the store and opening up future merch revenue for content customers (and rewarding “earlier birds”). What’s the issue?
That they are obviously for profit, silly.
But in all seriousness, I've never understood why people get upset when companies announce merch. Merch has to be the least anti-consumer way to earn more revenue. It's way better than locking more features behind paywalls etc (Kagi already does this, but it's mostly related to features that actually cost Kagi more to operate and they are very transparent about how much each search costs them etc).
So, I actually pay for Kagi's unlimited searches plan. I like their product.
That said, I don't want to be a "fan" of a company to the point of buying merch and being a free walking billboard. Frankly, I don't think it's a healthy mindset for anyone to be in.
Give me a good product for a fair price, and if not I want to walk away without feeling like I was complicit in helping something grow that I now no longer trust.
That's totally fair. A company isn't entitled to you advertising their product in any way. My point is that merch existing doesn't hurt you if you don't want it, but if you do want their merch then it's a win-win (assuming the quality is good and it's reasonably priced).
In at least 90% of cases, “merch” is fairly low quality goods with branding on them that make them completely undesirable for the secondhand market most of the time even if they did last. They are also made in sweatshops internationally that abuse their workers in ways that would be completely illegal in the US. The people who sell them are not in the business of manufacturing them and rarely ever have any actual ways of proving that slave labor is being used to produce those goods.
And even if the shirts are being produced ethically and with good quality, the fashion industry in general is one of the planets most polluting market segments, so literally any unneeded additional production is bad for the planet.
This isn't specific to merch but modern fashion overall. Almost anywhere you go your clothes will be low quality and involve worker abuse. It's just the world we live in. You can get high quality and more ethical clothes but they're hard to find. I just wouldn't put that as a negative towards merch specifically.
Promotional clothing makes up a sizeable portion of the fashion industry, though, and there is more merch than just clothing articles.
Not OP but, the issue to me was doing a round of fundraising, and spending ~1/3 of that on the freebie campaign.
As a paying customer, as much as I appreciate the free t-shirt, would rather that money getting spent productively on improving the product or selling it to more people. Obviously not as dumb as spending all the funds on a domain name, but the roi feels like it's in the same ballpark.
That's 20,000 billboards right there, ones which will organically rave if someone asks 'whats that on your shirt.' Thats advertising money well spent.
Especially given they have a marketing budget of zero, i.e. nothing at all, otherwise, as far as I’m aware.
Wow, I was not aware of that.
Last I heard they were profitable (https://blog.kagi.com/what-is-next-for-kagi), and were only doing financing rounds to boost growth, not because they needed them to survive (unlike a lot of startups especially in tech…).
But as @vord’s commented already pointed out, I’m not sure what better alternatives there might be to spend it outside of directly going for more full-time hires.
I can order 20,000 custom-printed T-shirts from Alibaba for $15,600. If Kagi could only raise less than $50,000, I'd be worried about them regardless of what they spent that money on.
From their 670k round, they spent 1/3 of that on t-shirts, so ~220k on t-shirts? https://blog.kagi.com/celebrating-20k
That seems like a lot. Did they pay retail price for the shirts?
They actually started as an AI company, and most of their announcements recently seem to be expanding their LLMs or adding them more places. I agree that it's not shoved down user's throats, and I hope it stays that way. As you said, it's only used by a minority of users and that should be reflected in their resource allocation. Instead, it seems like a major focus and marketing ploy.
As others have covered, the shirts were directly subsidized by a new injection of investor money. I generally support businesses diversifying their revenue sources through avenues like merch (if ethically sourced, which is often difficult), but this seems like a poor use of these funds specifically. At the end of the day it's more important for Kagi to be a useful tool instead of a strong brand.
These are just my opinions. I've read other blogs & posts online with similar sentiments over the last year, but unfortunately I didn't bookmark them.
As someone who pays for Perplexity (search engine with AI) and uses Google at the same time, here's my two cents about search engines with AI VS traditional search engines.
Usually my searches fit into these four criteria, it's been like this for years now:
I don't have a habit of doing any complex queries, at best I use "site:reddit.com" for example, but nothing more.
Eventually I tried out perplexity, and it changed how I do searches. It has both good and bad sides, which is why I still rely on Google.
Starting out with the bad:
- their response quality are highly dependent on the quality of the search results.
I know, "no shit sherlock", but let me elaborate. AI, no matter what model or what company, has trouble understanding things like jokes, sarcasm and lies. Some deal with it better than others, but it's a common problem.
I should mention, Google's Gemini is particularly terrible at this, Perplexity (which uses Sonar, Claude and GPT) in my experience doesn't have as many problems as Google's Gemini have.
So, if among those search results pops up a reddit thread, and there's a comment that says some joke, the AI may take it at face value and insert it in your response. Depending on the rest of the response, you may be inclined to believe it (if the rest looks ok, why would you mistrust one random part in the response?).
It's why, depending on what I'm looking for, I filter by results that I know that will be relevant. If I'm searching for opinions, I filter by reddit. If I'm searching for more factual results, I filter by academic links. For news, news websites, etc.
- they tend to treat the search results as the gospel of truth
This goes hand in hand with the previous point. The AI is instructed to read the sources and pass a summary of it to the user. And the AI will do just that.
But also, it depends on the model and topic. Claude Sonnet 3.5, the model that I mainly use in Perplexity, will every once in a while clarify nuances about what it said, without referring to any source, especially in more controversial topics. A few days ago I searched about if Pitbulls actually eat their litter mates (obviously not, but I was fact checking a photo), I had Sonnet clarify several times that there is a syndrome that could explain it, but that it's not unique to pitbulls.
But, in more niche topics, or that aren't controversial, the AI will generally trust what it read. Unfortunately I cannot remember what it was, but several weeks ago I searched about a topic, and the AI read a random comment on reddit and passed it on as if it was true. I didn't believe it and asked it fact check that point, and sure enough, it couldn't, there were no sources that could confirm what that redditor said. (I do remember the AI saying something along the lines "it could still be true", as if it was a prideful kid unwilling to admit they may have said something wrong lol)
The good:
- it's great at introducing you to a topic, and summarising it
Perplexity does not substitute proper deep dives, I still will read the articles if I need more details, but it's great at helping you get started. If you know nothing about a topic, it will give you a general idea of what it is. I found myself getting to know better several topics, which previously I wouldn't have bothered because I would have to deep dive a bit into some pages.
Some time ago, I found out that the Japanese in WW2 had plans to invade the USSR. Normally, finding out about this would involve searching google, learn about the plan's names, search about the plan, click the links, read about it. With perplexity it was like "Did the japanese plan to invade USSR?", and it answered yes, elaborated on it and provided the sources. Just like that I had everything I needed in one spot after one interaction.
- it's useful for comparisons, especially in type of products that I'm not familiar with
At least in the case of perplexity, there's a feature called "Pro Search", which basically turns your prompt into several different search queries and analyses all of them.
So for example, you want to compare product A with product B, the pro feature will search first about product A, then about product B, and then a final search about people comparing product A and product B, and gives you a final response with all the sources combined.
And when it's about products I'm not familiar with, I can ask about little details to understand why they matter and what's the difference in practice. E.g. why it matters if headphones are open or closed.
- because it reads and quotes sources, there's little to no hallucinations
Hallucinations is when an AI makes stuff up. They tend to happen after long interactions with them, and/or when you start going deep into niche subjects.
But in this case, it doesn't happen. Or at least, in my experience it never happened.
Because the AI will be always reading the sources and essentially repeating what they said, it never gets the chance to make stuff up. Obviously the new problem is that the search results have to be good, but assuming that you do a decent job at filtering by sources that are relevant, you will get decent responses.
- it's good at giving you references in documentations
I work with looker and lookml. Without going too much into it, it's a dashboard platform, and you need to code to make dashboards. For one reason or another, there isn't a lot of resource material readily available on the internet about lookml. In python you are almost guaranteed to always find someone with a similar problem to yours, in lookml it's rare.
Because of this, there's the expected consequence: none of the models I tried know a lot about lookml. Back in early 2023, I remember the first iterations of GPT 4 knowing absolutely nothing about lookml. Edit: hallucinations were constant. LLM's don't know that they don't know, so chatGPT would always make up functions that didn't exist
Thus, there's only the documentation. Thankfully, perplexity saves me a lot of time. Whenever I have a doubt or don't know how to do something, perplexity has been very useful at referencing the docs and telling me where to look at.
Where I am going with this
Idk if I'm doing a good job or not - probably not - but what I'm trying to say is that at the end of the day, AI in search engines are still imperfect and they don't substitute the traditional search engines. But they do introduce a new dynamic in searching.
I consider them a different tool, not a replacement, at the moment at least.
Whenever I want to find out something really quick, I use Google. For example, stock value, weather, translation of a word, meaning of a word, etc. Google will in all these cases show the information right at the top.
For a bit bigger/complex topics, I use perplexity. What's the function in lookml that does this, give me more information about this function in python pandas, what's a library that let's me do X, etc.
This new dynamic has been useful for me, both in my day to day and work, enough that I pay 20 euros a month. If Kagi offers something like this, a traditional search engine and an AI on the side, it's honestly the best of both worlds (that said, for transparency, I never used Kagi, I will research more after posting this comment)
Kagi's main LLM feature I use is Quick Answer which works much like Perplexity and summarises the top few search results. I've used Perplexity's free version as well, and I find it's usually comparable to Kagi's quick answer. The only exception is Perplexity is much better at digging up scientific studies for complicated scientific questions.
Kagi has blown Google out of the water for me. I was skeptical, but the ability to rank websites and set up custom bangs really lets you dial in exactly what you want out of a search engine.
you can get custom bangs completely for free in your browser by making bookmarklets with keywords in firefox
for example, make a bookmark for
https://en.wikipedia.org/wiki/Special:Search?search=%sand give it they keywordwthen type ctrl+L (focus url bar) then
w bananas+ enter and u are at the wikipedia page (or search results) for "bananas"bangs are just a worse version of this built-in functionality
TIL. I guess the main advantage of website bangs in Kagi is that it's easy to add bangs that have already been configured by the community.
I also use Kagi bangs to go straight to the image results (
!i ripe heritage tomatoes) or get international results (!es,!jp, etc).you can do the same thing with bookmarks, for example this is my google images search (keyword
i):https://www.google.com/images?oe=UTF-8&gfns=1&q=%sfor an arbitrary country code, bookmark the following url with the keyword
kint(or your choice). usage is like:kint es como adoptar un gato en espanaand the first word afterkintwill be used as the country code, the rest is the search term, so itll take you to https://kagi.com/search?r=es&q=como+adoptar+un+gato+en+espana (note i swapped the region & search term from kagi's default but query params like this are order agnostic)more info in a blog post I wrote a few years ago, I copied from my bookmarklet for searching on an arbitrary wiki.gg wiki (sadly RIP gamepedia) and just changed the url to be kagi (reordering the params means I didnt have to edit any code)
Letting you edit questions in the middle of a chat is something a lot of AI chats do. It gets more interesting if you can also edit the AI’s answers. You could correct its mistakes, or add some new ones :)
I've been paying for a chatgpt subscription for about a year now, I'm curious to hear what people think about switching that over to kagi now (I'd rather not pay for both)
The Kagi assistant is currently not very polished, and it also doesn't yet support image recognition. I'm only paying for Kagi Ultimate and it works fine for using Claude 3.5 and GPT4o, thought a bit buggy.