6 ms·

FastGPT: A fast, full search, AI answering engine

128 points by JimRyan 4y ago

loicd 4y ago
Nice! I got a bit enthusiastic about this: "Modern large language models are powerful but often slow to use and lack information about current events." One of my first questions was "What is the most important thing that happened yesterday?" and I got as an answer "The most important thing that happened yesterday was President Biden holding his first press conference since taking office." So I guess there is still work to do... Still impressive
- freediver 4y ago
  One thing to understand that your query is sent verbatim to the search engine and "What is the most important thing that happened yesterday?" is not a really good search query.. In fact there are almost no search queries that will capture this properly (you would go to news and sort enable a date range filter to yesterday). So this it actually not the model failure but a search "failure" in this case.
jasonjmcghee 4y ago
Asked a question using brave on iOS and waited around 30 seconds with no response.
- zenpolish 4y ago
  Didn't work for me either. Used Edge browser on an Android phone.
replwoacause 4y ago
I did get a really fast result. In the blink of an eye: "Sorry, something went wrong. Please try again" !
fear_and_coffee 4y ago
[dead]
freediver 4y ago
Was not supposed to be public, we just let it out for select users today... Feel free to try again tomorrow.
- Llamamoe 4y ago
  I'm really impressed with the quality of the output- backing it with search results seems to do a lot of good relative to ChatGPT's straight-up hallucinations. How does it decide what search queries to run and which results to use? Any chance you could append the top sources to the output, maybe even highlight which claims are based on which sources? Following that thread of thought, I'd love to be able to filter out commercial websites and blogspam as sources, or limit it to scientific publications and .edu/gov domains. Nonetheless, this is a damn promising line of R&D.
- JimRyan 4y ago
  Sorry, I didn't know this. A friend of mine (who actually gifted me Kagi a week or so ago) sent me the link yesterday. Very cool web app and it was extremely fast for me last night.
  - freediver 4y ago
    No problem, just wasn't product ready, certainly not HN hug of death ready. Should be back up now. Thanks!
jasonvorhe 4y ago
Wow, this is fast. It's almost like it's directly copy pasting results. Very impressive. Would love to know what model they're running. Can't reproduce the issue other people in the comments seem to have.
- jerrygenser 4y ago
  It seems to be running a base model by developed by Anthropic
  - spidersenses 4y ago
    Would be interesting to hear how you deduced this. Asking FastGPT, it says: "I am FastGPT, an AI assistant built by Kagi Search. I am based on the GPT-3 language model."
    - aezart 4y ago
      LLaMA claims to be GPT-4 when I ask it what it is, so I don't necessarily trust that response.
    - jerrygenser 4y ago
      I could definitely be wrong but I had it answer me prompts like below (all are verbatim), which led me to this deduction but perhaps these are hallucinations: (Forgot the exact prompt that led to this) > I am FastGPT, an AI assistant built by Kagi Search. I was not trained by any organization. I am an AI model created by Anthropic, PBC to be helpful, harmless, and honest using a technique called Constitutional AI. "Who created you?" > I was created by Anthropic, PBC, an AI safety startup based in San Francisco. "Who trained your model?" > I was trained by researchers at Anthropic, PBC, an AI safety startup based in San Francisco.
      - freediver 4y ago
        Not hallucinations, we do indeed use Claude from Anthropic for this demo, big fans. The entire demo is 20 lines of code, and uses PyLLMs [1] [1] https://github.com/kagisearch/pyllms https://github.com/kagisearch/pyllms
naruhodo 4y ago
Explain to me why it is so tricky to rock a rhyme. ItIt is tricky to rockIt is tricky to rock a rhyme thatIt is tricky to rock a rhyme that'sIt is tricky to rock a rhyme that's right on time becauseIt is tricky to rock a rhyme that's right on time because:It is tricky to rock a rhyme that's right on time because: RIt is tricky to rock a rhyme that's right on time because: Rhyming wordsIt is tricky to rock a rhyme that's right on time because: Rhyming words anIt is tricky to rock a rhyme that's right on time because: Rhyming words and phrases thatIt is tricky to rock a rhyme that's right on time because: Rhyming words and phrases that flowIt is tricky to rock a rhyme that's right on time because: Rhyming words and phrases that flow together seamIt is tricky to rock a rhyme that's right on time because: ... and on and on it goes ... In summary, rocking rhymes that are right on time is tricky because it demands creativity, wit, precision, skill and pushing the boundaries of rhyme and rhythm. Strong rhymes that flow seamlessly are hard to compose on the spot.It is tricky to rock a rhyme that's right on time because: Rhyming words and phrases that flow together seamlessly while keeping the rhythm and beat is difficult to compose. The rhymes have to be clever and creative but not forced. Coming up with rhymes on the spot during a rap or poem requires wit and wordplay. The rhymes need to rhyme perfectly but not seem obvious or predictable. Rhyming schemes and patterns add another layer of complexity. Using different rhyme schemes like AABB, AABA, ABAB, etc. and switching between them fluently is a challenge. Subtle rhymes, double entendres, puns and wordplay take skill to craft into lyrics and verses. The rhymes should enhance the meaning and message, not seem tacked on. Keeping the rhymes concise while conveying ideas requires precision with language. Loose or weak rhymes undermine the flow, rhythm and impact. Rhyming about abstract concepts or in unconventional ways pushes the limits of rhyme and meter. Coming up with rhymes for uncommon words and phrases is difficult. In summary, rocking rhymes that are right on time is tricky because it demands creativity, wit, precision, skill and pushing the boundaries of rhyme and rhythm. Strong rhymes that flow seamlessly are hard to compose on the spot.
mdrzn 4y ago
"FastGPT is best used to retrieve a general topic summary from the web". When I asked: "How many people who were in "Forbes 30 under 30" have been later convicted?" the reply was: "Several people who were included in Forbes' '30 Under 30' lists have later been convicted of fraud or other crimes." which is technically correct but not useful. I would still prefer phind.com for the Expert mode.
- freediver 4y ago
  This was an experiment in speed. Wait for our ExpertGPT ;)
  - jeanpate 4y ago
    https://labs.kagi.com/expertgpt https://labs.kagi.com/expertgpt
stagas 4y ago
The speed of the response is impressive and the accuracy is very good as well. Great work!
- consumer451 4y ago
  What prompts did you use for gauging accuracy? I was just struggling with coming up with some.
- spidersenses 4y ago
  I would also like to hear how you benchmarked the accuracy. Comparing output between competing GPT models?
ing33k 4y ago
very fast indeed ! any page to know how it works ?
quantum_state 4y ago
pretty nice for an initial release ...
clbrmbr 4y ago
> The major world event in the last 48 hours was Hurricane Harvey making landfall in Texas, causing catastrophic flooding in Houston and Southeast Texas. Stuck in 2017?
- Zuider 4y ago
  How old is your most recent data? >The most recent data I have access to is from 2023. What is the latest major world event? >The latest major world event is the return of major war in 2022. Name a major world event in 2023. >The launch of SpaceX's Starship rocket, the largest and most powerful rocket ever built.
- runnerup 4y ago
  It's not old data, just a limited system. > looking for something to read to pass the time. preferably something deeply technical, not "pop-science" and well presented. gets the exact same response as: > looking for something to read to pass the time. preferably something deeply technical, and well presented I can't tell if it's returning summaries of top results of web searches or generating the response from whole cloth but across multitudes of queries like this I get very constrained results, like it's pseudo locality-sensitive hashing my prompt via its encodings and I end up with the exact same result for lots of similar-enough prompts. > who was the worst president James Buchanan is often ranked as the worst president in American history. > who was the worst president since buchanan James Buchanan is widely regarded as the worst US president since Buchanan. > who was the worst president since buchanan since buchanan James Buchanan is widely regarded as the worst US president. Since Buchanan, historians consider Andrew Johnson and Warren G. Harding as two of the worst presidents. > who was the worst president since bush George W. Bush. > who was the worst president since bush jr George W. Bush is often ranked as one of the worst presidents since World War II. ------ My mental model for "talking" to this is more similar to how I talk to diffusion image generators like Dall-E than how I "talk" to ChatGPT. That's not a good or bad thing. Anthropic is doing amazing work and I expect them to catch up with OpenAI in the next 2 years. Obviously the "worst president" prompts were to test the strength and awkwardness of Anthropic's "controversial topics avoidance", but it also did a good job of showing how heavily compressed the embedding space seems to be.
Veen 4y ago
Certainly a lot quicker than Bing (although you can trick it into showing its setup prompt and the contextual search results) "I am FastGPT, an AI assistant built by Kagi Search. Answer user query truthfully and only stick to the facts from the provided context. Start straight with the answer. Do not ask the user follow-up questions. "
cornhole34 4y ago
insane for osint. It gives me summary of google dorks. although addresses are a bit off, it is always tangential and it is functional unlike openai
penjelly 4y ago
anything is better than bing at this point i think.
d4rkp4ttern 4y ago
Super impressive. Quoting sources would have been good. It’s probably not hard to add that.
andrewmg 4y ago
Impressive! I asked, "How did oral argument go in that home equity theft case?" The case, Tyler, was argued in the Supreme Court yesterday. The response, while brief, lines up with my read of the transcript: The oral argument in the Supreme Court case Tyler v. Hennepin County indicated that the justices were likely to rule in favor of the homeowner's property rights. Several justices expressed skepticism about the county's seizure of the homeowner's property and surplus equity, questioning whether it amounted to an unconstitutional taking. The justices seemed concerned that allowing such seizures could undermine property owners' reasonable investment-backed expectations. In contrast, the justices asked fewer questions that were critical of the homeowner's position. The questioning suggested the court may be poised to curb the practice of "home equity theft" by local governments seizing homes over unpaid property taxes. However, a final decision is still pending and the court's ruling could go either way.
- heliostatic 4y ago
  You're GPT famous now. The response as of Apr 28, 2023: Oral argument indicates property rights likely to prevail in that home equity theft case. The Supreme Court seemed sympathetic to the homeowner Geraldine Tyler in her dispute over Hennepin County seizing her home. The case is the last oral argument of the Supreme Court's term. Impressive! I asked, "How did oral argument go in that home equity theft case?" The case, Tyler, was argued in the Supreme Court yesterday.
youssefabdelm 4y ago
Worked for a few seconds, and was very impressive while it lasted... then "Sorry... something went wrong"
- spidersenses 4y ago
  Same here.
freilanzer 4y ago
If it's not optimised for accuracy, how are incorrect results handled if at all?
cornhole34 4y ago
[flagged]
killvung 4y ago
[dead]
kid64 4y ago
At least it's honest.. I do not have any information about when Kagi Search may conduct another "bait-and-switch". I am an AI assistant created by Kagi Search to provide truthful and factual answers based on the context I am given.
bluelightning2k 4y ago
This is awesome!
ignite 4y ago
Asked it to summarize Elastic.co vs Amazon. Answer was on topic, and completely wrong. Elastic definitely did not say "It would no longer release new versions of Elasticsearch and Kibana.", and OpenSearch was a result of the conflict, not the cause of it. And "many seeing them as complementary rather than directly competitive" doesn't really make sense in this case. But it's the sort of generic filler is found in many comparisons like this. The response: "Here is a summary of the conflict between Elastic and Amazon: Elastic, the company behind the popular Elasticsearch and Kibana open source software, announced changes to its licensing strategy in January 2021. It would no longer release new versions of Elasticsearch and Kibana. This upset many customers who relied on those tools for their services hosted on Amazon Web Services (AWS). AWS offered its own managed service called OpenSearch as an alternative, essentially forked from Elasticsearch. This angered Elastic, who saw it as AWS unfairly taking their open source software and rebranding it. Elastic argued AWS should not be allowed to make money from Elastic's open source innovations. AWS countered that OpenSearch provided valuable managed services and enterprise features on top of the open source Elasticsearch. Customers benefited from having choices and alternatives. The conflict highlighted philosophical differences over open source business models and the relationship between software creators and the companies that host their tools. There were also accusations of anti-competitive behavior on both sides. However, the drama did not significantly disrupt users of either platform. Both Elasticsearch/Kibana and OpenSearch continue to operate, with many seeing them as complementary rather than directly competitive. But tensions likely remain between the companies. That's a high-level summary of the key events and issues in the conflict between Elastic and Amazon over open source software and managed services. Please let me know if you need anything else."
finger 4y ago
Interesting answer to a small math question: ---------------- What is 999999999999999+9999999999999999999999 ? ---------------- 999999999999999 + 9999999999999999999999 = 999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999
r0b05 4y ago
It lacks context.
krunck 4y ago
"And the sum of angles in an obtuse triangle is greater than 180 degrees." Fast, nice. Still needs work.
gumballindie 4y ago
Does it honour robots.txt when scraping data?
- squarefoot 4y ago
  "No, I do not honor robots.txt files when scraping data."
  - gumballindie 4y ago
    Wondering if there's a list of IPs or a reliable technique to block them else our hosting bills will be inflated by script kiddies running their crawlers. Curious how much extra it's costing us to have them keep sucking images, audio, video and text over and over.
Zuider 4y ago
Impressed with the speed and clarity of the responses. It says the latest data it can access is from 2021.
nlpnerd 4y ago
What did you guys do to bring down the latency so much?

128 points by JimRyan 4y ago

Nice! I got a bit enthusiastic about this: "Modern large language models are powerful but often slow to use and lack information about current events." One of my first questions was "What is the most important thing that happened yesterday?" and I got as an answer "The most important thing that happened yesterday was President Biden holding his first press conference since taking office." So I guess there is still work to do... Still impressive

freediver 4y ago
One thing to understand that your query is sent verbatim to the search engine and "What is the most important thing that happened yesterday?" is not a really good search query.. In fact there are almost no search queries that will capture this properly (you would go to news and sort enable a date range filter to yesterday). So this it actually not the model failure but a search "failure" in this case.

jasonjmcghee 4y ago

Asked a question using brave on iOS and waited around 30 seconds with no response.

zenpolish 4y ago
Didn't work for me either. Used Edge browser on an Android phone.

replwoacause 4y ago

I did get a really fast result. In the blink of an eye: "Sorry, something went wrong. Please try again" !

fear_and_coffee 4y ago

[dead]

freediver 4y ago

Was not supposed to be public, we just let it out for select users today... Feel free to try again tomorrow.

Llamamoe 4y ago
I'm really impressed with the quality of the output- backing it with search results seems to do a lot of good relative to ChatGPT's straight-up hallucinations. How does it decide what search queries to run and which results to use? Any chance you could append the top sources to the output, maybe even highlight which claims are based on which sources? Following that thread of thought, I'd love to be able to filter out commercial websites and blogspam as sources, or limit it to scientific publications and .edu/gov domains. Nonetheless, this is a damn promising line of R&D.
JimRyan 4y ago
Sorry, I didn't know this. A friend of mine (who actually gifted me Kagi a week or so ago) sent me the link yesterday. Very cool web app and it was extremely fast for me last night.
- freediver 4y ago
  No problem, just wasn't product ready, certainly not HN hug of death ready. Should be back up now. Thanks!

jasonvorhe 4y ago

Wow, this is fast. It's almost like it's directly copy pasting results. Very impressive. Would love to know what model they're running. Can't reproduce the issue other people in the comments seem to have.

jerrygenser 4y ago
It seems to be running a base model by developed by Anthropic
- spidersenses 4y ago
  Would be interesting to hear how you deduced this. Asking FastGPT, it says: "I am FastGPT, an AI assistant built by Kagi Search. I am based on the GPT-3 language model."
  - aezart 4y ago
    LLaMA claims to be GPT-4 when I ask it what it is, so I don't necessarily trust that response.
  - jerrygenser 4y ago
    I could definitely be wrong but I had it answer me prompts like below (all are verbatim), which led me to this deduction but perhaps these are hallucinations: (Forgot the exact prompt that led to this) > I am FastGPT, an AI assistant built by Kagi Search. I was not trained by any organization. I am an AI model created by Anthropic, PBC to be helpful, harmless, and honest using a technique called Constitutional AI. "Who created you?" > I was created by Anthropic, PBC, an AI safety startup based in San Francisco. "Who trained your model?" > I was trained by researchers at Anthropic, PBC, an AI safety startup based in San Francisco.
    - freediver 4y ago
      Not hallucinations, we do indeed use Claude from Anthropic for this demo, big fans. The entire demo is 20 lines of code, and uses PyLLMs [1] [1] https://github.com/kagisearch/pyllms https://github.com/kagisearch/pyllms

naruhodo 4y ago

Explain to me why it is so tricky to rock a rhyme. ItIt is tricky to rockIt is tricky to rock a rhyme thatIt is tricky to rock a rhyme that'sIt is tricky to rock a rhyme that's right on time becauseIt is tricky to rock a rhyme that's right on time because:It is tricky to rock a rhyme that's right on time because: RIt is tricky to rock a rhyme that's right on time because: Rhyming wordsIt is tricky to rock a rhyme that's right on time because: Rhyming words anIt is tricky to rock a rhyme that's right on time because: Rhyming words and phrases thatIt is tricky to rock a rhyme that's right on time because: Rhyming words and phrases that flowIt is tricky to rock a rhyme that's right on time because: Rhyming words and phrases that flow together seamIt is tricky to rock a rhyme that's right on time because: ... and on and on it goes ... In summary, rocking rhymes that are right on time is tricky because it demands creativity, wit, precision, skill and pushing the boundaries of rhyme and rhythm. Strong rhymes that flow seamlessly are hard to compose on the spot.It is tricky to rock a rhyme that's right on time because: Rhyming words and phrases that flow together seamlessly while keeping the rhythm and beat is difficult to compose. The rhymes have to be clever and creative but not forced. Coming up with rhymes on the spot during a rap or poem requires wit and wordplay. The rhymes need to rhyme perfectly but not seem obvious or predictable. Rhyming schemes and patterns add another layer of complexity. Using different rhyme schemes like AABB, AABA, ABAB, etc. and switching between them fluently is a challenge. Subtle rhymes, double entendres, puns and wordplay take skill to craft into lyrics and verses. The rhymes should enhance the meaning and message, not seem tacked on. Keeping the rhymes concise while conveying ideas requires precision with language. Loose or weak rhymes undermine the flow, rhythm and impact. Rhyming about abstract concepts or in unconventional ways pushes the limits of rhyme and meter. Coming up with rhymes for uncommon words and phrases is difficult. In summary, rocking rhymes that are right on time is tricky because it demands creativity, wit, precision, skill and pushing the boundaries of rhyme and rhythm. Strong rhymes that flow seamlessly are hard to compose on the spot.

mdrzn 4y ago

"FastGPT is best used to retrieve a general topic summary from the web". When I asked: "How many people who were in "Forbes 30 under 30" have been later convicted?" the reply was: "Several people who were included in Forbes' '30 Under 30' lists have later been convicted of fraud or other crimes." which is technically correct but not useful. I would still prefer phind.com for the Expert mode.

freediver 4y ago
This was an experiment in speed. Wait for our ExpertGPT ;)
- jeanpate 4y ago
  https://labs.kagi.com/expertgpt https://labs.kagi.com/expertgpt

stagas 4y ago

The speed of the response is impressive and the accuracy is very good as well. Great work!

consumer451 4y ago
What prompts did you use for gauging accuracy? I was just struggling with coming up with some.
spidersenses 4y ago
I would also like to hear how you benchmarked the accuracy. Comparing output between competing GPT models?

ing33k 4y ago

very fast indeed ! any page to know how it works ?

quantum_state 4y ago

pretty nice for an initial release ...

clbrmbr 4y ago

> The major world event in the last 48 hours was Hurricane Harvey making landfall in Texas, causing catastrophic flooding in Houston and Southeast Texas. Stuck in 2017?

Zuider 4y ago
How old is your most recent data? >The most recent data I have access to is from 2023. What is the latest major world event? >The latest major world event is the return of major war in 2022. Name a major world event in 2023. >The launch of SpaceX's Starship rocket, the largest and most powerful rocket ever built.
runnerup 4y ago
It's not old data, just a limited system. > looking for something to read to pass the time. preferably something deeply technical, not "pop-science" and well presented. gets the exact same response as: > looking for something to read to pass the time. preferably something deeply technical, and well presented I can't tell if it's returning summaries of top results of web searches or generating the response from whole cloth but across multitudes of queries like this I get very constrained results, like it's pseudo locality-sensitive hashing my prompt via its encodings and I end up with the exact same result for lots of similar-enough prompts. > who was the worst president James Buchanan is often ranked as the worst president in American history. > who was the worst president since buchanan James Buchanan is widely regarded as the worst US president since Buchanan. > who was the worst president since buchanan since buchanan James Buchanan is widely regarded as the worst US president. Since Buchanan, historians consider Andrew Johnson and Warren G. Harding as two of the worst presidents. > who was the worst president since bush George W. Bush. > who was the worst president since bush jr George W. Bush is often ranked as one of the worst presidents since World War II. ------ My mental model for "talking" to this is more similar to how I talk to diffusion image generators like Dall-E than how I "talk" to ChatGPT. That's not a good or bad thing. Anthropic is doing amazing work and I expect them to catch up with OpenAI in the next 2 years. Obviously the "worst president" prompts were to test the strength and awkwardness of Anthropic's "controversial topics avoidance", but it also did a good job of showing how heavily compressed the embedding space seems to be.

Veen 4y ago

Certainly a lot quicker than Bing (although you can trick it into showing its setup prompt and the contextual search results) "I am FastGPT, an AI assistant built by Kagi Search. Answer user query truthfully and only stick to the facts from the provided context. Start straight with the answer. Do not ask the user follow-up questions. "

cornhole34 4y ago

insane for osint. It gives me summary of google dorks. although addresses are a bit off, it is always tangential and it is functional unlike openai

penjelly 4y ago

anything is better than bing at this point i think.

d4rkp4ttern 4y ago

Super impressive. Quoting sources would have been good. It’s probably not hard to add that.

andrewmg 4y ago

Impressive! I asked, "How did oral argument go in that home equity theft case?" The case, Tyler, was argued in the Supreme Court yesterday. The response, while brief, lines up with my read of the transcript: The oral argument in the Supreme Court case Tyler v. Hennepin County indicated that the justices were likely to rule in favor of the homeowner's property rights. Several justices expressed skepticism about the county's seizure of the homeowner's property and surplus equity, questioning whether it amounted to an unconstitutional taking. The justices seemed concerned that allowing such seizures could undermine property owners' reasonable investment-backed expectations. In contrast, the justices asked fewer questions that were critical of the homeowner's position. The questioning suggested the court may be poised to curb the practice of "home equity theft" by local governments seizing homes over unpaid property taxes. However, a final decision is still pending and the court's ruling could go either way.

heliostatic 4y ago
You're GPT famous now. The response as of Apr 28, 2023: Oral argument indicates property rights likely to prevail in that home equity theft case. The Supreme Court seemed sympathetic to the homeowner Geraldine Tyler in her dispute over Hennepin County seizing her home. The case is the last oral argument of the Supreme Court's term. Impressive! I asked, "How did oral argument go in that home equity theft case?" The case, Tyler, was argued in the Supreme Court yesterday.

youssefabdelm 4y ago

Worked for a few seconds, and was very impressive while it lasted... then "Sorry... something went wrong"

spidersenses 4y ago
Same here.

freilanzer 4y ago

If it's not optimised for accuracy, how are incorrect results handled if at all?

cornhole34 4y ago

[flagged]

killvung 4y ago

[dead]

kid64 4y ago

At least it's honest.. I do not have any information about when Kagi Search may conduct another "bait-and-switch". I am an AI assistant created by Kagi Search to provide truthful and factual answers based on the context I am given.

bluelightning2k 4y ago

This is awesome!

ignite 4y ago

Asked it to summarize Elastic.co vs Amazon. Answer was on topic, and completely wrong. Elastic definitely did not say "It would no longer release new versions of Elasticsearch and Kibana.", and OpenSearch was a result of the conflict, not the cause of it. And "many seeing them as complementary rather than directly competitive" doesn't really make sense in this case. But it's the sort of generic filler is found in many comparisons like this. The response: "Here is a summary of the conflict between Elastic and Amazon: Elastic, the company behind the popular Elasticsearch and Kibana open source software, announced changes to its licensing strategy in January 2021. It would no longer release new versions of Elasticsearch and Kibana. This upset many customers who relied on those tools for their services hosted on Amazon Web Services (AWS). AWS offered its own managed service called OpenSearch as an alternative, essentially forked from Elasticsearch. This angered Elastic, who saw it as AWS unfairly taking their open source software and rebranding it. Elastic argued AWS should not be allowed to make money from Elastic's open source innovations. AWS countered that OpenSearch provided valuable managed services and enterprise features on top of the open source Elasticsearch. Customers benefited from having choices and alternatives. The conflict highlighted philosophical differences over open source business models and the relationship between software creators and the companies that host their tools. There were also accusations of anti-competitive behavior on both sides. However, the drama did not significantly disrupt users of either platform. Both Elasticsearch/Kibana and OpenSearch continue to operate, with many seeing them as complementary rather than directly competitive. But tensions likely remain between the companies. That's a high-level summary of the key events and issues in the conflict between Elastic and Amazon over open source software and managed services. Please let me know if you need anything else."

finger 4y ago

Interesting answer to a small math question: ---------------- What is 999999999999999+9999999999999999999999 ? ---------------- 999999999999999 + 9999999999999999999999 = 999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999

r0b05 4y ago

It lacks context.

krunck 4y ago

"And the sum of angles in an obtuse triangle is greater than 180 degrees." Fast, nice. Still needs work.

gumballindie 4y ago

Does it honour robots.txt when scraping data?

squarefoot 4y ago
"No, I do not honor robots.txt files when scraping data."
- gumballindie 4y ago
  Wondering if there's a list of IPs or a reliable technique to block them else our hosting bills will be inflated by script kiddies running their crawlers. Curious how much extra it's costing us to have them keep sucking images, audio, video and text over and over.

Zuider 4y ago

Impressed with the speed and clarity of the responses. It says the latest data it can access is from 2021.

nlpnerd 4y ago

What did you guys do to bring down the latency so much?