jedwhite

19 ms·

1. 151.▲
   by jedwhite 5y ago

   As a coder, I don't think there is anything I could read anywhere that would make me smile more. Thank you seriously! We've had many requests for enabling URL history and shareable URLs. It's not exposed in the UI yet but you
2. 152.▲
   by jedwhite 5y ago

   Thank you for giving us helpful feedback, and taking the time to try out Andi and give it a good workout! Your example questions are a great chance to explore how you can get better results with language models and semantic search by asking
3. 153.▲
   by jedwhite 5y ago

   Thanks. There are definitely edge cases from a technical perspective that are common in the wild. As the models get better these will get easier to handle, but we think that we can help guide users to asking more specific questions in order
4. 154.▲
   by jedwhite 5y ago

   Thanks for trying out all these questions on Andi and posting the results here. That was really exciting to see!! If you have access to any of the GPT-based Playgrounds, you'll see that large language models on their own tend not to be
5. 155.▲
   by jedwhite 5y ago

   I didn't realize when I first replied that you're Vlad who runs Kagi. Thanks for taking the time to try out the new Andi Q&A features and give us feedback on some of the trickier types of NLP queries. As someone running an alt
6. 156.▲
   by jedwhite 5y ago

   Yes! We're adding Report and Blacklist tools that let you block sites locally, and working to get that live asap. We were trying to have them included for when we posted this, but I was having problems getting the Q&A feature stabl
7. 157.▲
   by jedwhite 5y ago

   Thank you! Yes that's a great suggestion to add a "flag" feature on the answers specifically. We're adding tools to report and block sites locally (for spam), but "report a bad answer" vs "report a bad sit
8. 158.▲
   by jedwhite 5y ago

   Thank you! We know we have a lot of work to do on the UX. I'm the sole developer, and I'm definitely a back-end guy. We're working on improvements based on suggestions and feedback. The feedback so far is that the UX works be
9. 159.▲
   by jedwhite 5y ago

   Thank you sincerely for trying out Andi, and for the encouraging feedback! We know we have a lot of work to do. As a tiny team with a single developer, we're not expecting it to be perfect on day one. But as you say the results are pre
10. 160.▲
    by jedwhite 5y ago

    A friend mentioned to me that you're a software engineer at Google. Thank you for taking the time to have a look at what we're building with Andi, and giving us detailed feedback on the results. I mentioned that we don't log
11. 161.▲
    by jedwhite 5y ago

    Why, yes. Yes, I am :)
12. 162.▲
    by jedwhite 5y ago

    Thank you! It is so different it takes a while sometimes for people to grok it. But I'm 100% with you on that. A community member on Discord told us about how the conversation session with quick nav between searches along with reader-m
13. 163.▲
    by jedwhite 5y ago

    Interesting article because it talks about all the things that happen with the use of large language models on their own. Large language models are amazing at mimicry and composition, and are a key part of getting to great Q&A. But on t
14. 164.▲
    by jedwhite 5y ago

    Thank you!! It's not deterministic, and it's retrieving live content in real-time with each search, so there will almost always be variation if you do the same question multiple times in a row. I love both these answers. The GPT p
15. 165.▲
    by jedwhite 5y ago

    Thank you, we really are under no illusions as to how crazy-impossible taking on Google is. We don't really see them as a competitor. They're more the problem we're trying to solve. Google was once a search company. Now it&#x
16. 166.▲
    by jedwhite 5y ago

    Thank you. We really are trying to keep a good balance here as you say - strong branding for the original source using their own Open Graph tags for title, description and image (we try to respect the intent of the content producer). And so
17. 167.▲
    by jedwhite 5y ago

    From the Smithsonian Magazine no less :) The semantic search approach does throw up some interesting things sometimes! Thank you for trying out Andi. And sharing something I never knew too! I'm a little shaken by the four headed image.
18. 168.▲
    by jedwhite 5y ago

    I don't know picodguyo either, but to be fair, there really is a flourishing of innovation finally happening in search. Finally. After 20 years of zip innovation. Marginalia, Brave Search, DuckDuckGo is doing more than ever, Mojeek is
19. 169.▲
    by jedwhite 5y ago

    Advertising is the corrupting influence that has ruined the Internet and made using the web awful :( Google can only get away with it because they have a browser distribution monopoly with Chrome, Android and their deal with Apple. If Andi
20. 170.▲
    by jedwhite 5y ago

    Thank you. Reader view is my favorite thing we've built. The web has become a cesspool of ad-tech to the point that most of the news or articles you try to read are simply unusable. Pop-over/under/auto-roll ads, no visible co
21. 171.▲
    by jedwhite 5y ago

    Thanks for trying Andi. Depending on the query, you will get very different sets of results. Currently we fall back to traditional-style web search about half the time, but as we build out our own data stores and vertical indexes with speci
22. 172.▲
    by jedwhite 5y ago

    Thank you so so much! We know this is a very different approach to search, and we appreciate and want to talk with every single person who is giving Andi a good work out. So please let us know if we can help in any way at all!
23. 173.▲
    by jedwhite 5y ago

    That's a great question. Two things here: 1. The reader view only displays publicly available web content (it is made available to search engines and any public web client. We render through an anonymous proxy to protect user privacy.
24. 174.▲
    by jedwhite 5y ago

    Hey thanks for trying out some searches. Depending on the intent of your search, we try to predict the best source of knowledge to find an answer. Computational or simple knowledge graph queries like those will often get routed to Wikipedia
25. 175.▲
    by jedwhite 5y ago

    Thanks for trying it out. I'm a Type 1 Diabetic, so health search is one of the areas I really want us to help improve for the world. We have a long way to go but it's one of the suckiest areas of search with spam and clickbait ta
26. 176.▲
    by jedwhite 5y ago

    Oh no, so sadly we have our own Blue Screen of Death! And it's reproducible as I just got it for that search too! I'll report back once I've fixed that! Thank you sincerely for trying out Andi. There is a lot more coming with
27. 177.▲
    by jedwhite 5y ago

    Programming search is both one of its strong areas (we built Andi using Andi), but also the one with the most still to improve. So I'm grateful you tried it for that. We're doing pretty well with binning copycat and link farms in
28. 178.▲
    by jedwhite 5y ago

    Thank you so much! Andi's strong point is these type of factual answers, and I think you're 100% right to be optimist about models being within reach. The question answering feature has only just been released for the first time t
29. 179.▲
    by jedwhite 5y ago

    I think these go in our "let's break the AI" hall of fame :) Having search results alongside answers does lead to interesting serendipity, particularly if you asking compound questions that span a few topics. When I was testi
30. 180.▲
    by jedwhite 5y ago

    Thank you! Andi is starting to get really good at those sorts of serious questions, and that's a great example. The trick questions that everyone likes to ask any new AI give us great training data for improving the intent-detection ed