Launch HN: Hello (YC S22) – A search engine for developers

Thought I'd try this on a problem I've been researching today (which I resolved) where my service worker for offline PWA usage was working for everything except audio files.

I searched the following in say hello.so.

"Service worker fails on request for audio file"

I got back a couple of results related to general service worker use but none that get close to discussing the core problem that lead to the solution.

The same query in Google returns several results that together pointed me to the solution (it was around range headers in requests for media data types).

This is just one example though. I think the problem you are trying to fix is worth the effort. I just wonder if this is where humans are still stronger than computers - gathering unstructured data to use in problem solving.

The description of the steps you took is super helpful feedback - thanks! Hello performs best on "how-to" questions at the moment. We're still working to improve troubleshooting type queries.

That'll be a difficult adaptation for potential users to make. I think most of us have been conditioned to phrase our queries a certain way to achieve the best results from Google.

Then again maybe that's just me.

You're right in that moving away from Google is a huge behavioral change. While our model does prefer natural language queries, what I was really referring to are the different categories of developer searches. Ad-hoc how-tos are a type of search where the developer knows what to do but not exactly how e.g "how to set a cookie in fastapi." Troubleshooting searches are things like copy-pasting a compile error, or "why is X behavior not working."

I may not be representative, but the vast majority of my Google searches are troubleshooting.

If I need to know how an API works, I go straight to the docs.

Put another way, troubleshooting is a several times per week task for me, and doing something net new is a once a month task which may not result in a search query.

You’re going to have a hard time staying on my radar focusing on the less frequent case.

Possibly the traffic patterns are different for more junior devs, but my hazy recollection of my early career is spending most of my time troubleshooting.

Thanks for the perspective - we're working on improving our functionality for troubleshooting.

Is the assumption you are making that most developers would go to search first? rather than when they hit a blocker or error?

We will support troubleshooting type searches in the future. It's just not available right now.

Congrats on the launch! I love this idea. I've thought for a long time that something like it should exist. Google results are often lacking in this realm.

I've played around just a bit and clicked some of the preset examples and like what I'm seeing so far. I bookmarked it and will try it out more as I code over the next few days.

Main initial feedback: I'd really like to see version/last-updated-at info accompanying all results. One of the biggest problems with Google for code stuff is finding outdated examples and docs. Even better would be a dropdown that lets me see results depending on the version of the language/framework/tools I'm using.

Thanks for trying it out and good point - we'll look into adding version info

Hey so I built a search engine doing largely the same thing (and also interviewed with YC during the W20 time) and ultimately we pivoted away due to lack of interest from the developer teams we were pitching, often the startups we were pitching didn't have enough accumulated internal knowledge for the paid plan to be useful. For the ones who did (at the like 200+ person 5+ yrs in business mark) we still weren't seeing the problem being painful enough where companies wanted to pay to solve it.

How do you see navigating this space when this can be considered a nice to have versus a strict need?

Right now we're primarily focused on building a search tool that developers love. Would love to chat more about your experience - shoot us an email at [email protected]

Personally I’ve never really had an issue with Google - I think my mental model with how it works is to the point it makes sense.

It would be amazing if this could be used for internal documentation however. Like we have so much documentation on our wiki which is just disorganised.

Stack overflow offers a version for companies. I've never used it, but it sounds like what you might want

I can just imagine it will spawn the same "closed as off topic" and other similar responses for most questions :P

Also, stack overflow's search has always sucked. The way to find stuff on stack overflow has mostly been to use google.

Indexing internal docs is one of our ideas for how to monetize. And it'd be great if you could tell me more about your mental model while using Google -- are there inconveniences that you brush aside?

This seems oriented towards answering a question like “how do I…” where you expect an explanation, but I’d say most of the web searching I do is pretty specific.

Maybe half the time I know what I want (eg: the order of values in the animation CSS property), and from who (eg: MDN), so I just go to the relevant docs page via google with something like “MDN animation css”.

The other 50% of the time, I’m searching an exact error string, probably in quotes, on google. For that I also don’t really want a knowledge graph answer, I’d much rather see a GitHub issue or stack overflow post and I’ll derive the context I need.

We empirically find that most developer searches fall into a few categories consistent with this paper [0]. The author explains these categories nicely in her blog post [1]. It sounds like you're describing 4/6 of these categories: ad-hoc how-tos, understanding an API / recalling forgotten details, and troubleshooting.

We currently do well at answering "ad hoc how-to" questions and are working to improve our answers in the other categories. This could either look like augmenting our existing natural language answer, or building a separate view specializing in official documentation or errors.

[0] https://elisehe.in/i/googledrivendevelopment.pdf [1] https://bootcamp.uxdesign.cc/the-hidden-insights-in-develope...

I ran a search for 'How to scrape Linkedin Profiles'. At a quick glance 1 out of the top 4 results actually contained some instructions. The rest was more SEO-bait. Just thought I'd share with you - not sure if this is the exact use case.

Congrats on the launch! I can vouch that this is indeed a problem devs face, especially new devs. The number of times I've had to steer away my students at Edyst away from SEO optimized articles!

Possible edge cases that aren't handled yet. https://imgur.com/a/mieqZIU

Though I know the first search term was unrelated but I tried it for some time as a regular search engine, tried a bunch of random keywords apart from what it was meant to do as well as some that can be legitimate questions as well.

I am liking the product quite a bit however. Good stuff.

Congrats on the launch! Looks promising, so I'll try it out for a couple of days.

One feature request at first glance: please default to the system font stack for code snippets. I see you're currently using Consolas, a Microsoft typeface, which is not pleasant to see as a mac user.

You can use this to default to the system font on every platform:

    font-family: "SF Mono", "Monaco", "Inconsolata", "Fira Mono", "Droid Sans Mono", "Source Code Pro", monospace;

Why do you consider it unpleasant? I’m a mac user and I really like Consolas. I like to use it in VSCode or when building websites that display code blocks.

Thanks for the feedback, we'll take a look at that :)

Not to be too critical but the results I got so far have been subpar. Seeing a lot of hyperbole/clickbait articles.

Let's say I'm searching for front-end frameworks. Each article has the word "best" in the title, yet doesn't link to resources like State of JS, Stack Overflow Survey or other similar sites. So, in this context "best" is subjective. I can't be bothered with subjective results when I'm trying to find out what is actually considered "best" or in this case popular.

Those articles are coming from Bing as of right now. Our offering is based on analyzing those articles and summarizing them/picking out the most relevant parts. We definitely plan to augment (and eventually replace Bing) with our own index.

We have considered blacklists but haven't looked into Goggles yet. Thanks for mentioning this

Really curious how this worked, the query 'FHIR appointment spec' produces the following. It actually did get the right result as the first result.

"fhir appointment spec

I'm not sure what you're asking about, but I'll try to answer it as best I can. __ Appointment is a FHIR data type. It's a way to describe a time slot for a patient to be seen by a healthcare provider. Appointments can be booked, cancelled, rescheduled, or canceled and rebooked. It can also be used to describe the location of the appointment. "

Pretty impressive summary given that it doesn't exist in any one specific page.

Thanks :) We've gone all-in on using our AI to answer questions based on information from multiple sources.

Pretty decent overall. Though some code that comes back is a little dense and hard to read. This appears to be caused by how the scraper/parser handles whitespace. For example, the query "make a curl request in php" returns an example from the comments section in the PHP docs where the entirety of the comment has been returned as code, but the <br> tags have not been converted to newlines.

It would be a good idea to preserve whitespace, or arguably better, integrate optional syntax formatting

Overall, this search engine looks promising

Thanks :) we'll take a look at better syntax formatting for the code snippets

UI/UX first impressions feedback: Unnecessary animations are annoying especially, you can rid of all the animations with no any exceptions (think of all the applications that we use everyday and designed by biggest companies, they all are free of animations). The same thing applied to shadows, someines shadows add extra complexity to the UI.

Thanks for the feedback. I agree that the animations can get old if you're using the search every day. We'll look into simplifying.

Yep, the AI definitely prefers fully formed sentences as it is right now. We know it's important not to force users to change how they word their queries, so making it less sensitive for phrasing is a priority for us.

Searching "meta class python" gives better results, which seems reasonable to me.

I see whole solutions copied from other websites displayed on your site.

Is that legal?

Isn't there copyright on those?

Ding ding ding. This is the exact issue that a vocal minority are whingeing over github copilot for. It's automatically pasting results from websites without embedding the necessary attribution - so if you copy entire functions from this search engine (which may be coming from stack overflow for which attribution is required), then you're guilty of the same thing.

So I only see one of two outcomes:

1. Courts rule copilot is fair use in which case your search engine becomes largely superfluous

2. Courts rule copilot is infringement in which case all of these types of applications cannot be used commercially

I would hope it's not legal.

> Hello pulls in the raw text from all five search results linked on the search page to generate...

Not to be negative but I think I'll stick to the sites and people that made the results and not a middleman that intends to charge for other people's work.

I think that framing is missing some nuance. Seems more like they would be charging for the process that goes into sifting through those results and pulling out other people's work on the user's behalf.

Yeah, I can see there's a lot of work going into the process but it's still using other people's efforts without permission or compensation.

> using other people's efforts without permission or compensation

Like every web search done on Google? That said, I think attribution links should be displayed, license too if available. Copilot should be doing the same to defuse this discussion.

First Impressions:

* I won't use a different search engine for programmers stuff vs everything else. So while this might be targeted toward software developers I can't see myself using it unless it can handle normal searches.

* UI/UX - I hate the progress bar, I'm not sure at all what it's telling me as there are results shown while it's still completing. The results are way too spaced out. On my 27" 2K screen I can only see 3 results, the search bar takes up way too much space and there is way too much padding on the results. Don't move the DOM on me, removing the progress bar is jarring as is "Was this answer helpful?" popping in, I'm here for results, not to train your ML.

* Trackers - Using the default installs of Privacy Badger and uBlock Origin meant no results ever loaded. I'm not sure what was being blocked that caused the issue but cookies from bing [0] and a request to cloudflareinsights.com [1] should not hamper showing results.

Search is a tool and one that I need to be quick, simple, and informationally dense. This checks almost none of those boxes. I'm even open to using a different search engine (I semi-recently switched from Google to Ecosia and it's been near-seamless), but I don't see any "pro" to using this engine and I see a ton of "cons".

[0] https://cs.joshstrange.com/V5uiyM

[1] https://cs.joshstrange.com/GkPuap

EDIT: I did a few more searches because I realized I wasn't getting the "info box"/ML results on my first few searches and I wanted to be fair. Sorry but that made me dislike this even more. I really, really hate content moving out from under me. My eyes start reading one of the 3 results that was shown then they got pushed down for another overly-padded box that tried to "answer" my question. The results were worse than "grab the selected answer from the first SO that matches this query". Maybe it would be better if that info was shown off to the side and didn't move the results when it loaded in but again, I didn't find it useful in the queries where it showed up.

EDIT2: I posted a follow up comment about what, specifically, I think should be changed: https://news.ycombinator.com/item?id=32005841

I definitely see the benefit of a separate code search engine and would use it. Google has, as you have notice, gotten way less useful. A big part of that is trying to target the wider market to the exclusion of less popular searches, like developers'. A dedicated engine would help with that.

I'm more than willing to open another tab to not have a search result page full of YouTube videos.

To each their own. I have 1 flow for search which I don't plan on changing and I personally have no issues with google search (for code, technical, or otherwise). I still consider it to be one of the best and I don't agree with the "google search is getting worse" crowd. Maybe at some point I'll use some kind of "search hydra" that hits multiple engines and either combines the results or shows me the results based on the type of search it thinks I'm doing but I don't imagine I'll ever want to consciously switch engines based on task.

Also I've never seen a "search result page full of YouTube videos" no matter what the query was. Sometimes there will be 3 or a carousel of them near the top but I can easily ignore that (assuming they aren't relevant or useful to me). I can't remember the last time I got a video for a code/technical query on google, I just did some testing and only a few queries showed videos, always 3, always partway down the page so that the search results at the top answered what I needed before I even got to the videos.

In your first comment, you say you want high information density. In this one, you say if there are irrelevant results on Google, you'll just ignore them and scroll past.

I get that there's always a high bar to switch to a new tool, and Google obviously has certain advantages that are tough to replicate, but it seems like you're applying a bit of a double standard. If Google shows me a bunch of ads and irrelevant results that I have to parse through to find what I'm looking for, that's not high information density; it's quite the opposite.

Both things can be true and you are omitting/ignoring part of my comment about irrelevant results. Higher density of information means I can scan faster and see more with less/no scrolling. As for irrelevant results I called out that the relevant results were above the videos (irrelevant, to me in this context, content). Lastly, I very rarely get ads when doing technical searches so that doesn't really factor in here. I'd wager I get what I want within the first 3-4 results on google reliably (at least when it comes to technical/programming-related searches) and when I don't it's normally super niche (it also rarely has videos or other stuff in the results, think: error messages).

Right, but you still have to click on those links manually and potentially scan lots of text yourself for relevancy. Our goal is to automate that.

I understand that, unfortunately your goal is still in the future (be it a week, month, year, or decade I can't tell you but it's not there today). In the interim (and if you want users while you refine) your search results should at least be on par with Google/Bing/etc. That way the "happy path" is your ML spits out the right answer and no links need to be clicked but if your logic can't come up with an answer or if it comes up with the wrong one you need the results to be a viable fallback.

EDIT: Building on what I said:

I use Github CoPilot and have been very happy with it. It's far from perfect and even when it spits out good code I have to do minor cleanup but it does save me time and "sparks joy" when it works. When it doesn't work it doesn't really get in my way. If CP required I change my entire method of programing, IDE, or if I had to go into a "special mode" to use it then it would be next to worthless to me.

As it stands you don't have a good fallback (regular results). Your product should be additive to what currently exists in the space. Not "a step forward if we guess the correct answer and a massive step backwards if we don't". I 100% believe you can make changes such that the results function as a perfect fallback (I've outlined them in various places of this thread).

Not the parent poster, but I think there is an important distinction between information density and information relevancy. Google is information dense (compared to this), but not all of it is relevant. This search engine aims for higher relevancy, but the density suffers from the stylistic choices of the creators.

This is exactly what we think. And yes, we definitely plan to do better with space efficiency

With one dominant search provider, we are constantly conditioned to parse the results from Google. It is very easy to ignore certain parts of the page and mentally process which parts of the page your brain is interested in.

So, while it’s not necessarily “information dense”, finding the information you need is comparatively “cognitively light”. Or at least predictable…

The devil you know…

You're right -- we are all conditioned by Google. Yet, using Hello for my own technical searches these past few weeks, higher signal and lower noise is much better for me personally. At the end of the day, Justin and I are making the search engine that we want to use ourselves as developers.

Fair point, but I think the technical term for that is "Stockholm Syndrome". Or perhaps in this case, we should call it "Mountain View Syndrome".

I think you missed the forest for the trees... it is a Q&A system not just a search engine. You can talk to it, you can refine your queries.

To the SayHello team: kudos for being faster than Google to release a Q&A+search system. I was expecting something like this for a couple of years wondering why Google was sleeping on its mountain of papers and not doing it.

Search was the first step in finding information, Q&A is the next logical step. Language models+search such as DeepMind RETRO have shown this approach to be very efficient: 25x reduction in model size for the same perplexity and verifiable correct answers with source document references.

In the future I expect search to become more like an assistant with context and language abilities. Retrieving a bunch of web pages is so 2000's. Q&A is especially relevant for mobile use with speech interface (hello Siri and Google Assistant).

> I think you missed the forest for the trees... it is a Q&A system not just a search engine. You can talk to it, you can refine your queries.

From the creators:

> We're building a better search engine for software developers.

Also no you can't "talk to it", I'm not sure where you got that idea. It has a "Ask a follow up" but that performs a new search with none of the context of your previous search (also this UI of sliding a modal up from the bottom and layering the results is terrible).

> Search was the first step in finding information, Q&A is the next logical step.

And we are clearly not there. Not only does this not allow you to ask follow-ups to refine but it doesn't give good results in my testing.

Could you post which examples you tried? It's not perfect, but the "Ask a followup" feature is usually smart enough to use existing context to refine.

I asked a question with "php" in the search then asked a follow up without including "php" in the search and it gave me python results. Also, and I'm sure this is a code formatting issue, the PHP code is invalid (newlines appear to be missing, the important one being after the opening tag) [0]

[0] https://cs.joshstrange.com/RP4B59

I don't see a followup in the screenshot you provided. Again, we're definitely not perfect with followups, but we generally do capture context. It would be very helpful if you could tell us your original search query, your followup, and your intent with this question.

That wasn't what I searched for and did a follow, here is a repeat of the original and follow up I tried (or as best as I can remember, I know it actually showed me python code last time, now it's showing SQL and python results).

Original (php get end of day timestamp): https://cs.joshstrange.com/50y9Pf

Follow up (end of month timestamp): https://cs.joshstrange.com/hStUEm

Sidenote: This slide over modal (that hides the results of the first query, you cannot get to them once you do a follow up) has got to be the worst of all worlds. It's very unintuitive, doesn't provide any value, makes your initial query only as useful as what you can see "before the fold". Just redirect to the new results or append them under. This modal is frustrating to work with, it's a weird parallax-type thing: https://cs.joshstrange.com/d6rhPZ

> * I won't use a different search engine for programmers stuff vs everything else.

I'd use this as a ddg bang[1]. I don't use them often, as ddg is a great search engine, but some search engines handle certain queries better and ddg lets you route queries efficiently.

https://duckduckgo.com/bang

Thanks for the feedback. We agree that speed is incredibly important, and we're working on making searches much faster. We'll be iterating on the UI/UX as well, as we think that we can definitely do better and be more efficient with space.

I'd love to hear more about what you mean by "informationally dense" -- some search engines simply show more information on the results page, but that doesn't make results inherently better in my opinion because it frequently simply increases noise relative to signal.

Our current approach is to provide only the most relevant answers/code snippets and nothing else (high signal with low noise) as opposed to cramming in every Stack Overflow answer we can find. We realize we still have a long way to go to make it magical for every search, but we're working on it.

Consistent, non-moving (as things load) UI is super important. Also something that bugged me but I didn't realize why until now: don't use the full width for the description under links (or the titles for that matter). We've known for some time now that if you make text too wide it becomes harder to read.

My suggestions:

* Kill the padding/margins, it's pretty for demos or certain cases but I want to be able to see more information, heavy padding/margins have no place in search results.

* Shrink the search bar to the upper left like every other search engine. Keeping it centered with tons of padding wastes space. Take your logo and put it to the left of the search field, take the buttons and put them to the right. On my screen you are burning a little over 500px of vertical space with things that don't matter, the results matter.

* Shrink your "regular" search results to be half the width of the screen (on desktop, something like a max of 700, Google uses ~640 as does Ecosia). Use the space to the right to show your AI/ML results. This means no content will jump around and people can more easily read the results, full-width is very hard to read. Also shorten the "description" under the links. 2 lines max (at 640px width).

* Either don't ask "Was this answer helpful?" (use hints like: Did they click the link? Did they leave the site after seeing the results?) OR don't make it move the content (hold the space empty if you must animate it in, just don't let the content shift multiple times after doing a search).

Here is your default result for "this is a test" search query: https://cs.joshstrange.com/oKbz6G

Here it is with a bunch of padding/margins removed: https://cs.joshstrange.com/VEVXGh

Yes, I removed the logo/buttons because that was faster than moving them to the left/right of the search but the end result is the same. In my tightened up version you can fit 8+ result links where the initial version could only show 3, also all the results are easier to read.

Thank you, the cleaned up version is helpful. We definitely have a lot of UX work to do :)

> some search engines simply show more information on the results page, but that doesn't make results inherently better in my opinion because it frequently simply increases noise relative to signal.

But if the "automatic" answer fails and I need to skim results, as I'll often need to do, you put 3 result previews in a space DDG and Google fit 5. They also apply reasonable defaults for the max line length - a basic typography thing that improves quick readability a lot.

To comment on the dynamic DOM - we're displaying up to three answer types (text, code, links) for each question. We're loading them independently to get information to the user as fast as possible. The alternative (in this state) is to have all of them wait until the slowest component finishes. We're still in the early stages of development, so either way it's not going to be perfect. I can see how this can be a poor experience for some - we're working on it.

Side-by-side is the best way to handle this. Show results on the left and load in the ML stuff on the right after it loads. This prevents content-jump and makes the results less wide (you want to aim for <700px to be more readable).

Maybe not everyone understands that the Q&A responses have to go through a large language model before they are displayed. This takes time, showing something while the LM is churning away is a good idea.

Yep, the current loading bar attempts to do that (replaced by the slowest generated answer when available). Definitely looking into new ways to improve speed + loading experience if necessary.

A placeholder empty box that gets populated when the content arrives would be an improvement.

Also the scraped snippet appears at the top of the results a couple seconds after the results load and it causes all the results below it to suddenly jerk lower on the page

Lowest common denominator and "one click" is killing the internet for me. A lot of times I am the lowest common denominator and so that's fine. But when I am a specialist I want something that lets me be specific and gives me specific results.

How much more specific do you need than having the ability to refine your question iteratively? I think regular search engine only allow for a few special keywords. Here you can use natural language to refine.

Sorry, I meant this in response to the part of the parent's comment about not wanting a specific search engine for code. To make it clear - I could see myself using this engine.

UI: Feels overpadded to me, I'd like to be able to see more stuff without scrolling so far

a search for v8 gave me:

* juice

* v8 engine

* juice

* v8 engine

* juice

so definitely some non-programming searches showing up, unfortunately none of the documentation sources for v8.

Yep, I saw non-programming stuff but my worry is, if they are branding themselves as "better search engine for software developers", that programming-stuff will be weighted higher than non-programming stuff even if the search term has nothing to do with programming (or a tenuous link). Though your search examples seem to prove the exact opposite.

All that said the UI/UX is too frustrating to use (as-in) even if they don't promote programming content over non.

We learned many of these lessons at https://you.com/code:

* we also needed to build a strong "everything else" search engine and then

* have great results for coding with specific search apps like StackOverlfow, AI code complete, ++

* be very fast (we messed that up when we first launched)

* have great scores on Privacy Badger, be compatible with uBlock, etc.

Last week we've started opening up our platform to collaborate on results with outside developers and have gotten a lot of interest: https://about.you.com/developers/

Maybe we can collaborate also with you guys at sayhello. Ping me at [email protected] if you want to compare notes.

I work on building language models at FAANG for my day-day job and I'm very curious about this - how does finetuning on T5 remove hallucination? Can you elaborate on this?

This is an amazing product, btw. Let me know if you're looking for people to hire :)

We don't use T5 exactly, we use a derivative that has a similar (but not identical) architecture and is also pre-trained differently. That model, combined with our factual generation dataset and clever prompt engineering, seems to be the secret sauce for reducing hallucination.

And thank you :) it's comments like this that really fire us up

> We found that GPT-3, GPT-Neo/J/X, and similar autoregressive language models that predict text from left to right are prone to “hallucinating” and generating text inconsistent with the “ground truth” document.

The term hallucinating is brilliant for how these AI systems seem to generate output.

Your product is very interesting, seems to work nicely on easy queries "how do I sort an array of objects in JavaScript". But was quite confusing for complex queries.

The UI doesn't work too well on mobile, but it's a beta and software is written on the desktop.

I also think making this a specific search engine for companies internal messy data would be a very useful tool as well.

Thanks for trying it out! We've gotten some solid feedback about the UI and will be working to improve it. Could you tell us a bit more about what you were trying to do and how it was confusing for complex queries?

Hallucination is, funny enough, the technical word for this phenomenon from NLP research :)

This suffers from one of the things I've been hating about Google lately: it doesn't use exactly what I typed in. Case in point: "react-native-navigation" is an entirely different package than "react-navigation". I query about RNN and get results about RN. I get this is due to Bing, but it could be a fundamental flaw with the approach (for me at least)

The animations and page jumpiness are a bit off-putting and slow, but it is a beta!

Yep, we feel that frustration -- it's our intention to use exactly what you typed in, but here Bing is messing up. Using results from our own index should make this less of an issue in the future.

https://beta.sayhello.so/search?q=how+to+base64+encode+a+str...

Query: how to base64 encode a string in ruby

Response: I'm not sure what you mean by "base64 encode a string in ruby" - that's a bit of a misnomer. Base64 encoding is a way of storing data in a form that can be decoded by a human. It's not a secure way to store data, but it's useful if you want to send a message to someone who doesn't understand the language you're using.

The right answer is in the third link provided but it's not exactly correct.

Google gives back the Ruby Module Base64 docs as the first hit.

Our index is based on Bing as of right now -- if they give us low-quality results, our generated answers will be low quality as well. We're definitely aware of this and are working on developing our own index to augment Bing's in cases like this.

while i assume there are good business reasons you're basing your stuff on Bing, it's notable that as a general rule developers don't use bing. In my experience the google results are radically better.

> while i assume there are good business reasons you're basing your stuff on Bing

I'd strongly suspect it's the same reason DDG does so: Bing offers a search API, and Google has no incentive to offer access to its index and results

There are plenty of companies that make a living scraping results out of Google, so I don't mean to say it's impossible, it's just a monster amount of energy playing cat and mouse with what is effectively an unlimited budget to stop one from accomplishing that goal

How do you plan to index internal documentation given that you currently use Bing?

Maybe I'm misunderstanding the intended scope of this engine, or I just ran into a bad result page, but:

https://beta.sayhello.so/search?q=Java+aot+compile

Does not seem to mention graal anywhere. (It's just a random test query that popped into my mind)

Asking a full question for a code snippet seems to work: https://beta.sayhello.so/search?q=How+do+I+sort+a+map+in+Jav...

How do you deal with licensing for these snippets though. Is that up to the user to verify?

Because we're currently built on top of Bing's index, we're somewhat dependent on the raw pages they provide. If those pages don't mention graal, neither will our AI. Building out our own index is something we're working on.

It is currently up to the user to verify licensing for the snippets, but we try to make it easy (using the See Reference button) to go to the original source.

> we’re planning to charge teams per user/month to use on internal data scattered around in wikis, documentation, slack, and emails.

Do you have any idea of how you're going to go the "enterprise integration" route without hiring an army of implementation consultants?

best of luck, I'm sure there are many teams on Confluence that wish they had a functioning search without moving everything off Confluence at once

I'm seeing the same page as result 1, 2 and 3. Interestingly only 1 out of those 3 results were scraped from that page. Even more curiously only 1 out of those 3 results were even valid code.

https://beta.sayhello.so/search?q=hello+world+in+brainfuck

Nice idea for the project though. Good luck with it

Our code extraction/ranking model hasn't been trained on that language yet, so it's definitely an out-of-domain example. We'll keep working on expanding our repertoire!

Ahh that’s fair enough. I think it’s fair to say “brainfuck” is outside most peoples domain. I was just curious how your search engine performed on less common search queries (the kind that are trying to debug a rarely hit problem with an otherwise popular framework or language and thus you often spend hours digging through irrelevant answers before you find that one blog post that solves it) but couldn’t think of a more realistic example off the top of my head.

Specialised search engines are the future in my view. Google is an ocean of data great for generalised search. But as soon as you want to narrow down to a specific domain it’s mostly noise - in my personal experience.

Hey so, I have been working on a specialized search engine (at least you can think of it in this way). And for me a lot of the gains came because we could structure and restrain the search space in a meaningful way, such that we could build better distance metrics than a pure text search could do.

I wonder what you think about that. Maybe one could submit a code snippet, or mark something as an error, or ask for a refactor of some code. But then again, this gets close to what copilot is doing.

We are absolutely experimenting with non-text inputs (and outputs as you can see). A big problem we see in mainstream search engines is that syntax is not parsed well. Understanding code as context for the query could be huge for developer search.

I love the idea <3

It may be a weird suggestion, but if the query to general topics returns something like this https://unzip.dev/archive (check how compact it is and delivers almost all you need to know about the subject to get you going), it would be perfect.

Thanks for the feedback - I agree that explanations aren't the most appropriate for every kind of search. We're definitely considering different forms of search inputs and outputs (text, code, lists, etc.)

And what about books, just links to the most important books on the subject, without getting too philosophical about how to determine the most important ones.

One of your examples at the bottom of the page, is the "IsPalindrome". I guess the assumption is I would click on the "see Reference" as it doesn't provide any context for the code. This context of a real person explaining it is one of the benefits of sites like Stack Overflow, so I would think about this element of UX.

Also, I noticed in your palindrome reference example, it didn't choose the accepted answer from Stack Overflow. How did it choose the example? Also, the 2nd 2 reference panes, I can't tell what value they are adding. They seem like a list of random outputs of the ispalindrome script.

Our code ranking algorithm uses an NLP model (trained on a code dataset) to pick the most relevant snippets. You're right that the accepted answer on Stack Overflow is a good heuristic, and it's something we'll add to our ranking algorithm in the near future.

Showing an answer written by a human as a part of the code snippet is also a good idea.

Can you give an example of using the context of a previous query?

I applaud you for trying to make a new search engine - it's not something sane people would try to do because of a certain behemoth eating everyone's lunch. It's going to take extraordinary insight and out of the box thinking to get something really good.

Our initial prototype focused on conversational search, with the ability to ask follow-up questions.

Here's a rather trivial example: Q: "Who founded Y Combinator?" A: "Paul Graham founded Y Combinator with Jessica Livingston and Trevor Blackwell."

If you scroll to the bottom of the answer page to ask a follow-up question: Q: "How old is he?" A: "Paul Graham is 57 years old. He founded Y Combinator in 2005."

Glad to see better search tooling for programmers since it's an essential task we do every day. How do you compare yourself to you.com's specialized search engine for developers? https://you.com/code

You.com is too similar to Google imo; we go significantly further than they do in terms of synthesizing explanations. Same goes for snippets; on You.com, you usually need to click on a button (.e.g "Open Side Panel") to get a code snippet which adds friction. Furthermore, they seem to be simply showing the full Stack Overflow page; our approach is to find and rank the most relevant code snippet while offering a "See Reference" button to make it easy to go to the original page.

you.com is a lot more informationally dense, feels quicker to find the right answer, and I rarely have to open a new tab because of the (open side panel) button. Beyond programming, it doesn't seem like I can use hello.so on a regular basis for normal searches compared to Google and you.com.

While I'm not directly affected by this, a blank white page is always a symptom of careless error handling, even in the case where the user has JS turned off. The <noscript> tag exists expressly to present information about your site's need to have JS enabled

We've tested it on the latest version of Safari. Maybe try on an updated browser?

It's totally up to date. From the console:

"TypeError: N.at is not a function. (In 'N.at(-1)', 'N.at' is undefined)"

I searched for 'iife javascript' and although it understood what I was searching for, it's 3 code examples only showed an iife on the 3rd code example. the 2nd one didn't make sense at all [0]

[0] https://beta.sayhello.so/search?q=iife+javascript

Thanks for trying it out - we're still pretty early so the code search feature won't be perfect. The model works best with natural language queries, however, so you might find better results by giving the search engine more to work with.

https://beta.sayhello.so/search?q=Immediately+Invoked+Functi...

On mobile browser I noticed that autocorrect was enabled, which isn't great for technical inputs. I want "hcl", not "hello".

Using latest Chrome gives "Uncaught (in promise) TypeError: Failed to fetch" console error.

The same is with any input, even with predefined ones; the progress bar gets to the end (slowly) and nothing happens.

Firefox works.

Thanks for letting us know. We're under load too so I'm not surprised there are some hiccups. I just tested with Chrome Version 103.0.5060.66 and the fetch works - will have to investigate the latest Chrome version later.

[Y-Combinator] is pushing up daisies now. Too damn slimey for the future. I get that you love your Simoleons, slavs.

It says start typing to search. So i started typing and it didn't search. I really expected it to be some sort of typeahead search without requiring focus on the the search field :)

The odd thing to me was not missing autocomplete, it was that it says "start typing to search", but when you type, nothing happens. This is of course because the search field is not focused when you load the page, until you click it. It should maybe be "type here to search".

That shit drives me absolutely crazy when the one thing a user wants to do is interact with a search field, but the page doesn't choose to put focus on the input

Ah, I see what you mean. We'll look into focusing the search field or changing the text to be less ambiguous.

Maybe someone from the team can help me with a question regarding the privacy page/policy.

>The searches we anonymously log be used to improve our product.

>Your data will not be shared with any third party unless we are required to respond to subpoenas, court orders, or legal process, to establish or exercise our legal rights or defend against legal claims.

>We will never sell your data to any third party.

The first sentence is at odds with the second two.

If you say you are only collecting query and query response data, but then assure me you aren't selling my data, I can't help but wonder which is true:

1) The site is actually collecting data, but is not selling it. 2) The site is not collecting data, and the privacy page is outdated/incorrect/inaccurate.

The same question obviously applies to court orders and the like. Which is it? Do you store data that would be material to me if the site was presented by court order, and that is why you have given me a disclaimer? Or do you not store data, and so the disclaimer is meaningless?

We only collect anonymous queries + explicit feedback provided. No data is material to you. The privacy page was made pretty early on - we'll update it to be more precise. Thanks for pointing this out.

> Training a sequence-to-sequence language model (T5 derivative) on our custom dataset designed for factual generation yielded much better results with less hallucination.

Could you elaborate more on this or point to a paper/benchmark results?

I'd be happy to talk a bit about how we evaluated the model. The task we're performing is fundamentally long-form question answering (LFQA), and recent papers (https://arxiv.org/pdf/2103.06332.pdf) have shown that metrics such as ROUGE (used for the KITE benchmark) aren't great at evaluating the quality & truthfulness of generated answers. On our dataset, our approach is to use a combination of human evaluation (which is still arguably the most reliable metric used by the NLP research community to evaluate generated answer quality) and an entailment score (checking if a generated answer is consistent with a "ground truth" context document).

Good idea. Although it seems it's language querying is pretty poor.

If I specifically list Python/Javascript the first couple results are not even in that language, 3rd/4th are. And you have to click link/see reference to even see the language

You would think if your language is included in the query it should be heavily prioritised

Thanks for the feedback. Could you post what queries you tried?

I'm not a fan of the UI. The concept is great, but imo this doesn't differentiate itself enough for me to stop using Google or even searching via a site like Stack Overflow

Doesn't appear to work in Safari at all (at least, for me here - some JS bundle error).

Same issue, Version 15.1 (17612.2.9.1.20)

    TypeError: N.at is not a function. (In 'N.at(-1)', 'N.at' is undefined)

I searched the walrus operator and all results were unrelated

What was your query? I just searched (https://beta.sayhello.so/search?q=walrus+operator) and got "The walrus operator is a new syntax for assigning variables in the middle of expressions." with one of the code examples being "(walrus := True)"

You.com has a similar code search specific product. What do you plan to offer that they don't?

There are a few key differences. The main one is that our AI generates an explanation to answer your question directly; we go significantly further than they do in terms of synthesizing explanations. Same goes for snippets; on You.com, you usually need to click on a button (.e.g "Open Side Panel") to get a code snippet which adds friction. Furthermore, they seem to be simply showing the full Stack Overflow page; our approach is to find and rank the most relevant code snippet while offering a "See Reference" button to make it easy to go to the original page.

Overall, our goal is to have the highest signal-to-noise ratio of any search engine when it comes to developer searches.

I quite like this. Looking forward to continuing to use it.

Very interesting.

Some of my results with code examples looked awfully similar to GitHub CoPilot output.

Is that being used to generate results sometimes?

Thanks :)

We actually do have a code generation model similar to CoPilot but it's not active yet on the backend. All of the code snippets you see are pulled from other websites.

Nice project. FYI Search Returns empty page on UC browser mobile.

Would this be useful as a vscode plugin perhaps? I would much rather search this on an IDE imo but I am just one random dude.

We've actually had a bit of feedback related to IDE integration. We're focusing on developing the search technology itself right now but vscode plugin could be interesting to look at in the future.

How do you think your product will fare in the wake of the backlash and legal saber rattling against GitHub Copilot?

cool! some of the code samples shown aren't always the most relevant, but looks promising.

The answer never loads if using Brave with Shields up, even after allowing JS.

Just installed Brave and it seems to work even with Shields up. The site is under load so maybe try again. Are any of the other things (answer/explanation, code snippet, links) loading for you?

I press search, get a 10sec progress bar loading and nothing happens.

Congrats for the launching.

Anyway, just want to comment the product niche. Vertical search engine? Interesting. Will we see another vertical for a search engine product?

> center a div

#myDiv{ margin:0px auto; }

We’re still very early - of course it can't be as good. Could you tell me what you were trying to do and how it didn't work for you?

Not sure what the original commenter was looking for but I can give my thoughts:

- stackoverflow's UI actually serves well to provide a sort of "ambient" information that rapidly indicates not just the best answers, but the best most-recent answers. Oftentimes, and especially in rapidly-evolving dev languages/frameworks, what was the best answer a few months ago may no longer be the best answer and the ability to rapidly scan the comments that would indicate this is valuable. - in addition those stackoverflow comments and links within them can point to additional info that can save the dev time (potentially pointing to the dev misidentifying the problem: "don't do this, this is the real issue <link>).

I think with the traditional google->stackoverflow or google->[some documentation site, forum, etc] user flow you actually get layers of ambient cues as to relevance, recency and quality that we've grown accustom to. Even if your product ultimately serves better answers I'd worry that lacking these cues would make a user like me feel as though I'm blindly trusting an answer that seems to have come from the ether (sort of like github copilot).

As low-hanging fruit maybe adding level-meters beside each result that indicates these dimensions could help (like npmjs.com does with npm pkg results in their ui).

I love the product idea and it looks like a strong start! Good luck!

I agree that software documentation is constantly changing and a mechanism for evaluating the "freshness" of an answer could be very useful. Right now we provide an easy way to find the source of a code snippet but source attribution/evaluation is something we're actively looking into, especially for the natural language answer. A quantitative score could be interesting too. Thanks for the feedback!

it just won't be as good as the refinement in searches i can do with appending stackoverflow at the end of a google query and github copilot already does what you are trying to do