sethkim

I have a really hard time understanding how this, and similar products like Scale Donovan make sense in a defense capacity. Even more so than previous AI waves, LLM's are highly non-deterministic and are way more suited to low-risk use cases (for now). I understand the appeal of giving more data-driven capabilities to non-analytical users, but when it risks returning hallucinated results that could lead to actual human casualties - that quickly crosses a line.

I have a ton of respect for both Palantir and Scale as companies - but this just looks like a cash grab race to see who can dupe the Pentagon the fastest into believing they need some technology that's years away from anything they should be using in practice.

It seems like Hacker News might be broken, since all of the top posts have like one or two points. Anyway, seems like a good opportunity to help someone out!

Everyone has their own value function in life, and it will change over time. My advice is generally to do what feels important. I wrote about this recently: https://sethkim.me/l/thesolutionspace/?postnum=7.

If the internship opportunity seems incredible, it probably feels important to do, right? And I'd imagine if the pay is really good, you'll be able to travel home a handful of times over the summer and spend time with family/friends. But if spending the whole summer with them feels more important than the internship, it probably is. It's your question to ask yourself, and answer honestly.

On the question of growing up too fast: you will most likely have to get a start on the career you think you want at some point, so the internship sounds like a smart move. I say "you think", because that's likely to change as well (sorry to say). Internships are a great way to dip your toe in the water of a number of career paths to start to figure that out.

Life is a great balancing act of doing what you have to do, should do, and want to do. Balance wisely, but don't forget to have fun along the way!

There's been a new media for music every ten years for the past fifty or so. Records, cassette's, CD's, mp3's, streaming. I've been doing the thought exercise with some of my more creative and/or musically oriented friends recently to try to determine what comes next (if anything).

The only conclusion we're able to come to is some kind of procedurally-generated music that is more specifically catered to each listener. I'm not surprised this wave is starting to break and it's a great time for artists and legal teams to figure out how to work with this technology if there's any possible way to do so.

You can feel free to reach out on one one of the email addresses listed on the website.

You could assume zero, and start from there! People may be using it; I'm not sure. I'll occasionally check out the site and there are roundtables that have been started.

I usually refer to this distinction as "tall" vs. "wide" problems. Tall problems require having at least one exceptional or specialized person to solve it, whereas wide problems can be cleared by most generalist problem-solvers. The time required to solve wide problems is simply one of aggregate bandwidth.

Unfortunately, I think society has a ton of people who can solve super tall problems working on super wide problems, because the largest (and most well-funded) organizations have the widest problems. In some cases they have super tall problems too, but past a certain size the lights stay on by solving the wide problems.

I actually built something in this space in the last two years called roundtable.audio (https://roundtable.audio/). I actually also created an HN extension found here: https://hackernews.roundtable.audio/. See this post: https://news.ycombinator.com/item?id=26856883.

I've thought about it quite bit, and I think one of the long-term strategies to revive the space is having live-audio conversations attached to specific news articles, tweets, and other places on the web which prompt discussion. I also considered creating a chrome extension at one point which allows someone visiting an arbitrary webpage to be able to audio chat with those who are also on the same page.

I'm planning on shutting down the deployment for roundtable.audio in the coming months, so if anyone wants to take these projects and run with them, I'd be happy to chat about how to collaborate or hand things over. The repo is open-source as well, but I'd certainly like to hear what you plan on doing with it or how I can help!

Contrary | Sr. Product Engineer (Full-Time) | San Francisco, CA or Remote

Contrary is a venture capital firm with a strong focus on talent and technology. Our internal engineering, product, data science, and design team builds tools to provide superhuman leverage to our investing team, productize our value to all of our stakeholders, and relentlessly support and cultivate the next generation of top talent.

We're looking for 1 - 2 exceptional full-stack, product-minded engineers to join our team, and take ownership of one or more of our core products (to be discussed in first call). We're tech-stack agnostic (you can build in the languages and frameworks you're comfortable with).

Email hiring [at] contrary [dot] com for more info.

Would be curious if anyone has run https://www.mightyapp.com/ on a Pi, had success, and saw the combo as a suitable thin-client replacement for a bigger machine.

I totally agree with this - but with my current understanding of WebRTC, there is no "efficient" way of decentralizing a multi-party real-time app (1:1 is only so useful). It requires an SFU or MCU architecture.

Finding an efficient way to chain peers without a centralized server would be a seriously transformative piece of engineering (hence the title).

Are you sure you're looking at the right Twitter account?

Thanks for this comment. I think there's an important distinction that's worth clarifying.

Automation is great in most circumstances, and truly does free humans to do more meaningful work. But my take is that you can quickly push up against cases where it's unclear that you've improved a situation. It's easy to make the judgement that you have when you reduce time spent on doing "mundane" work.

Perhaps there is a heuristic that can be developed to let someone quickly assess the usefulness of automating a certain task. Would be happy to discuss this more.

This is interesting. You are the second person on this thread to think that Mantis outsources calls to someone else (it does not).

What on the website makes you think that? Important for me to know so I can clarify it to visitors.

Yes, exactly right.

I'm curious what would indicate otherwise...

I like the idea of "abtracting" a human with automated processes. I think you're right that a semi-automated chatbot is exactly that. The thing to be super careful about in such cases is whether the automated component is purely beneficial, and be rigorous with data to ensure it is.

Glad to hear!

A lot of my outbounding efforts with Mantis have fallen flat when I sent something that was more or less templated, but have had better results with something far more personalized.

Here's a fantastic article that might be worth a read: https://cloutly.com/blog/cold-email-template/.

Can't say I didn't take some inspiration from that idea!

Awesome story. Gives me a lot of confidence in the value that Mantis provides.

There will always be people who don't need or want to speak to a human, but having the outlet to do so when it's useful to the customer can make a huge difference.

Yup. It's natural human tendency to reduce mundane work. But it can be illuminating how much you can improve results by stepping in and getting your hands dirty first.

Hey! Good question!

No, I'm certainly not available to take calls 24/7 right now. For a smaller scale business, it's certainly seems reasonable to only take calls during "normal business hours." That being said, during those hours when businesses are normally responsive to a live chat feature, Mantis is meant to provide a higher throughput medium to answer questions and solve issues faster than someone can via text.

Hey diskzero - thanks for adding the questions here!

- Q: I would like to set up office hours where I can guarantee that I will be available to take calls. - A: Right now, you might have to publish on your website the time of such office hours. If this is something you and others would find useful, we can add a configuration option to allow you to state when you're typically online.

- Q: What if I am on a long call and several people call in? - A: We allow "concurrent" agents at the moment, meaning you can have multiple people answering calls so people don't have to wait in the queue. Another feature I'm thinking of adding in is a way to set a time limit per call. That is, when someone calls in and you answer, it will display a timer (say 3 minutes) that the user has. This seems like a not so rude way of preventing individuals from hogging all of your time.

- Q: Is there a way to transition from the call to a screen share if I have a way to show a customer how to solve their problem? - A: There is currently an "interactive canvas" feature where you can use a WYSIWYG editor to send live visual aids to the user. I think a screen-share feature is also very sensible and something I'll likely put on the feature roadmap.

tl;dr: Mantis is a live-audio chat widget which can be embedded into any website with two lines of code. It's primarily intended for sales and support, and can be used to quickly answer site visitor questions and learn more about user needs. This will hopefully generate more leads and conversions, and give your customers the painless, white-glove support experience they desire. Come give it a try - I'll be around the rest of the day taking calls!

More details:

I'm Seth, the creator of Mantis. In the spring, I (re-)launched roundtable.audio here (https://news.ycombinator.com/item?id=26856883), and decided shortly after to pivot to solve a more immediate business problem.

If you're anything like me, you hate live chat solutions as they exist currently, especially when automated. When possible, I always prefer to pick up the phone to have my questions answered quickly by a real human before making a buying decision. It's hard to believe that in 2021 the best we can do is put 10 digits on a website when we want to enable audio-based support.

As I was building this out, I realized it can also be a powerful tool outside of the typical sales and support contexts. Everyone knows it's important to talk to your customers when launching a new product or startup. As a founder, it's still possible to have these customer conversations yourself (doing things that don't scale - the YC gospel). Mantis can offer a customer interview hotline directly to your site. I plan to use Mantis to iterate on Mantis, so perhaps it can be useful to your tech-oriented product too. This re-surfacing of an HN post last week captures some sentiment around this idea (https://news.ycombinator.com/item?id=28594313).

The core product is ready to go, and I'll be rolling it out slowly to start. If you're interested in trying it, add yourself to the waitlist on the landing page or email now@mantis.chat with your intended use case. There's so much cool stuff to build on top of this, and I'm excited to share more as I progress.

Come by and have a chat with me by clicking the black button in the bottom-right corner and clicking "Join Audio Call". I'm excited to hear what you think!

Honestly, the browser/platform test coverage isn't broad yet. Not too sure why Firefox is having difficult but I know some browser and platform combos are quite finicky with WebRTC.

Which link are you referring to?

Neat idea!

Sort of - this specific extension "adds" live audio to HN, but broadly speaking it's meant to enable live audio to otherwise text-based sites.

Thanks for pointing this out. I think there is a possibility for both preserved, text-based comments to exist alongside live, audio-based discussions.

I wouldn't want the HN experience being degraded by not being able to access relevant discussions about posts in the future, but I think those who are deeply interested could benefit from the ability to speak live about posts with others (including the OP) in a more intimate and fast-paced setting.

The roundtable for this post is a great example. There have been some super valuable comments in this thread as well as great discussions held in the audio chat room (except for the rick-roll, that is!).

Interesting comment - someone else pointed that out too. Could be a good feature to explore.

WebRTC only works on certain browsers and platforms.

Something like that!