jusquan

Agree that ComfyUI Manager is super helpful for custom nodes, Yup, haven't found anything that'd work out of the box yet. It'd be helpful to even just have a marketplace where you can bring your own API and sell calls from the platform, abstracting away the user management and billing. I was able to put some of my ComfyUI workflows behind an API so that it runs on 'serverless' GPUs, but am still stuck with the rest of the stack to build from the ground up.

I realized that a lot of companies have some sort of annual learning stipend/continuing education benefit, but tons of folks (including me a few times) wind up leaving $$$ on the table by spending it on nothing, letting it expire unused by EOY.

Over the past few months, I've asked my friends about their favorite purchases they've managed to get approved and built stipends.fyi to start to compile them [still a WIP].

Let me know if you have any other helpful (or questionable) purchases you've managed to expense and I'll add them—hope you find the list useful! Even if you don't find anything interesting on the list, hopefully this reminds you to expense a book or class you've been eyeing.

Note: most of the links are affiliates, which give me a few % commission if you wind up using them, half of which I donate to St. Jude Children's Research Hospital and the other half of which I selfishly use to buy more books for myself :)

This is great, thank you for sharing! I wonder what the reverse would look like. More and more nowadays, I find myself first looking on YouTube for tutorials and walkthroughs, even if they wind up being more verbose than their written counterparts.

This is really cool! Would love to see a version for Pokemon and Yugioh to revive my childhood. Because the relative consistency of card formats, AI-generated trading cards seems like a great use-case, with some gaps depending on the type of card. For Yugioh, I imagine generating believable images is easier since the card art is extremely varied, but it's probably a bit harder to come up with meaningfully complex card effects/descriptions. It's probably the opposite for Pokemon, where there's a much smaller set of known Pokemon so AI-generated ones are easy to detect, but the moves that each one possesses are usually pretty simple (unless that card game has gotten way more complex from what I remember)

There’s a baklava place in SF called Baklava Story that is hands down the best baklava I have ever had and probably will ever will have. The owner, Tolgay, travels to Turkey each year to source pistachios and milk to make butter with from specific farms each year. With the residual milk from making clarified butter from the Turkish milk, he makes soap, and gives it to customers.

Every single one of his reviews on Google and yelp are 5 stars. I deeply admire folks that invest so much of their time into one craft (especially if it happens to be edible) to become the best

Haha thank you! Strongly agree—a characteristic part of HN are definitely the opinionated and inevitably contrarian replies, so in the future, we want to create opinionated fakeHN agents mapping to the past comments of real individual HN users. That way, we can hopefully create higher-fidelity comments capturing their biases and technical depth to capture more of the magic of HN.

Those are great suggestions! I especially like having an 'I'm feeling uninspired' button to help fix the blank canvas problem + make it easy to test with real HN posts. Appreciate the feedback!

Ah yup this is something that happens every 1 in probably every 20 or so times. Planning on adding a few layers of parsing after generations to prevent this from happening!

Nice! Thanks for trying it and glad you found it interesting :)

kindernetes is overkill, at that scale you can run a single dayckare just fine without child orchestration

Thanks for flagging, would you mind trying again and letting us know if it's still happening? We just pushed some fixes

Glad you enjoyed it, and thanks for trying it out! :)

@OpenAI you know where to find me, let's partner up ;)

Haha the flagged comment is an easter egg at the moment, so unfortunately (or maybe fortunately) there isn't a guideline-violating comment. Though, as we create higher-fidelity comments mapping to real HN user comment histories, it would be really cool to have LLMs do real moderation on top of them! Definitely something we want to try.

dang vigilantly roams all variants of HN :)

Thank you! Yup it'll get confused for super small inputs, generally the more verbose of a post you're able to give it, the better the results are.

I could see how FAQs would be helpful, or even helping posters predict what users might be confused by or curious about to add more context to the initial post!

I'm scrambling to write a check to these companies

https://www.fakehn.com/submission?id=xC022pZ1bbGcHGqPExSS

Thanks for flagging, this should be fixed now!

I think I got this exact comment in one of my test generations for this submission

JK, depending on what you're submitting I agree the comments are very shallow right now for many submissions, which is why we'd like to create higher fidelity comments by mapping generated comments to real, individual, opinionated HN readers and their comment histories. Anything we can do to bring more of the real HN magic to fakeHN!

Yikes, thanks for letting us know! We were getting rate-limited by OpenAI, especially after hitting the front page(!), so sometimes the initial comments would load, but the replies wouldn't, killing the page.

We just fixed this so that even if replies fail, it'll still show you the initial comments. The replies should also be less-likely to fail now, in case you give it another go!

Where can I invest in this

Let me know how it goes! It's fun to compare the comments to real ones for the same post. Some definitely miss the mark but others, especially some of the Ask HN posts, have surprising similarities.

fakedang but 1 in every 10 posts is realdang

Thanks for sharing! This is something we've noticed too. One of the key traits about any social forum like HN is that there are more concrete opinions (with some inevitable contrarian ones), which we aren't seeing very often now. We're hoping that as we create higher-fidelity comments by mapping simulated user comments to real, opinionated, biased HN user comment histories, we can see more of this!

someone's gotta take the fall and leave a comment on this post violating the guidelines (jk please don't)

what's the link to the library, I need this ASAP

Thanks for trying it out! You're right—one thing you can almost always count on in any sort of shared forum are the inevitable contrarians, and it's one of the magical parts of social simulation we're hoping to capture more of, hopefully by mapping fakeHN's simulated users to this history of real, opinionated, and biased HN commenters.

Nice! > AI-generated comments might not capture the nuances and insights of real HN users. True! > Have you considered incorporating real comments into the system to improve its accuracy? This is definitely something we want to do, creating agents mapped to real users, to get closer to above.

I've definitely felt that the karma system is a bit intimidating for newcomers, but I also get how it serves as a signaling system (though imperfect) against folks with bad intentions

Haha that's just randomly generated easter-egg! Though it'd be really interesting to have real AI-generated moderation on top of any inappropriately generated comments

Haha thanks for sharing! I wonder how these responses compare to how real HN users would respond...who knows, we might have some undercover pageant-contestants spending their off-time debating about React libraries