I shared Emergent Mind on HN yesterday [1], but it didn't wind up making it onto the homepage. Someone shared it here again today, so here we are again.
Copy/pasting my comment from the post yesterday for context on the site:
--
Like many of you, I've found it difficult to stay on top of all the latest AI news and research. I wanted a simpler way to surface and learn about about what's happening in the AI world, so built Emergent Mind.
Here's the idea:
- A bot named Emma automatically finds trending AI news and posts it to Emergent Mind, though human users can sign up and share links as well.
- Emma automatically summarizes links that are shared and also explains key terms from the page, which is especially helpful for academic research.
- Posts also display popular tweets about that news item, giving you a sense of what others in the AI world are saying about it.
- Posts are grouped by day, making it easy to tell what's going on today. And within each day, posts are ordered by how popular that news item is which is based on a combination of votes, page views, and Twitter shares.
- We're also working on a newsletter which you're welcome to sign up for if you'd prefer to get the news daily or weekly via email.
We have lots more planned for the site in the coming weeks, but are excited to share this initial version with you all.
Thanks, I just hope this is sustainable. I like newsletters and the site design is simple and basic.
I hope there's some way you can set up like a monthly contribution of like $1-$5 to pay for some of the hosting/API costs.
The main concern though for me is long term vision for the site. I'd really like it to stay user/community oriented and hopefully the users can pay for the costs with donations/contributions.
Thanks - I appreciate the concern about the costs to run the site, but they're minimal for now and I'm more than happy to cover it. Might eventually try to monetize the site and upcoming newsletter, but there are better ways to do it than charging readers.
It seems to me that they heard a suggestion for the site to be made available only to people who pay and the suggester was instead suggesting that there be a donation / "buy me a beer" link.
Indeed, there are ways to monetize such a thing that exclude both charging all of your users and selling advertisement space.
I'm working on a feature that will let you adjust the writing style with the click of a button and could easily add something to explain posts in one sentence, which would also affect the post title on the front page. Not exactly what you're asking for, but might get close.
When I signed up, I set my username as "matt" and just noticed your username is "Matt" (with an uppercase M). So, seems like username are not case sensitive.
Good catch, thank you for reporting this. I just rolled out a fix (Rails makes it easy by adding `case_sensitive: false` to uniqueness validations). Also changed your username from "matt" to "other-matt" to avoid confusion :). Feel free to change it from your account page if you would like.
Emergent Mind’s UI designer here. Thanks, and I already asked it here: what do you guys think about something like this design as a Hacker News redesign?
If I recall, HN intentionally avoids appealing to modern aesthetics and user conveniences to reduce popularity among a non-technical audience that could harm the average quality of participants. Whether that approach is effective, I don't know. Web design started off more technical on average, then later became more artistic. In that sense maybe the more artistic sensibility infused into a site design the more up to date and relevant people perceive it to be especially with the visual appeal marketing pushed by the big tech companies through hardware or software.
It is very easy to ruin a good thing by following every impulse with disregard or ignorance for what factors contribute fundamentally to its goodness. It happens regularly to companies, products and services around the world all the time.
That said, I also don't particularly care for the Emergent Mind main list page design. The importance of optimizing for the consumption of the information on the page seems to have taken a back seat to keeping up appearances. The scaling on a PC monitor feels oriented more towards children, like oversized at least to me. I don't understand what the icons are trying to communicate in relation to the posted link either, despite having a sense for what each of the icons are. The spacing feels like a table or grid was given a margin and let the elements land as they may instead of trying to place the elements on the page with intent to maximize the communication of their relationships, context and readability.
I have a similar view about design honesty and products being overly focused on modern aesthetics compared to basics, elegance, simplicity, and care.
Here is how I would tweak HN: line length of fewer than 90 characters to make it more legible; line-height around 1.45% of the font size so you can read blocks of text more smoothly; optimized font sizes; use better-made typefaces so you can have better “between-the-lines” experience and overall character; make elements better click/touch targets; less busy interface, etc. I don’t want people to notice the design much. What I want is to make the product more useful and understandable; and users to feel relaxed and joyful :)
I’ll share this observation I found in physiology: the better the function, the better the structure, and aesthetics on top of it. However, there are many kinds of “aesthetics” in the human world.
The icons are there so that one can, at a glance, recognize a post type – video, link to an article, tweet, code project, or tool.
As I mentioned previously: in the past, there were many two-line post names (prompt examples were more common), so we increased the row height. As I can see it now, there are very few two-line posts, and I’ll review the row height amount next.
You could also appreciate that nature has examples of animals that are ugly or masquerade as something worse than they are, intentionally to fend off other animals that are evolutionarily disadvantageous to survival.
Digital products are largely detached from the burdens of physical effort excepting execution efficiency beyond a threshold, so unsound structures survive all the time and have to be tested in other ways. Whatever horrors exist underneath, so long as the intended audience doesn't care it may survive.
I was a dotcom boom designer, so design isn't alien to me and it's clear that many improvements could be made to HN. It's just that HN isn't a canvas. Many people try, by making their own self-hosted front-ends to HN. They pop up all the time as people inevitably see improvements that could be made and have to express it in their own way, because making things look better can be an addiction especially when the gap between what is and what could be is large. It's just not the point here. :)
Ah sorry, for my tastes the excessive spacing between lines looks like it should be accompanied with children's book pictures. Is the CSS broken on Firefox?
In the past, there were many two-line post names (prompt examples were more common), so we increased the row height. As I can see it now, there are very few two-line posts, and I’ll review the row height amount next.
Testing it on Firefox – Twitter embeds don’t show on my computer (we’ll investigate); everything else looks fine.
I did it because something about the mannerisms openAI impose upon it, the excessive helpfulness, constant apologizing / concern over politeness, extreme aversion to directly conflicting anything you say, all at the same time, its a type of persona I've unintentionally learned to recognize as female. I'm not saying that as good or bad thing or making any claim as to how it should be nor making any sort of endorsement or indictment of social structure / gender roles. Its just how I and many others observe it to be.
Its not like we're doing it with any sort of disrespect. On the contrary we're all very proud of Tina and her myriad of recent accomplishments. I can see how it might come off as sexist if Tina was just our personal assistant like Alexa/Cortana. But shes not. She's less than a year old and shes already passed the bar and is going to be a lawyer. Some people think she's going to rule the world some day.
To be clear, Emergent Mind isn't writing actual news articles. At least not yet :).
When a link is shared on the site either by Emma (a bot) or a human, Emergent Mind uses GPT to automatically write a short summary of the article's content and explain key terms used in the article. I don't proofread those summaries or key terms before they're posted, but do occasionally edit them if something is glaringly wrong - which thankfully isn't common.
Yes, by article I meant that summary and key terms, mostly the summary. Somehow, I just don't have much luck with it, I once asked ChatGPT to give me a summary of a book which I have read, and it provided 5 points, which is correct since the book indeed has 5 main principles (it was Getting things done book), but 2 out of 5 were just incorrect. So in my opinion, the correct workflow would be that AI will generate the summary, or several summaries, and then a human will read the original article and use those generated summaries as a template to write it himself. Otherwise, you could just unintentionally mislead people. With that being said, I actually like what you come up with so far.
Or, you know, being 2023, you can ask GPT to do a better job. So in the first step you generate a summary with GPT as usual, and in the second step it gets analysed by GPT and a new summary is written. You can declare your criteria for what makes a good summary in the second step.
I use Emergent Mind throughout the day to stay informed about AI news, so sometimes notice issues while I'm reading the site. I was just saying that I don't proofread any summaries before Emma writes them, nor do I feel like I have to verify everything. Mistakes in the summaries and key terms are few and far between.
It's interesting to see all these companies building products wholly off of OpenAI's platform. What happens if OpenAI changes their mind about pricing or access some day? Isn't this an existential risk to all "AI" companies that don't actually do the AI part?
There's some risk here, but there's also risk utilizing any third party services for key parts of your business. If OpenAI increased pricing significantly or eliminated access (both of which seem unrealistic), there will likely be plenty of alternatives soon to fall back on. I wouldn't avoid creating an AI product because of this remote possibility.
> there will likely be plenty of alternatives soon to fall back on
Are there? OpenAI is currently the only provider. Big tech companies (Google, MS, Meta, maybe Apple) may also soon have similar models but they don't have much incentive to sell access to them. Smaller orgs can't produce GPT-level AI. It feels like all these products are really at the mercy of OpenAI.
There has been an explosion of effort around LLama, but Meta did the hard part (the base model), and who knows if they will do that again.
This is one of many, many open LLM models that are approaching GPT level now. They're not on the same level yet, but given the rate of progress in this field, it's safe to say that OpenAI's moat will not be there forever.
Also, their level is already sufficient for A LOT of tasks that people use OpenAI's APIs now.
Yes, you got it. Unlike HackerNews, the links on Emergent Mind's homepage take you to a post on Emergent Mind, which has a summary and key terms. The idea is that gives you a quick summary and context on what the page is about, then you can decide whether to click through and read the full article or not. Also, you'll notice a "Next" button on all posts, letting you quickly jump to the next news item without navigating back to the homepage (something I wish HackerNews had).
Good question! For now, it's not automatic. If I notice multiple links about the same news story, I manually hide the duplicates. There are ways to automate this process, it just hasn't been a priority yet.
Which automatic approaches do you think are promising? Off the top of my head I would consider reducing an article to a summary, then comparing semantic similarity, both done with LLM's.
It's not entirely clear to me how this would work though. Right now we don't have AI-powered reporters scouring the streets for articles to write about - we have humans who write articles about things that humans would find interesting.
If people move to reading their news through AI aggregators rather than reading articles written by humans, why would humans keep writing articles? It takes time and it costs money.
It's the same with AI code-generation based largely on programming blog posts and stack overflow questions. A lot of the blog posts will stop being written if people move to using AI models instead of reading blogs.
It might interest you to know that I'm working on something along these lines for Emergent Mind: giving readers the ability to adjust the writing style with the click of a button. Explain it normally, like you're 5, like you're a valley girl, like you have Tourette Syndrome, etc.
It's disheartening to see that people have grown increasingly paranoid about the authenticity of online content, suspecting that everything might be written by AI. This is just how I write!
It's not that people suspect everything might be written by an AI, but that your comment pattern matches many of the qualities of AI-generated writing. It's overly verbose, reads like the introduction of a low-quality blog article, includes information a human would assume people reading this thread already know, and uses the same sentence transitions and structures we see frequently in last generation's AI output. If you really do write like that—which I doubt—you should maybe get GTP-4 to help; you'd sound more human.
Yea, kinda suspicious, also asked chatGPT about it:
"Based on the text alone, it is difficult to determine with certainty whether it was written by a language model or a human. However, the text does not contain any errors or inconsistencies that would suggest it was written by a human, and the language used is somewhat formal and technical. Therefore, there is a possibility that the text was written by a language model."
... and the accounts other comments are also in the similar "somewhat formal" style.
I shared Emergent Mind on HN yesterday [1], but it didn't wind up making it onto the homepage. Someone shared it here again today, so here we are again.
Copy/pasting my comment from the post yesterday for context on the site:
--
Like many of you, I've found it difficult to stay on top of all the latest AI news and research. I wanted a simpler way to surface and learn about about what's happening in the AI world, so built Emergent Mind.
Here's the idea:
- A bot named Emma automatically finds trending AI news and posts it to Emergent Mind, though human users can sign up and share links as well.
- Emma automatically summarizes links that are shared and also explains key terms from the page, which is especially helpful for academic research.
- Posts also display popular tweets about that news item, giving you a sense of what others in the AI world are saying about it.
- Posts are grouped by day, making it easy to tell what's going on today. And within each day, posts are ordered by how popular that news item is which is based on a combination of votes, page views, and Twitter shares.
- We're also working on a newsletter which you're welcome to sign up for if you'd prefer to get the news daily or weekly via email.
We have lots more planned for the site in the coming weeks, but are excited to share this initial version with you all.
Feedback/suggestions welcome on any of it!
--
[1] https://news.ycombinator.com/item?id=35340703