Hacker News new | past | comments | ask | show | jobs | submit login
Launch HN: Type (YC W23) – AI-powered document editor
194 points by stewfortier on April 4, 2023 | hide | past | favorite | 231 comments
Hi HN, we're Stew and Stefan from Type (https://type.ai/). We're building an AI-first document editor that helps you write. It's similar to Notion, but focused on building a solid authoring experience.

Here’s a general demo: https://www.youtube.com/watch?v=MpK9PWo0lUw

And here’s a demo that includes math and code blocks: https://type.ai/code-math-demos

There are a lot of AI writing products out now, but we've found that most of them treat writing like a one-shot activity that should be delegated to AI. We don't think that's the optimal way to write. We think of great writing as the product of clear thinking, which requires a lot of time tinkering with and refining ideas. So we’re building a user-friendly document editor that puts the author front and center.

As you write in Type, you can press cmd+k to summon simple AI commands. Most of our commands are grounded in familiar writing primitives (ex. “Write paragraph”) and attempt to understand the context of your document.

Type supports multiple rich block types, including code and math and our commands are able to both interpret and output these block types. So if you're writing an introductory essay about machine learning, for example, you can use Type's chat feature to generate and refine equations and code blocks you'd like to include in your document. Once you’re satisfied with what Type has generated, you can drag and insert the block anywhere in your document (as seen in the demo video above).

We’ve also built a "what to write about next" feature in the document sidebar that offers suggestions on ideas you may consider adding to your document.

We’ve built some editor features that aren’t AI-specific but which we think make for an enjoyable authoring experience: (1) Type is built from the ground up to be offline-first. This means most interactions (search, loading documents, etc.) are instant; (2) Mobile support as installable PWA; (3) Keyboard shortcuts for the most useful commands; (4) Markdown copy/paste support.

We designed Type to be most useful for longer-form writing, so we encourage you to try it out in the context of something like an essay or a technical tutorial. If you try it out at https://type.ai, we’d love to hear what you think. We think Type feels pretty different from other AI writing tools that produce fairly shallow content, but would love to get your honest feedback on whether we're hitting the mark.

Each account comes with a free allocation of AI commands, after which you can activate a paid plan for unlimited AI usage (you can still create and access unlimited docs on the free plan). If you'd like some additional free credits, please just drop us a note at founders@type.ai and we'll refill your free credits.

We'd love your feedback on what feels helpful and what feels confusing or missing. Thanks!




First of all, good luck and the implementation looks pretty slick.

I relate to the other comments that (1) Explosion of AI generated blogs / copy etc is not what we need and it's hard to see the value of it long term (2) this looks like a simple usage on top of GPT4, no real IP / innovation - this is risky from a business model perspective.

Good luck!


I'm _extremely_ surprised to see all of the YC companies that are essentially just layers on OpenAI.

I think it's safe to assume more models will be introduced and vendor lock-in can be avoided, but I find it hard to believe some of the "simpler" ideas can create compelling, VC-scale businesses.


I think it’s both correct and realistic to assume that most ideas won’t reach venture scale size. After all, the vast majority of startups fail. However I’d like to point out that any software product that is built is per definition a layer on top of something else and it will always start small. The key question is whether that “layer” is useful and if it can keep getting increasingly useful with time.


While integrating with OpenAI is just an API call away I don't think it takes away from the user experience. Sure its less of a moat but plenty of products have won by having the best user experience and not many new features (Notion, Digital Ocean, etc.)

Not every experience is going to be best suited to fit in a chat box like ChatGPT which opens the door for startups like this one to build something new.

I'm excited when I see products like this, and I think we really need to retire the critique of "Its just an API call to OpenAI". While yes that is a core part of it, there is a lot of time and effort that goes into developing these experiences that has value


I suppose it's inevitable that the first uses are going to be the trivial ones. They'll be first to market.

I'm sure they imagine that when they figure out something actually worth doing, they'll already have a user base and revenue stream and reputation. If there's somebody out there doing something more innovative, they'll either buy them out or reproduce their idea in-house.

Me, I'm skeptical that there's a "there" here. But that's why somebody else is getting rich and I'm not.


Why are you surprised? OpenAI is Sam Altman's (YC's last leader) pet project. Of course, YC is highly incentivized to produce more customers for OpenAI


Incentivized how, specifically?


Well there's this thing called friendship. You see, executives at VC firms and such are typically pretty friendly with each other, and view their fellow executive's success as potentially pulling them up the ladder. AI is the next big thing. Sam Altman and his company is in a nice spot to profit. Moreover, YC research is an investor in OpenAI.

Thus, because the executives want Altman to succeed (to presumably advance their own career) and because they're in charge of YC, which is an investor in openai, they have every incentive in funding startups that then use OpenAI as the main platform.


This hypothetical incentive structure seems tenuous at best and not the open and shut case it sounded like in your original comment. I’m sorry but I don’t buy it.


You don't buy that people want their investments to succeed and their friends to become rich? Those seem like some of the strongest incentives to me.


So the YC Partners are actively investing in worse companies that will degrade their returns so that OpenAI/Microsoft can add new customers that do almost no volume compared to ChatGPT or Bing?

That's not how YC or VC works.


There's more to businesses than simply the underlying tech. If technical superiority were the sole determinant of business success, the tech landscape would look a lot different today.

Rather, YC is banking that some of the use cases get people hooked, thus causing vendor lock in, to these new apps, which end up benefiting YC both by benefitting the companies they're funding as well as benefitting open ai.

I don't share the sentiment that companies that are 'just' layers over OpenAI are incapable of building moats.


Thanks! I do agree with #1, and I think we need to make that clearer in our messaging. For context, I'll re-share one my replies to a related comment:

> The thing that excites me most about generative AI isn't "more," it's better. I often use Type to write satire and now whenever I hit a block, I don't tab over to Twitter – I have Type generate some ideas. Often, I don't use them as-is but they do inspire a new angle I hadn't thought of.

One #2, I think there's some truth to that today. But our belief is that over time, these products will start to evolve into something more advanced and useful. A product like Type, for example, won't really look like Google Docs + AI in 3 years, it will start to feel like a more novel category of tool. We'll see, though!


How will it look in 3 years then, and why doesn’t it look this way today?


Because we're still figuring it out and it takes a while to build new stuff (not a great answer, I know - but it's the honest one)


Sounds like vaporware.


That's not what vaporware is. A product that just released on a technology that has only been available for a few months isn't vaporware simply because it doesn't already achieve the full roadmap goals


What is the roadmap? The author’s response doesn’t say anything about what plans they have.


And the "roadmap" is for paying customers only.


In 3 years it will also come complete with fake images, diagrams, photos, sound bites and video clips.

We'll all be locked in to walled gardens (figuratively) since anything out on the wild/open/unverified internet will just be fake. This is what could rescue traditional media if they play their cards right. Providing genuine content by authenticated and verified-human authors.


"fake thoughts"

"fake emotions"

"fake ideas"

"fake attractions"

"fake memes"

"fake dreams"

I imagine the Luddites were themselves angry about fake work and fake souls. We did okay though.

Humans will cope [1]. We are resilient as fuck. We once had to fend off lions and bacterial infections and getting throttled in the night. Now we worry about lattes, stock prices, and political hullabaloo. We'll be fine.

[1] (Not just cope. I'm willing to bet that it'll be better than everything that came before.)


It’s bizarre that we have to talk about coping in the context of technology. As opposed to how it’ll make lives better.


Luddites were executed or deported [0] though.

[0] https://en.wikipedia.org/wiki/Luddite


>This is what could rescue traditional media if they play their cards right.

Hahaha haha no

Their hallucinations are worse than ChatGPT's


It looks like just a feature of Word/Ai Writer/etc indeed.


> this looks like a simple usage on top of GPT4, no real IP / innovation

You can generate the first draft in the chatGPT-4 window, then ask it to add a section or expand a list and it works pretty well in the chat interface. So it's not really hard to write articles with bare chatGPT UI.


It is marginally better than ChatGPT, though, for writing.

I would happily pay a token or credit-based fee to use this while in heavy writing mode.

A MRR or ARR license for this? hmmmm... it would need to be multi-modal with dynamic page layouts and images to justify a subscription (for me).

Or auto-update a doc in the background via async AI.


To each their own :). I do think it's a big space and plenty of folks will find ChatGPT helpful enough to not warrant a more vertical / specific solution like ours.


Explosion of blogs and cheap copy was already rampant, but it's true that it is magnified. On the other hand, do you consider niche sort of content[0] a net negative, when these tools let fewer writers or writers with fewer hours create content for, perhaps untapped, niches? The newsletter linked below is not something I would have been able to do or justify doing before ChatGPT. Can niche usecases make up for the explosion of samey stuff?

0: https://chinesememe.substack.com/p/learning-a-chinese-song-w...


1. *Not that special* Honestly, this isn't too impressive. It's just a pump that sends a request to GPT-4 and gets a response. The buffer isn't anything extraordinary.

2. *Sectional content generation* What's interesting is how you can format a prompt to give GPT-4 the context of a section. It then generates content relevant to that section and the overall document.

3. *Deciding what to write next* It's cool that GPT-4 can analyze a document and suggest possible sections to expand upon. We can then choose which direction to take next.

4. *One-click content generation* I like the idea of a simple button that generates context-appropriate content based on your current location in the document.

5. *Standardized text manipulation* This approach offers a standardized way to select a region, generate a prompt, and manipulate text. You can shorten or lengthen it, change the timing, fix grammar issues, or even adapt it for posting on Twitter. It's a versatile method for content generation.


Congrats on the launch! Your product appears well-crafted and polished. I have a suggestion I'd love to share. As others have noted Type.ai doesn't seem to be highly differentiated from existing products like Notion AI, Lex, and many more. One intriguing direction to consider is focusing on knowledge organization and integration.

What if the AI can sort and use all the information in the system in their writing (bookmarked webpages, other notes and text in type..). For example, when you were demonstrating the launch blog post creation, you likely had plenty of related content saved in folders within the system. It would be great if the AI could organize and incorporate that material into the blog post.

Achieving this will require significant optimization and fine-tuning, but it could potentially create an interesting competitive advantage. Mem AI is moving in a similar direction, but not specifically for writing purposes.


So basically, do something else.


What a thought, huh? "If you want to differentiate, you need to come up with something else"


not just something else, something more.


Looks awesome! Congrats on the launch.

Just one small silly bit of feedback - in your demo video you show one of the use-cases as showing you coming up with fake user testimonials, maybe not the best use-case to show!


On the contrary, it’s where the money could be. ChatGPT didn’t show that AI is “smart” now. It showed that many more jobs can be automated with a machine that can’t reason, but is good pretending it can. LLM aren’t going to replace developers soon. It’s Filipino freelance content writers from Upwork who should be worried.


I really need to re-record that, I agree! I make a passing comment to replace them with real quotes, but at that point I guess the AI wouldn't have been all that helpful.


Ah, I had audio off! :)


What I would pay for is the exact opposite of this.

I am not looking forward to being deluged by pages and pages of AI generated content that will surely be sent out by MBA types looking to make a name for themselves as visionaries and flooding my inbox.

What I want is a tool that reads documents or corporate memo or email and extracts the key message into a small paragraph or two.


My co-founder and I both resonate with this and we probably need to make our positioning clearer given that it didn't come through here.

The thing that excites me most about generative AI isn't "more," it's better. I often use Type to write satire and now whenever I hit a block, I don't tab over to Twitter – I have Type generate some ideas. Often, I don't use them as-is but they do inspire a new angle I hadn't thought of.

That said, I can see how we have a lot of work to do making that more clear. And I can certainly see how a product like ours or ChatGPT could be used to produce lots of mediocre writing – which is undesirable.

Appreciate the feedback!


Can't bing chat do that already? You can tell it to summarise the page you are looking at.


I think the core idea is strong, but it's missing several features that could differentiate it as a product:

1) adding an "advanced" tab which enables the user to input more information about the document, e.g. formatting, length, etc...

2) learning the user's writing patterns and style - I'm not sure if this can be done by prompting, but my guess is that there are methods that you can use to steer the prose towards the user's default style

3) plagiarism detection - this would require a large document store and could be implemented using fuzzy/semantic search via Milvus followed by string matching

Keep up the good work.


Agree on each suggestions. I’m especially excited about building #2, which I think is a fairly subtle, hard problem to solve.


My thought on #2 is to continually evaluate and store off a 'best' sample of the author's writing taken from everything produced that hits on a number of key metrics (ie. general wordiness, adjective usage, humor, unusual quirks that are statistically significant), then compile that down to where it might seem to be complete gibberish but an effective and performant prompt.


Many of us have 20, 30, or more years of written electronic communication.

Main issue preventing just* chucking this into one of the LangChains is (a) unrolling email threads mixing many people's writing, (b) keeping partitioned tones (workplaces, topics, networks, cultures) distinct.

* No such thing as "just".


This looks really cool! I think the UX around selecting a subpart of the text and asking it to rewrite that is very promising (also for stuff like code editing; it'd be awesome to have this built-in to your IDE, not sure if VS Code already has it).

My only worry with this is that I'm not sure what the long-term edge will be. This whole product looks a bit like just a feature that will soon be added to MS Office Word. I'd love to hear more from the authors about how they plan to differentiate themselves here.


We think about that a lot! I think this reply from my co-founder summarizes our answer well: https://news.ycombinator.com/item?id=35442714


I'm getting burned out on the AI stuff. Every day it's half the posts on the front page. Startups, tools, it's neverending. It's somewhere on the spectrum between huge hype and real paradigm shift, so it's not unexpected, and very possibly not unwarranted, just tiring. If I could go a week without hearing about ChatGPT and AI tools that would be nice.


It's a tough one because follow-ups [1] and repetition [2] aren't in the spirit of HN (it's hard to stay interested under a repetitive barrage), but significant new information is [3], and there has been a ton of that too. So we're in tradeoff land. If it helps at all, we've downweighted a ton of AI-related posts that don't really contain new information. I know that many are still making it through though.

The Launch HN posts are a special case because that's one of the things HN gives back to YC in exchange for funding it; and as you can imagine, there are a ton of AI related startups wanting to launch right now. However, Demo Day starts tomorrow, so you have two more threads to endure today (the other one being https://news.ycombinator.com/item?id=35445097) and Launch HN season should simmer down after that.

[1] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

[2] https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...

[3] https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...


Maybe HN should make a category. I'm feeling your fatigue too.


We don't have topic categories. We do try to downweight topics that have had large amounts of coverage, except when there's significant new information. The links in https://news.ycombinator.com/item?id=35445497 point to lots of past explanations, if anyone wants more.


One of the first times I recall hearing someone gripe about the frequency with which a topic was appearing in news was during the Eurocrisis, when a news presenter quipped about the Greek financial situation being in the news rundown once again.

9/11, the 2nd US-Iraq Gulf War, the 2007-8 Global Financial Crisis, and Covid-19 come to mind.

I consider this a strong indicator that there is in fact a significant story developing, and whilst AI has certainly presented us with sixty or seventy years of false starts, the prospect and impacts of fast take-off make it the sort of thing that 1) is hard to judge either in advance or as it's happening and 2) would absolutely change the world. Exceptions include fad/fashion, political, and some investment hype-cycles.

Even if the current OpenAI / LLM fuss is a false alarm / limited breakthrough, it's breathtaking enough as it stands and as has already been demonstrated to take very seriously. Soaking up solid takes (and yes, they're hard to sort from the chaff, another characteristic of major stories) does seem worthwhile.

You can flag or hide stories on HN, so long as you're logged in (and have met the fairly minimal karma thresholds).


I’m with you on this, especially when people don’t realize that plugging GPT into some arbitrary product isn’t an original idea. It’s a bit cringe.


I'm looking forward to the collapse of the generative AI bubble (like the cryptocurrency bubble it's replacing) and the coming AI winter.


It’s not a bubble. People consume a ton of crappy content. Generative models can produce such content at massive scale. There’s money in that.

Just think about Marvel movies, it’s a lot of CGI and mediocre story.


congrats stew! (good to see you back with a new idea)

sooooo. this is a classic business strategy sort of thing. you, an AI startup, have to build Notion, faster than Notion can build AI features.

your work is cut out for you. i dont have any suggestions but would love to hear your thoughts on how to outcompete massive incumbents.


Are people happy with Notion? We didn't renew our (extremely expensive) subscription. The product is clunky and slow. As a team, we never needed the collaboration features. In the rare case where we do need some real-time shared writing space (mostly just for taking notes during Zoom meetings), we make an O365 document.

As individuals, a lot of us have moved to Obsidian, but we aren't using it collaboratively. Personally I just use it as a simple note taking space; I keep a note for everything I've googled multiple times, and I can pull it up quickly with a simple cmd+p or cmd+shift+f. Notion provides basicallly the opposite experience (open a website, wait for it to load, use their shoddy search, wait for that to load, then maybe find what you're looking for).

As a team, we don't feel like we're missing anything without Notion. We collaborate in markdown using GitLab pull requests and Mattermost chat messages. Some of us write that content in Obsidian and then paste it into the GitLab text input (have you ever tried pasting markdown into Notion? Good luck with that!)

I think some non-devs might prefer Notion, but as a dev, the idea of using some proprietary React frontend as a note taking tool is the opposite of what I want. Obsidian is great.

IMO Notion got distracted with this "database" idea, where everything is a "block," because the reality of it is that the experience of everyday text editing becomes infuriating. Nothing will make me resent a product like unintuitive shortcuts that hijack my return key and closing backticks (also see: ClickUp).


I find that it's great for keeping track of my D&D campaigns. It has a lot of features that make it great for managing that small group of people. I also co-manage a WoW guild with it. Roster, todo's etc. Great for a small number of collaborators.

I don't really produce documents with it. That does seem annoying with all of the blocks. I just use Google Docs for that use case. I find Notion to be something like Evernote with Airtable dropped right in. I don't think I would use it to replace Confluence or whatever. But as a way to share my Org-mode oriented brain with other folks, it works nicely.


Hey swyx. Stefan, co-founder and CTO chiming in here. I think our strategy here is pretty simple (although not easy). We have our opinion and our vision on what this product ultimately should look like, we trust that that opinion coupled with our capability to execute and listen to our customers will at the end of the day deliver a product that has enough differentiated value that is carves out it’s own segment. This probably sounds a bit hand-wavy but I really think that is how you need to operate. If you’re too focused on what the competition does the product loses its soul.

At the end of the day though that thinking obviously needs to translate into a set of features that sets us apart. When comparing to Notion specifically we already have a few of those that make us stick out and that our customers appreciate such as offline first support, instant search, writing suggestions, and most recently our chat integration.

Btw huge fan of your new podcast! :)


Stefan pretty much captured my perspective! I might just add a couple of related things:

We have a subtle but important difference in focus compared to a product like Notion. We're not aiming to build the best knowledge or workplace management product. We're really focused on building something that helps you author high-quality content (usually, that will be shared publicly).

Secondly, IMO the end-state of many of these products won't look like Microsoft Word/Notion + AI. I think entirely new interfaces and workflows will be discovered over the next 2-3 years that wouldn't have been possible without today's LLMs. The one advantage we have is no priors – we can take big swings on "risky" ideas.

Like Stefan said, I know both of those probably still sound a little hand-wavey but it's part of what keeps us motivated to keep building.


thanks for very thoughtful response and for listening! still trying to figure it out and let me know if theres any topic or guest request you have!


I like this. Keep it up guys!


It certainly has potential, I like it. There are also possibilities for additional features like prompt repository etc. What doesn't work for me is the pricing... normally, one already pays for ChatGPT+ and ChatGPT APIs, and now this wants additional more than $20... I would consider it, if it would somehow use the chat window or APIs with my API key, and the software will be $10 max.


Appreciate the feedback on pricing! Will keep this mind as I imagine we'll offer a wider range of options in the future.


I've been using Type for a month now and it has really helped me. Its nice to just have a fully feature rich editor that is modernized with AI /ChatGPT too. Nice work to the team!


Glad you've been enjoying it. I'm continually surprised by how much more I feel I can do working with ChatGPT/AI in a doc editor interface versus exclusively chat (though each have their unique drawbacks).



Yep! We're definitely not the only show in town. I'd love to know if there are any Lex features you prefer over Type.


Two comments:

1. This should've been an addon/plugin for the top-5 most used text editors (Word, Google Docs), potentially also a plugin for WordPress/Drupal/Facebook/Twitter/Instagram, instead of a standalone text editor that nobody's going to download.

2. Looks like every YC startup now is going to be a thin wrapper around OpenAI's GPT endpoints. "Dump your ideas into this textbox and let the magical black box add some fluff". Things are going to get boring, old and non-original very quickly.


Literally nobody installs addons for word or Google apps. Nobody cares about WordPress. You can't have plugins for social networks, and this is for long form text anyway. Bad ideas all around


Didn't try it, but isn't it how Grammarly works? Just an addon for any input field


This. Should be a browser plugin for maximum deployment impact.


I think a docs plugin would be pretty sweet. Would install. It's probably not the kind of thing you can start a company around, though. More like a side project.


Google will soon be adding this sort of stuff directly to their office tools anyway, just like Microsoft is.


> Nobody cares about WordPress.

29.41% of the world's top 10K websites use WordPress. 29.65% of the world's top 100K websites use WordPress. I wish nobody cared about me like that.


And nobody cares about those statistics too


"Nobody cares about WordPress"

Are you sure about this?


Yes


And companies even block them for security.


> Things are going to get boring, old and non-original very quickly.

I’d say they already have. I have half a mind to write a HN front end that filters out any posts with the phrases “GPT”, “LLM”, and “AI”.


Nice. I do hope you'll let GPT4 do that for you.


I'll use it more than this type or whatever document thingie


I went ahead and threw one together: https://save-buffer.github.io/no_ai/


I would gladly use it


Agree. The AI stuff is pure novelty, I’m skeptical it’s going to be truly adopted in its current form in any substantial way.


This is HN... People enjoy not using Microsoft or Google products. Do it on neovim and people will be happy.


Feature, not a product. Not hating on the founders, but has YC gotten so big it just accepts projects based on keyword matching now?


YC accepts about 2% of the ones applying, and plan for only 2% to become home runs.

I like this idea. So simple, but a great product on top of ChatGPT

(Cited from one of the lates YouTube videos, don’t remember which one)


Tons of products have been built on a single great feature.

Tons of amazing products have also completely failed to gain traction.

In this case, the goal is to build a business. If they can get enough people excited about their offering, it doesn’t matter if you consider it a feature or a product. You are not their target market, and that’s ok.


There are large encumbents in this space that can't be hand waved away with 'not their target market.'

FWIW I am their target market. I will use AI powered document editing when Atlassian, Github and Notion integrate it as a feature.


Hi Alan. Our target audience is people that write long form content. In order for the product to provide a great user experience for that audience we feel that we need to build an opinionated editor and integrate AI in an opinionated way into that editor. It’s hard for me to see how that is a product inside GitHub or Atlassian.

Notion (and others) is undeniably an incumbent that as you say can’t be hand waved away. With that said we are already getting feedback from paying customers that feel we’re better at some things, including a snappier editing experience and a more natural way to interact with the AI portion. I also think our chat integrates in a way that sets us apart.

If you really think you are in the target audience I would encourage you to give it a chance. We’re very open to feedback and if there’s anything specific that you think other products are doing better, we want to address it.


Hi,

When I read the description I immediately thought of Ben Evans talking about how what happens when your entire company is rendered into a feature by an encumbant [0] (FWIW I disagree with his attack on antitrust).

So, how are you going to make sure Type survives Microsoft adding ChatGPT to Word, or Substack offering something similar, etc.

[0] https://www.ben-evans.com/benedictevans/2020/12/21/google-bu...


> FWIW I am their target market. I will use AI powered document editing when Atlassian, Github and Notion integrate it as a feature

You just contradicted yourself. You said you are the target market and then immediately described why you are not the target market right now.

If you are not an early adopter of this thing, or if it doesn't have an immediate application for you, then you are not the target market. Whether you like it or not and whether the project owners know it or not.

When you consider it a viable alternative to solve your needs, then you will be their target audience.


Also, so many are building their whole business around an API they do not control. They are one OpenAI decision away from being deleted.


OpenAI may remain the leader in the space, but there plenty of alternatives already. Give it a another minute or two and the alternatives will be as good as the current GPT4.


It’s now start to exist open source versions of ChatGPT and competitors, so I don’t think this is going to be a significant risk in 1-2years


So far it's looking like OpenAI will have a surprisingly shallow moat. Open source is already right on its heels. Midjourney is killing it in image generation with 11 employees.

Then in a few years Apple will do what it does and step in to make billions off of a nearly mature technology.


Is it looking like that? They are still way ahead of Google, yet alone the open source alternatives. Of course things could change quickly.


There are alternative models and with something like langchain you could swap them or run multiple in parallel.


People said the same about iOS apps, but some of those companies were eventually purchased for billions.

In general, if you want guaranteed success this is the wrong industry for you. Sometimes you've just got to accept the risk.


Complete with fake testimonials. I'll be downvoted for this, but that product is borderline unethical.


That's what I thought. The first headline it generated was "How Type Works". But how would it know? It's just spewing out bullshit.


How you tell


Looks very similar to what Google will be adding to Google Docs—what makes Type worth almost $300 a year?


The honest answer is that it'll be hard for us to answer that until we try out what they've built!


sorry but this seems so easy to copy and not really a something new, we havnt seen so far.

I am going to wait until smbd makes it on github for free


Yea, and I can easily imagine in the near future you don't even have to wait for somebody to copy these simpler "app-around-an-API" apps because AI models will generate the interface and the simpler logic for you on the fly. There will be still value in the human input: the ideas of the creator / prompter about what kind of features should the AI include, but if others too have access to the same app-generating AI then they can just generate it based on the same idea. So copying simpler apps will become a lot easier and you would need some complex feature as the core value of your app to compete with AI generated apps.


Same feeling here: a simple ChatGPT prompt, masked as an Editor, having a Pricing page on it.

I might be wrong. And I had another feeling: soon every YC startup will do the same thing over an over again: pick any idea + chatgpt + pricing


Seems like their new criteria of entrance might strike the quality :(

>52% had nothing more than an idea.

https://twitter.com/paulg/status/1642566043053826048?s=52&t=...


This is quite interesting, the recent launches of AI API wrapper companies posted on HN draw my attention. I like the innovation and use of the latest AI technology, but at the same time, I see a wrapper around an external API. Yes there can be more services providing LLMs, but those services are part of huge tech cos with established software and customer/user bases. The GPT-effect on the market is good, maybe internal reorgs will focus on UX improvements to integrate AI more into existing products (at what speed/pace...can a startup steal the thunder or just show what UX is needed...)

Genuinely curious how this all pans out, it is an exciting time!


You can do this in emacs, it's not really a big deal. It's a very simple editor that takes the text you've typed (or the text that's in context) and formats it as a query for "Open"AI's API. The only thing they're banking on is the "type.ai" name which I can assure you, no one is going to remember in the face of multi-syllable, heavily marketed, non-common-verb company names.


smbd?


* somebody. the idea is simple: gpt4 + python + reactjs + PSQL + 2 evenings -> working demo.


You probably don't need python and P(ostgres?)SQL too?


"somebody"


[A bit out of scope] I suspect that in pretty short time people will stop reading any texts from the Web.

And indeed, human got a good skill to skip ad blocks on pages.

Next will be any texts on web pages: why to bother reading stuff that AI throws on us?


I was afraid of that too.

But then I remembered that I had already stopped reading some common Google search results because they seemed too low effort for a very long time. And they often looked like they were generated, not written. Just three examples from different fields: Quora, CNN, and CNET. But the list is much longer. In non-English parts of the Internet, there is also poorly auto-translated content from websites like Stack Overflow, which is weirdly high in the Google results. Fortunately, I found extensions to block these websites in Google search. So for me, and I believe for many people, it has already happened.

On the other hand, I enjoy reading articles from Simon Willison and Adam Johnson. And even though now we have very powerful chatbot services that can explain anything to you or effectively teach you some skill, I will most likely continue to read the content that they put on their blogs or elsewhere.

Reputation did matter, and it will matter even more in the future. I believe people will continue to read other people's texts. At least I will.


Or, what I believe to be a more plausible future, you just read, watch and listen to content from sources that you consider worthy of your time.

Example: videos uploaded in the YouTube channel Linus Tech Tips have a high probability of being made by the Linus Tech Tips people and not just being completely AI generated, unless of course if that made the videos better, but I still think that the videos would be curated by the Linus Tech Tips team.


Why bother reading stuff that some underpaid overworked copywriter throws at us at the behest of a growth hacker, with the sole purpose to get us engaged?

I, for one, would rather read an insightful piece by an author who has been helped by an AI than a soulless product of a content farm.


How soon do you think we'll have AI coming up with its own unique writing style?

When do we get AI Hunter Thompson?


Why? Texts generated by AI are pretty useful. I read ChatGPT answers to my questions literally all the time.


How sure are you it's not lying to you? I asked ChatGPT to write a description of common plasma cutter table features, and it didn't know the difference between initial height sensing and torch height control.

I closed the browser tab and haven't gone back since.


In many contexts it's easy to verify what ChatGPT tells you. There are ways to use ChatGPT as a tool that do not require it to always be right for it to be useful.

For example, the other day I asked it something about the Flask codebase, and it found the relevant part of the codebase immediately. When I asked it about the behavior of the code, it wasn't always correct, but it still showed me the relevant code so I could read it way faster than I would have found it myself.

Initially my impression of ChatGPT was the same as yours - I asked it some questions in a specialized domain I know well, and when it was wrong I decided ChatGPT is useless. But after enough people told me they find it useful, I took another look and tried finding more applications. And since then I've been impressed by what it can do.


I can definitely see it being good for assisted learning. Quickly groking codebases seems to be one of the most popular uses.


I just used it to write some tests, then I implemented and it failed. It then continued to explain what my setup was missing. Regarding coding v4 is become pretty accurate. And if it makes a mistake it's able to explain what went wrong.

Regarding certain medical conditions I've asked to list the studies and explain them, it does that pretty well.

But just as talking with a human, or with googling info on a website, or with Stackoverflow, Im always assuming I need to double check it.


True that. It is replacing or complimenting some of my teachers right now.


What makes you think the texts that aren't from the web are any less likely to be written by AI?


I really really liked https://reflect.app but needing an internet connection to access my notes is a miss for me. I don't need an app but I need something offline and locked down. I managed to replicate the experience of Reflect with Obsidian and several plugins such as a calendar, meetings and their canvas feature. I also don't even pay for syncing, I encrypt the vault (also a plugin) and keep the files in GDrive that I use and sync across devices.

I like this but would not pay for another note, bookmark, todo or markdown/rich text editor app service. Also arguably true to hn fashion someone could roll this or get very close as an obsidian or other widely used tool plugin in short time.

I also just checked reflect's page and they added a gpt4 prompt feature a couple weeks ago and you get all this for $10/mo. https://reflect.app/changelog


The pricing is too high. There should be some option to use existing chatgpt subscription. No way, $29/month but maybe at $3-4 / month I could see myself using something like this. That said, I am not sure how far away is microsoft loop away from this given they are starting copilot option.


I've been using this company's product for a few weeks now and I'm really impressed with its ease of use and functionality. The team has clearly put a lot of thought into designing a tool that is both powerful and user-friendly.

That being said, I do need a dark mode and markdown syntax conversion.


I love the workflow! This is how I'm writing articles now with ChatGPT's GPT-4, just with less copy-pasting and all in one place, and fewer Markdown problems.

The two things I'm missing, or didn't find how to use:

1. GPT-4 :) I know the API isn't public yet, but the reasoning abilities of GPT-4 are so much better that I'm having a hard time arguing with GPT-3.

2. I have a long prompt I give to GPT-4 (context on our product, writing style guidelines, text examples for style, words to avoid, etc.). It's about two pages long, in addition to the request for the specific article or paragraph I'm writing. How should I incorporate that into Type's UI?

3. How do I import/export a whole article, or paragraphs, as Markdown? EDIT: Copy-paste. Lol, simpler than I expected.


Glad you're liking it and thank you for the feedback!

1. GPT-4 is definitely on our radar and is something that we are planning to support as an option. There are trade-offs though as it's a lot slower and as people have mentioned, a lot more expensive.

2. You can use the built-in chat the same way you would use ChatGPT. With that said I know that is not the ideal way to achieve what you're looking for. We have some ideas around features that will address this specific problem. If that is something important to you and you'd like to chat about it feel free to email me and we can talk about it: stefan AT type.ai.


GPT-4 is way too expensive atm for anyone to offer it in their products, unless they're charging _a lot_, or are delegating to the user's API key. It's also very slow and unreliable for any production app.


$29/mo is more than my ChatGPT Plus subscription, which does include GPT-4 :)


ChatGPT Plus GPT-4 is heavily rate-limited, as well as slow for this use-case. As of now.

This is almost certainly operating off of <=gpt3.5-turbo.


They could have a button "think for more time" or something and allow a certain number of those per day. This is how Wolfram Alpha works used to work, for example.


I feel as if there's 100+ of these splattered across daily deals software sites and betalist.com.


We feel similarly. There are a lot of products in this space. Very few are enjoyable to use, though.


It's something I would use, but pricing is way too steep. Compare WebStorm+Copilot, which is $17, and has a tangible benefit. Paying per usage might be better because a fixed $29/month might barely be used at all.

There's also more value on mobile because of how much slower it is to type on mobile. I can get close to typing at the speed of thought on a keyboard anyway. Something that types faster than the speed of thought means that it has to read my mind and get it accurate more often than not. Copilot works great because it's filling in blanks, but it hasn't done well for more creative forms of writing.


Can you add a library of documents? The model should use search to pad the prompt with relevant demonstration examples before generating the answer. It would be much easier to draw from a known library of text than just using raw GPT.


This is something we're very excited about building. We haven't shipped it yet, but we have a pretty clear roadmap to get there. Stay tuned!


I'm not code savvy, would be lovely if I could point type at our drive account and have it use that as a source library.


Agreed! We’ve begun exploring some ways to do that.


I don't see how you can ever profitably offer unlimited AI usage for a fixed fee.


That's definitely fair, we do think we'll eventually need to develop a more sophisticated pricing strategy & usage limits


> "Generate a testimonials section."

What a perfect use case for AI hallucinations!


So basically you need to do the same amount of writing, except instead of practicing actual writing you are giving instructions to the computer. Good for spammers, but real writers don’t need this.


Is it using openAPI APIs behind the scenes?


Yep, we are primarily leveraging OpenAI's APIs. We've started to experiment with Anthropic's Claude as well and it seems promising.


Congrats on the launch, Type looks great!

I was curious, is there any difference you noticed for your use case with GPT vs Claude?


Thanks!

Claude seems to be especially strong with creative writing and writing in a wider breadth of styles, which ends up being really important.


The paragraph-by-paragraph contextual help provided here makes the interaction with AI much smoother! Makes me think something like this would also be a natural fit for Jupyter notebooks.


The "unlimited" usage is interesting - will you be checking out the history of the top 10 or so users to see if anyone is using your text editor to train a smaller model?


We will need to develop more sophisticated pricing and monitoring to make sure this doesn't get abused. For now, we haven't noticed any nefarious usage - but that's partially just a function of us just being small / under the radar up until now.


Maybe I haven't been patient enough but I use chat gpt 4 all the time for code.

But to write real quality articles is quite hard to get right.

It uses a lot of words, but a lot of fluffy filler sentences.

With code it just seems to get me, especially V4. But with writing it's always off.

Maybe it's me or maybe they've just put a lot of time in training it on coding feedback, or maybe writing is harder because it's much more interpretable what a good article is.

In some cases I actually got better results writing with 3.5.


There is this terse but dense sweet spot between Crichton and Sire (Naming the Elephant) that I aspire to.

Both require deep knowledge that surfaces in a few well chosen words that are packed together intelligibly. None of the GPT outputs can match that kind of linguistic mastery. I suppose I should add "yet", but I don't want to.


I'm not sure it will. I always thought coding was hard. But I came to conclusion writing is harder.

But it could also be that they are reinforcing gpt by senior devs.

But the writing in some cases maybe reinforced by "normal" people.


That's just ChatGPT with extra steps


Of course, it is... that's all it takes to get VC funding these days.


This tool is designed to make typing and editing documents faster, easier, and more intuitive than ever before. One of the key features of HN: Type is its ability to use artificial intelligence to suggest and autocomplete words and phrases as you type. This means that you can write more quickly and accurately, with fewer errors and less need for manual correction. The system learns from your writing patterns over time, allowing it to make increasingly accurate suggestions that are tailored to your individual writing style.

Data curation involves creating valuable data for users engaged in discovery and data analysis. It includes gathering, maintaining, and managing data in databases or data warehouses to make it useful.

<a href="https://techeela.com/data-curation/">Curated data</a> characteristics include identifying signals, robust data management throughout the data cycle, and supporting data governance.


Yet another shitty text editor riding on the faith that the common verb name and the "ai" suffix will make stonks go up. Delusional. And people say "AI" will somehow lead to good things...

Edit: Intense cryptobro "I have my own coin" energy


It's the same kind of bubble, involving a lot of the same people. We're seeing so many AI shovelware projects now because people are trying to get in while the bubble is still expanding, and get out before the crash.


I don't know how it's get into YC. Is getting into YC now-a-days so easy.

EDIT: I've already built something like this for my personal blog (similar to ghost like editor). That was before GPT-3/4. Only thing remaining is to hook with GPT-4.


Can someone please explain as I truly don't understand: Are thin layers over an LLM considered valid startups now? What's the big deal? Half of the people on HN can build something better than this in a weekend.


> What's the big deal? Half of the people on HN can build something better than this in a weekend.

Just because I can doesn't mean I want to. I have plenty of my own projects to work on; why bother wasting time down making something I don't care too much about when I could do something interesting instead?


It's not about how hard it is to make, it's about how willing someone is to exchange dollars for it. I think Stew and co. are just trying to rapidly find paying users that can help them steer to a better product vision, which they can then build out.

Don't make a startup to do hard things. That's what a hobby is for.


>Don't make a startup to do hard things. That's what a hobby is for.

Still learning this after 10 years of building something that didn't exist before.


The demo videos with different block types are super interesting. Assuming there are plans to add image blocks or other rich/dynamic media? Could see that being very powerful as new models are made accessible.


Yes! We do support basic image blocks today, but the idea is that we’ll integrate generative media types as new image/video/audio models become more reliable & stable.


is there any demand for this? The market for human writing services seems small. Why do you expect people to pay for inferior AI writing when so few are paying humans to write or edit for them?


Yep, we’ve acquired hundreds of paying customers since launching about a month ago. And we don’t see the market here as “people who hire writers” — we see it as “people who must write to get paid.”


Something about the idea of an AI powered text editor generating a testimonials section for itself doesn't feels wrong.


What is the editor built off of? Prosemirror?


Should I just apply to YC with a company that's nothing more than "(X) thing people do every day but powered by GPT-4's API"?


Looks pretty darn good. I love the flow of hotkey-prompt-output. I’ll try this for writing my sci-fi story.


Appreciate it, would welcome your feedback after you've given it a spin. You can email me anytime at [first name] at type.ai.


How much did you pay for the domain?


Too much. Mid-five figures


Why would you spend that much money before validating that anyone would pay for this?


We knew we wanted it and we figured there are more people like us out there.


Great piece of real estate!

Love the product demo/UI


Appreciate it!


Did you spend as much on product development too?


Haha tough crowd. (Yes)


What export formats does it support?


Hi! We currently don't have an export feature although that is something that we want to add. With that said we've made it as easy as possible for you to copy and paste our documents into other editors. If you paste a Type doc into Google Docs for example it will retain all your formatting and if you paste it into a text editor, it will land as correctly formatted markdown.


A .tex export format would make this super attractive to researchers writing scientific articles.


Good to know, thanks!



Cool, keep building! (This isn't a feature of Microsoft 365 or Google docs yet?)


Do you have a privacy policy?


Nice work. I would recommend a name that is more unique for lookup purposes.


That's fair! At a minimum, we probably should start saying "Type.ai" more often than just "Type."


Looks nice and fast, but I'm not interested for that pricing.


All the nay-sayers can go fuck themselves.

-

What about an AI built self-employment model.

--

What if you apply this to, say, making a decent self-shop for etsy-type users "Build me a hop using these product libs that I have made with descriptions and the picures" etc...


Now it is completely impossible to understand the meaning of mysterious definition "a solid authoring experience" because it's completely unclear who is the author.


No support for other than English?


It should support pretty much any language! The commands will attempt to generate text in whatever language is already in the document (though, it's certainly not perfect yet and will sometimes return English no matter what).


I'll try it.

Hope it's better than Lex.


Looks dope


You might have just given Microsoft Word, Google Docs and Notion free ideas to absorb.


-Honestly just me spouting off because I'm in a bad mood. Can't delete-


We thought it was pretty descriptive and at least as imaginative as "Word" or "Docs." But we're open to suggestions!


Honestly, fair. Sorry I'm a little off today but that's definitely fair. I ended up reading your description and your mission actually aligns with my problems with the experience of using LLMs for writing. The whole "one shot block of text" thing can be overwhelming and make it harder to use. I'm going to delete the parent comment but wanted to say that.


No worries! I really appreciate you giving it a closer look, and I’m glad our general philosophy seems to resonate. Please shoot me an email if you ever try it and have any feedback. I’m [Firstname] @ type.ai.


Every day for the last few months I've been wishing I had a cabin in the woods with zero modern technology. A big shelf of books, my acoustic guitars and some comfy places to sleep.


My internet died earlier this week. I put on a vinyl record (Paul Desmond) and sat on the couch enjoying the brief break from my digital life. Hoping to enjoy more simple times soon.


I’m literally planning to do this in a few days to get some solid focus time in. The thought of the disadvantage of not having an AI model to support already has crossed my mind (no internet access in the woods)

It’s noted that this editor supports offline - does this mean that the AI features also run offline? Or a limited version?


Unfortunately, everything but the AI works offline. Though, maybe that's a feature if you're planning a more mellow retreat :)


Have you considered a limited LLM that could run locally?

> planning a more mellow retreat

The objective here is to forcefully going to where internet is impossible (no phone reception, I don’t have starlink) with the objective of focused productive output with limited distractions.

The idea came to mind after reading about John Carmack doing this for a week, diving into AI using nothing but classic text books and papers as reference material to work off.

EDIT: here is the HN thread on Carmack’s week long retreat:

https://news.ycombinator.com/item?id=16518726


> Have you considered a limited LLM that could run locally?

I think there are two main issues here. LLM are large (the name even hints at it ;) ) and the smaller ones (still, multiple GB) are really, really bad.

Edit: and uses a ton of memory, either RAM if CPU or VRAM if GPU.


Are they all that bad? [1]. I’d be ok using a few 100gb on my laptop, given that storage is so cheap these days.

[1] https://github.com/nat/openplayground


Compared to GPT-4, most of them are not super great, yeah. I've tested out most of the ones released for the last weeks and nothing have been getting the same quality of results, even the medium sized (30GB and up) models that require >24GB of VRAM to run GPU. I have yet to acquire hardware to run the absolute biggest of models, but I haven't seen any reports that they are much better either for general workloads.


gotta love the fact that this is the top comment here


Winter is coming...

On a serious note, I hope AI will help me write documentation. But at that point, do we still need documentation?


If you use AI to write the documentation, then it's unlikely to be better than the documentation people could generate themselves with GPT-4. But it will be a while before it's better than documentation written by professional writers that have all the context, target audience in mind and know what they are documentation inside and out.


The big advantage of doing it ahead of time is that you can proofread and edit it. LLMs are good, but it'll be a while before they get to a point where a bit of human editing doesn't help.


Everyone wants to generate AI content, but no one wants to read it.


In popular music (and I mean that extremely broadly, as in most music, that is being sold) there is an interesting thing going on.

Listeners don't really wants to know the truth. They don't really wants to know that Justin Bieber had nothing to do with the genesis of Sorry. They don't really want to know that most "live performance" videos were created in studios and filmed afterwards. Nobody is really interested in how music is made, because studio work is mostly draining, moving through the dark and an endurance test. Nobody wants raw and honest, because nobody even has any idea how that would sound.

But everyone imagines they do. It's not even that information is well hidden. Everyone wants there to be a great story so much, that they are very happy to ignore the flimsiest of veils and just believe what they must. And the industry has always been happy to provide. It is the essence of stardom.

I am fairly certain we are not going to notice of even care about how it was generated, as long as we can be made to love it, which will still require an entirely different set of skills.


I think you are confusing apathy about attribution with quality/originality. Sorry was produced by Skrillex who is unquestionably a musical genius. Even though most people aren’t interested in the producer credits Sorry is only as good as it is because of Skrillex’ work


Where's your evidence?

AI content will surpass human content in quality and quantity. Moreover, it will enable vast numbers of people to do things they were previously unable to do due to budget, time, resources, depth of talent, opportunity cost, etc.

Meta point: I've never seen so many smart people so bearish about the future. The science fiction dreams and utopias we've celebrated since our childhoods might actually be feasible now, and yet so many are wearing frowns.


It'll surpass human content in quantity by a lot almost definitionally and in quality probabilistically.

But I'll have to read AI content on both sides of that probability.


> yet so many are wearing frowns.

I think the phrase "if it seems to good to be true it probably is" is one way (among many) to interpret the scepticism.

I think that it's easy to jump to conclusions about the implications of this technology. Towards either over-optimism or over-pessimism.

Expect the worst and welcome the best seems as good approach as any?


People are bearish for 2 reasons IMHO.

LLMs are in many people’s opinion overhyped. Their capabilities are overhyped.*

People don’t trust the powers to be to use AI to benefit the masses. Instead of the Star Trek-like utopias, we are more likely to get Blade Runner-like dystopias.

* Tech has a real “hustler culture” problem, I guess it was always a part of tech culture, but with the advent of cryptocurrency it really exploded. With the implosion of crypto (and real people losing a lot of real money), all the hype people need something to hawk. “AI” is that something, it’s the new “blockchain”.


I am very concerned about the future of jobs and workers.

Our capitalistic society is salivating at the idea of cheapening the cost of production by eliminating all the creative and generative work that workers make. The capitalist class doesn't care who gets hurt or what gets destroyed as long as they make more money.

Unfortunately the people that are hurt the most are the middle and working class folks who use their skills to create things, make money, and immediately spend it. These are the folks that keep the whole system running by using their money to buy more goods.

> I've never seen so many smart people so bearish about the future. The science fiction dreams and utopias we've celebrated since our childhoods might actually be feasible now, and yet so many are wearing frowns.

As it stands we aren't heading towards utopia, we are running head first to a bleak dystopia; where critical thought and creativity is authorized only to the algorithms and probability machines of the wealthy and powerful. While the rest of us are relegated to cheap, often dangerous, labor.

If we had support systems: Universal healthcare, wages, education, etc. I might be more supportive, but we don't because capitalism.


While such tools may have their benefits, they also have significant drawbacks that could ultimately make the world a worse place.

Firstly, AI-powered document editors rely heavily on algorithms and pre-existing templates to generate content, which means that the output can lack creativity and originality. As a result, we risk losing the human touch and the ability to express unique perspectives and ideas that cannot be replicated by a machine.

Secondly, relying on AI to generate content can lead to a lack of accountability and transparency. It can be challenging to trace the source of information, and this could lead to widespread dissemination of false or biased information. This could have disastrous consequences, particularly in areas such as politics or finance.

Thirdly, AI-powered document editors could also lead to job loss and exacerbate existing societal inequalities. It's likely that many jobs that require writing skills could be automated, leading to significant job losses. This could particularly impact those who are already marginalized and disadvantaged.

In conclusion, while AI-powered document editors might seem like a convenient solution, it's essential to consider their potential downsides. In my opinion, it's crucial to maintain the role of humans in creating content, so we can preserve creativity, accountability, and fairness in our society.


I assume this comment was written using AI, which gets the real risk of writing with AI across - there’s no substance to the writing, just generic words and sentences about a topic. And as always when you ask ChatGPT to write something, it starts the last paragraph with “In conclusion…”, as if to prompt itself that it’s time to wrap up.


Everything you say seems like made-up and irrelevant. In fact, an AI could write the same.

There is no evidence that AI-powered writing leads to lack of creativity.

AI is not responsible for the publication of documents. It is your duty as (co-)author to make sure that what you say is valid, and to make the necessary fact-checking before publication.

If anything, AI will on the opposite gives access to work to MORE people, possibly people that are less proficient in writing, but still might have interesting ideas.

All this AI FUD, it's really the history repeating itself.


I'm convinced this was written by ChatGPT.


GPT rolled


This is unbelievable! I think Type will truly empower its users to write stunning prose with minimal effort. It's rare that something comes along that generates value for the VCs while also making the world a better place. The SFBA is very special -- no place else in the world could such great ideas come together with flawless execution. The creators of GPT missed the boat here; competing tech is really catching up.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: