Is not having to build a dockers image worth $100 a month? I do find server setup to be a pain but I think if I will use a model for a year, I can take the time(3-4 hours) to set it up. Only with constant switching of models would I use a service like this.
I never set up bigger models like LLAMA on servers. Other hacker news people can chime in.
It's not only about "building a docker" but also maintaining multiple models, multiple environments and a lot of users. Imagine there is a group of engineers each needing to deploy their own models: one needs tensorflow 1.x, one needs tensorflow 2.x, one needs pytorch and one needs a very strange combination of dependencies. Trust me, things get complex very easily:
I definitely agree that for a fixed use case, building a docker once and for all is probably the simplest and best approach. However, it quickly gets more complex and out of hand.
Also the basic plan is free for independent developers. You don't need to pay more than as if you were using EC2 instances, but with the platform convenience - we definitely hope it's worth it!
Making a GPT + text to speech wrapper is not complicated.
Providing all other features (e.g video generation, podcast publishing, auto translation and many other features we’ve added that allow for higher quality pod creation) increase the level of complexity for reproducibility.
Ultimately, we aim to keep building features that lead to higher quality pods, easier to build, and integration of ansiliary (video, translation, show notes) that will enhance our moat.
Amazing work. Listening a bit to the HN podcast, I'm impressed by the natural-sounding pronunciation of technical terms with non-obvious phonetics like 'postgres'. Have you had to tweak a lot of these manually to get them sounding so good or is your model mostly getting them right?
Very cool. By accumulating lots of these tweaks, I feel like you're also going to have an opportunity to backdoor your way into a great text-to-speech API product as well if you have any interest in going that direction. It seems like the main challenge there is ironing out all the edge cases and you've created an excellent feedback loop for accomplishing that.
Webflow is so far ahead with CMS and now membership that it'll be hard for open source to catch up unless it moves significantly faster.
I think if this product integrates some kind of popular node CMS it might be able to compete. I might be wrong but it just looks like an front end editor similar to GrapeJS
You are correct regarding CMS. Same story with e-commerce and other things.
In fact fundamentally the architecture is built around the idea of not having those things built-in but rather providing APIs for others to create integrations with any data source: CMSs, Databases, E-commerce etc.
Webflow and others with built-in CMS have a massive amount of problems, because there is no one CMS that fits all.
Even harder is e-commerce.
So yeah, we are not going to compete on that because we are not building any of it, but instead focusing on architecture to support integrations.
Webflow CMS is expensive, I can see a small business starting out with Webflow and getting locked in. I can't see a more mature business doing this, as they'd want hosting flexibility. WordPress page builders don't have any hosting lock-in.
I think twitter now would be something like tiktok where relevant and entertaining stuff will be shown to me without browsing for it. That is the true value of Tiktok, saving the search time for something interesting.
Also one reason I think Tik Tok is successful is the lack of political content which is toxic and stressful. A lot of political content is designed to get you to be angry at the "opposition" which is not a healthy state of mind.
Does the new generated picture take into account of all previously generated image or just whatever is around the square, the first is amazing, the latter was a feature that was already there.
Regardless, this is a great way for people to fight the lack of detail in Dall-E which I think is one of it's largest flaw.
10 years ago I had the idea to write a children’s book about the joy and difficulties of building something; That idea was there but the hassle and cost of illustration always prevented me from finishing the book.
Recently, Dall-E was launched and I excitedly used it for my children's book illustrations. You can see the result and also issues I faced such as maintaining a consistent art style and face removal.
I used Dall-E a lot and get into a lot of the same issues, I think Dall-e needs parameters that are fixed for things like:
-percentage of the entire drawing that the image you want to draw should take; a lot of times I think the object I want is too "zoomed in" or large; a circle background is a good way to limit it but I think it should be more obvious
-No way to fix the color of the background so that it can fade in easily to other images or design
-Reuse drawing styles to generate further image to explore further and maintain consistency
A syntax could be: Octopus juggling blue database cylinders, digital art, cute, image-size:40%, background-color:#304324. With image-size, and background-color being keywords in the definition
I was looking for this comment, this is pretty much a copy of the MacBook 3 years back.
Although I am happy they are doing this since no one else is bringing a high quality metal laptop with the same "feel" that you would get from a Macbook aluminum Unibody laptop.
The best windows laptop before was a Macbook running bootcamp.
Unpopular opinion:
I don't believe that is the case. AI has always been in the realm of math and computer science. And the researchers fired was researching only the ethics of it. A lot of researchers are apolitical and only there for the challenge or the pay.
If anything, I think what happens deters investment in AI ethic research. For Google, it has only produced bad PR without any value to their influence and market value in the AI community.
Edit: Also Google's strategy of giving away a production level AI library like Tensorflow has basically locked up their status in the AI community for years to come.
The article mentions the impact on the broader research community too. Samy Bengio is extremely well known, not for ethics. He's one of the original authors of Torch.
They won't have trouble recruiting low and mid level AI engineers, simply because the pool is large. But the pool of highly experienced engineers and leaders in the space is much smaller, and even a moderate reduction in interest there could have huge long-term implications for the company--especially since those high-level contributors have greater financial freedom to follow their principles.
You live in la la land if you think they will have a problem hiring AI leaders. If you pay them they will come. Even the couple of researchers who got fired realized they made a mistake. Bengio left but he was already there for 15 years and is a multi-millionaire. He probably would've left anyway sooner or later.
I think you misunderstand how much leading AI researchers will be learning from complexity science and networks, and how much they'll understand [informational] diversity itself as critical to all networks (including their own community of practice).
My assumption is that there is a dovetail between understanding the role of randomness and diversity in neural networks, and understanding of its role in the social computation of society. Those who don't see that relationship and don't put it into practice in their own value systems will not be "at the top", because they are partially blinded.
Just my hot-take. Might be wrong. But I think there's a bias that favours these value systems amongst top tier researchers, unlike traditional hard comp eng studies.
I never set up bigger models like LLAMA on servers. Other hacker news people can chime in.