This is really cool, I'd love to see this pull data from [1] HuggingFace, [2] Replicate Hype, and [3] GitHub.

[1] https://huggingface.co/models?sort=trending

[2] https://hype.replicate.dev/

[3] https://github.com/trending

I'll try to accomodate this, thanks for the feedback!

Essentially, this would remove the requirement on tech companies to prove they cannot hire an American for the role before hiring from overseas.

As reference, here is Microsoft's response: https://www.regulations.gov/comment/ETA-2023-0006-0112

I’m struggling to understand from this paper whether the approach is better in the general sense (all cases, with wider models seeing greater benefits) or purely for wider models (with narrower models seeing detriment)?

If it’s the former this could effectively halve finetuning cost overnight which would go a significant way towards enabling a wider array of use cases for LoRA.

Don’t we already have open source achieving temporal consistency up to 15 seconds? I might be misreading the article but the main achievement in Sora is consistency over 30min+ generations

Where did you get >30mins from? OpenAI themselves said Sora is limited to 60 seconds, and most of their examples were less than 20 seconds long.

There are a lot of places e.g. Replicate where you can finetune and deploy language models. You’ll need GPUs, but you can simply select a hardware class and pay per second of usage, similar to AWS Lambda or similar serverless infrastructure.

Serverless AI is quickly becoming popular precisely because of the scenario you’re describing — it’s currently still pretty hard to deploy your own GPU stack, not to mention crazy expensive to run eg an A100 24/7, plus orchestration for scale up/down. It’s why so many model authors don’t host their own demos anymore and simply toss it on HuggingFace or Replicate directly.

Serverless providers will basically do the infra for you, as well as make necessary GPU reservations, then charge you a slight premium on the reserved price — so you’d pay less than on-demand on GCP/AWS, while they benefit from economies of scale.

I do imagine at some point soon GPUs will become cheap and more commercially available so you could away with hosting your own VPS in the distant (or near?) future.

Is it really a slight premium or more like “at least 2x of the cost”?

> Software Procurement by Federal standards is relatively straightforward

> FedRamp and FIPS compliance

It’s odd to see these in the same sentence. FedRAMP is so insanely complex/difficult to achieve in a straightforward way. Even by your own estimate for a series E startup (with lots of capital and the ability to spend >18 months< on compliance) there’s a 3M$ variation in cost.

That rules out every startup or SME in software and that’s why you have Palantir, half baked tech that rarely delivers/is somehow more universally hated in USG than ServiceNow. Yet able to seize the space and hike prices endlessly due to compliance being so difficult to achieve — they realize/accept this as their edge as well and it’s why they so aggressively pursued IL6.

The good news is that this is going away and USG is strongly reconsidering its approach here. CMMC, imo, is a huge step in the right direction.

> It’s odd to see these in the same sentence. FedRAMP is so insanely complex/difficult to achieve in a straightforward way

Agreed! Hence why I said "relatively". It's an easier procurement system than for other products in the Federal space.

> That rules out every startup or SME in software and that’s why you have Palantir

Tbf, Palantir's federal usage is kinda overstated from what I've heard from peers.

But yea, I agree, and made this point in another comment

There are a bunch of great upscaler models although they tend to hallucinate a bit, I personally use magic-image-refiner:

https://replicate.com/collections/super-resolution

As a huge asterisk this is a Tauri wrapper around Tiptap that does filesystem management. You can use Tiptap yourself easily as it's open source:

https://tiptap.dev/

https://github.com/ueberdosis/tiptap

Tiptap is immensely popular, supports both Vue and React, is super lightweight and has a huge community of extensions. I've been using this since 2019 and can't recommend it enough. A number of comments here are asking for functionality that is built in, easily available as an extension, or easily add-able!

Why do you need a huge asterisk about the editor control library this app is using? Many text editors follow this pattern:

    Notepad++ : Scintilla
    Notepad.exe : Win32 EDIT control
    TextEdit.app : NSTextView

It’s not like you can download and run Tiptap as an application, right?

Why is it a "huge asterisk"?

I'm also working on a WYSIWYG editor and using TipTap. It's great, but on its own, it's not a ready editor, but a great framework.

There's a lot you can do with TipTap (and ProseMirror) to make your editor stand out and fit your use case better than others.

> this is a Tauri wrapper around Tiptap that does filesystem management

Isn't Tiptap a wrapper around ProseMirror that does SaaS pricing?

Esri is an anticompetitive company leeching taxpayer dollars that only exists because ArcGIS trapped the government before many govcon laws preventing vendor lock in were put in place.

We would have 10x the mapping capabilities and much higher velocity on defense tech without this parasitic company.

Btw, also recently fined for blatant sexism against hundreds of female employees.

The open source geospatial community is pretty amazing :

https://wiki.osgeo.org/wiki/Main_Page

Esri in the cloud is terrible. Geoserver is much better.

Cloud Native Geospatial might break Esri's hold :

https://cloudnativegeo.org/

Yeah esri sucks and it’s software is worse than open source but it’s the only thing they teach because contracts

Not loving that there are more details on safety than details of the actual model, benchmarks, or capabilities.

> That’s why we believe that learning from real-world use is a critical component of creating and releasing increasingly safe AI systems over time.

"We believe safety relies on real-world use and that's why we will not be allowing real-world use until we have figured out safety."

Yeah, it would be way better if they just released it right away, so that political campaigns can use AI generated videos of their opponents doing horrible/stupid things right before an election and before any of the general public has any idea that fake videos could be this realistic.

Let's make it safe by allowing only the government (the side we like) and approved corporations to use it.

That'll fix it.

you joke, but the hobbling of these 'safe' models is exactly what spurs development of the unsafe ones that are ran locally, anonymously, and for who knows what purpose.

someone really interested in control would want OpenAI or whatever centralized organization to be able to sift through the results for dangerous individuals -- part of this is making sure to stymie development of alternatives to that concept.