nwoli

There always has to be an implicit totalitarian level of force behind such safety to give it any teeth

Another reason llama is so important is that once you’re banned from OAI you’re fucked for the entire future AGI products as well.

The logic of HN cheering this because elon is controversial

Loras on things most people don’t make loras for are already producing amazing art in my experiments (far away from the cookie cutter ai output style)

Hopefully it can help do this on emscripten files too and help adblockers dechipher obfuscated code for that purpose

Love every program having a little police bundled in them these days

Regulatory capture, monopoly prices when the earth is dependent on them and no new companies can form

This is kind of a HN dropbox comment. Very few non programmers would go through that

If you meant that last question sincerely: I’ve been able to improve the recommended tweets a ton (its basically good and useful now) by hammering on the “not interested” button a lot. It started out garbage as for you too though

Reddit reply

Base model LLMs are funnier than instruction finetuned

I just meant a synthetic human (a dalle/stylegan/stable diffusion face output).

One legitimate one I could imagine is if people want to pursue a career in the adult film industry but without having to reveal their true face (not with a celebrity face of course)

They barely make use of it for things like gemini (or else their model would likely surpass gpt-4) so I’m not that convinced by this

Brockman isn’t leaving just going on a sabbatical/vacation

One elegant approach for this I’ve found is this https://github.com/mit-han-lab/gan-compression They basically train an “all in one” network from which you can extract small or large models afterwards (with optional additional finetuning to improve the selected channel size combinations)

The path to too much friction to make new things is paved with regulation with good intentions

That’s not really fair to conclude that the training data contains vanity fair images since the prompt includes “by Vanity Fair”.

I could write “with text that says Shutterstock” in the prompt but that doesn’t necessairly mean the dataset contains that

Huggingface space to try it https://huggingface.co/spaces/stabilityai/stable-fast-3d

Pre generate a bunch of images via sdxl and convert to 3d and then serve nearest mesh after querying

> I think the world is being so disrupted by AI

Is it? I’m seeing a lot of potential for AI/LLMs to distrupt in the future but I don’t really see it happening yet though

Guessing the context here is that the RH was recently translated into Lean. Would be very cool if they threw their compute on that

Same could be said of wrong stack overflow answers or random google results. Clearly they’ll become critical of the results if the code simply doesn’t compile, same as our generation sharpened our skills by filtering bad from good from google results

Scaling linear algebra in the end is probably all we’ll need in the end. Only missing data and compute to get there

They’re definitely doing this on comments too. I’ve had negative towards altman comments at the top dropped to below the negative voted ones in the past

Add the sponsorblock flag in there too and its perfect

Weird since if image generation can run locally a simple classifier (eg clip) aught to be sufficient and run locally as well before presenting the image

Hopefully someone reverse engineers the model (unless this is a joke)

Kind of lost interest in blueskys future after they announced they hired twitters former trust and safety lead. Twitter has issues today for sure but that isn’t one aspect I’d like back (eg shadowbanning and lying about shadowbanning etc).

Funny to compare the signal to noise ratio on this blog vs the linked one to genkit https://firebase.google.com/docs/genkit/plugins/ollama A giant beta warning header and cookies banner on mobile. I’m seeing about one line of actual relevant info before scrolling