Anthropic released an api for the same thing last week.

This is actually pretty smart because it switches the context of the action. Most intermediate users avoid clicking random executables by instinct but this is different enough that it doesn’t immediately trigger that association and response.

All signs point to this being a finetune of gpt4o with additional chain of thought steps before the final answer. It has exactly the same pitfalls as the existing model (9.11>9.8 tokenization error, failing simple riddles, being unable to assert that the user is wrong, etc.). It’s still a transformer and it’s still next token prediction. They hide the thought steps to mask this fact and to prevent others from benefiting from all of the finetuning data they paid for.

The role of biodegradable materials in the next generation of Saw traps

It’s cool but it’s more or less just a party trick.

How many times is this same article going to be written? Model collapse from synthetic data is not a concern at any scale when human data is in the mix. We have entire series of models now trained with mostly synthetic data: https://huggingface.co/docs/transformers/main/model_doc/phi3. When using entirely unassisted outputs error accumulates with each generation but this isn’t a concern in any real scenarios.

Based on the pricing they’re probably betting most users won’t use it. The cheapest api pricing for flux dev is 40 images per dollar, or about 10 images a day spending $8 a month. With pro they would get half that. This is before considering the cost of the language model.

About a dozen methods they could use https://arxiv.org/pdf/2312.07913v2

New record for most buzz words in a headline.

I feel like they should at least provide them with a laptop If they’re going to do unpaid promotion.

The model does have a lot of advantages over sdxl with the right prompting, but it seems to fall apart in prompts with more complex anatomy. Hopefully the community can fix it up once we have working trainers.

Koboldcpp should allow you to run much larger models with a little bit of ram offloading. There’s a fork that supports rocm for AMD cards: https://github.com/YellowRoseCx/koboldcpp-rocm

Make sure to use quantized models for the best performace, q4k_M being the standard.

This is why you should always selfhost your AI girlfriend.

I’ve used the tplink ones that they’re using and they’ve been pretty solid. I can’t say how they’d fare in a 24/7 setup though since they’re not really intended for that.