Last week, two noteworthy releases managed to capture my attention. First up, we have HuggingChat, presented by Hugging Face and their ever-expanding repertoire. While its 30-billion-parameter count may not send shockwaves through the AI world, I initially thought the the inclusion of API access is certainly an intriguing development worth exploring, but I believe this to not be the case as I can't find any evidence of it. Please read this as well https://huggingface.co/chat/privacy
The second contender for our attention is DeepFloyd, which aspires to push the boundaries of text-to-image generation. Its apparent enhanced language understanding abilities sound impressive, however I was not able to confirm this. While the image set it's trained on may not exactly sweep me off my feet, it does follow in the footsteps of others such as , Stable Diffusion 1.5, 2.0, and 2.1. Rest assured, I'll be keeping an eye on their checkpoints, and perhaps even experimenting with LoRA, if fate allows. Although I must confess, adding text to images isn't at the top of my bucket list, but progress is always worth a tip of the hat. You can play with DeepFloyd here.
Alright introductions out of the way I want to express my thoughts on them.
Taking a closer look at HuggingChat, I must say that playing around with it was a fun experience, reminding me of the good old days of the wonky Bing chat. However, the 30B model comes with its share of issues – it tends to bug out quite often, making it difficult to get a straight answer. But hey, at least it was amusing, and that counts for something.
Now, about those licensing issues – they've left me utterly confused and quite off putting. I don't want to say things as I don't really don't know what is really cooking here.
The anticipation for DeepFloyd was palpable, but after giving it a whirl and witnessing the resulting images, I must admit my enthusiasm waned. However, let's not forget that even Stable Diffusion 1.5 wasn't all that awe-inspiring on its own – it was only when checkpoints were built on top of it that the real fun began. That said, we're once again confronted with some perplexing licensing issues. Of course, I might be somewhat biased, as I've been spending a considerable amount of time in this area and it's not quite catering to my preferences. But, it wouldn't be fair to judge it too harshly just because it's not doing "what I want." After all, variety is the spice of life. So I will continue to keep my eye out on this one as it may get me enthusiastic about it again.
Initially, I had every intention of publishing this piece on DeepFloyd's launch date. With HuggingChat making waves and DeepFloyd's sudden arrival, it was hard not to get swept up in the excitement. However, as fate would have it, life got in the way and this post was delayed. But as it turns out, the delay proved to be a blessing in disguise.
Taking the time to experiment with DeepFloyd and revisiting my notes helped me approach the situation with a more level-headed perspective. By not getting swept away in the hype, I was able to maintain a cooler, more composed outlook
I can't help but feel a tad singed by the uncertainty surrounding all this information. Even now, I'm left questioning whether my current understanding is accurate. In a constantly changing technology landscape, it's a challenge to stay on top of every detail