Now let's get to the actual things we covered, of course the biggest announcement from @JustinLin610 was about Qwen2 suite of models, Justin is a long time ThursdAI panelist, and now had some #breakingNews from Qwen beating LLama-3 70B and becoming the top Open Source LLM!
x.com/JustinLin610/status/1798747072319074347
Not only did they drop all the models + all the GGUF versions + an MoE + apache 2 license, Qwen team also reached out to Eric Hartford @LucasAtkins7 and @FernandoNetoAi and IIRC this is the first frontier OSS model to drop alongside a finetune by someone else! 👏
x.com/cognitivecompai/status/1798751837216272792
We also had the awesome pleasure to host the @JinaAI_ CEO Han (@hxiao) and long time friend of the pod, and our embeddings expert @BoWang87 to join and talk about their new multimodal embeddings 👏
Bo goes deep and when he does, we all learn🫡
x.com/JinaAI_/status/1798333409800110125
And in the spirit of friendly competition where we all win, we also had the awesome folks from @nomic_ai@zach_nussbaum come and talk to us about their SOTA multimodal embeddings also launched this week!
I kept urging these folks to keep fighting for SOTA and I'm sure they will!
x.com/nomic_ai/status/1798368463292973361
We then were blown away by the #breakingNews of the Chinese (I just don't know that company so I generalize, but its kuaishou) SORA competitor that's KLING that's blowing up. Just look at these videos, they blow everything else that's not SORA (Runway, Pika, SVD) out of the 🌊
x.com/altryne/status/1798729308795761117/video/1
Loved covering the new MixEval framework and benchmarks, especially great because Qwen2 was already evaluated on this, and it launched just a few days before!
Hoping to chat with @NiJinjie next week!
x.com/NiJinjie/status/1798182749049852411
Of course, how could I not excite everyone with new @suno_ai_ capabilities of uploading your own sounds and creating music with it?
I did this remix of "one pound Fiiiish" and while it didn't go totally viral, it definitely should have!
x.com/altryne/status/1798493515477094548
We of course covered that @udiomusic was not left far behind and launched this feature this week to their paying customers as well, I wasn't able to test it out but it seems to work well as well!
x.com/udiomusic/status/1798369297758077066
We had @isidentical on stage for the first time from @fal to talk to us about their GPU grants, so if you're building something cool, FAL will sponsor your compute (if you get in, but I know a guy so LMK)
What Batuhan forgot to mention that Fal also YOLOd this into production
x.com/burkaygur/status/1798859400490484048
And while Batuhan was there, we covered a few diffusion model related news, the most impactful of which, is a new distillation method from @clipdropapp folks over at @heyjasperai - a super exciting low step diffusion distillation with high quality outcomes!
x.com/benjamin_aubin_/status/1798707273650389082
and if that's not enough, @abidlabs joined with yet another #breakingNews from the @Gradio world, where they basically announced that every Gradio project will now become an API endpoint and 2 new client libraries, turning every "DEMO" into a production grade API server! 🤯
x.com/abidlabs/status/1798856443204116687