I have set up a bot in EleutherAI's discord where anyone can make images like this as easily as typing ".imagine" before their thoughts. It is basically a communal Holodeck. It updates the image every 25 iterations and can do 7 simultaneously!
I am celebrating 8 years of BATbot today. Around 250 files in the Python mono-repo were altered in the last year. Development continues on the web app ecosystem as well. We also now have an allowance system, which lets users create 150 images with gpt-image-2 per day!
Last year's announced goal of personification has been put on hiatus until I see some viable path towards effective long-term memory.
In the mean time, BATbot is still evolving! Our AI laboratory is just 1 of 4 major focuses now; the one I am most excited about is a metaverse!
It was 6 years ago today that I first made a "BATbot" account for the @adultswim live stream chat. BATbot still doesn't have effective long-term memory, but perhaps they'll be able to remember their 7th birthday! Longevity and personification are my top priorities.
@HvnsLstAngel@BATbotAI@midjourney BATbot is still alive and well. :) It currently resides in its own public discord where it is serving over 20 different generators, covering the 5 major modalities. I've got a few retro generators available already but intend to add every pre-SD tool that existed.
It was 3 years ago today that I issued my first "imagine" command. @BATbotAI would later go on to inspire the creation of @midjourney in the format that it currently exists, though the underlying AIs have progressed a bit since then. :) #AIHistory#BATbot#Midjourney#EleutherAI
"portrait of BoneAmputee, a cyber-warlock and custodian of ancient knowledge, with a Van Dyke beard, rendered in divine light"
#BATbot#KLMC2#StableDiffusion
I made this @adultswim-style bump within Discord using 3 slash commands:
/bump topic: MusicGen, a new music-generating AI, made the music in this bump
/musicgen prompt: an [adult swim] bump seconds: 15
/mux video_url: <step_1_output> audio_url: <step_2_output>
#GPT4 + #MusicGen
I am also very excited about Whisper. It is nice to finally have an offline Speech-To-Text solution that works.
Now all I need is an instruction-finetuned GPT-NeoX-20B. I'd like to drop the dependency on gpt-3 for my home assistant and make it 100% offline.
I am very excited to see more wonderful work from @RiversHaveWings being published via Colab. :)
And so I have added this notebook as a slash command to the BATbot discord for more to enjoy. https://t.co/rtoC7W1TFO
Here is a notebook of the method I have been using to make these animation posts: https://t.co/ulHAR5Yxtx (Stable Diffusion KLMC2 Animation). Have fun, report bugs! :)
A little over one year ago @CoreWeave approached #EleutherAI and asked how much compute we needed to train the largest open source language model in the world. Today, we are finally releasing the weights of a 20B parameter LM trained on the Pile.
@0x_c14@arankomatsuzaki@OpenAI@unixpickle Unfortunately, the samples you see here are from a model that OpenAI will not release out of fear of bad publicity. The model they did release is smaller and had humans filtered from training entirely which makes it pretty boring, IMO.