Excited to announce that I've founded new company! is dedicated to solving the AI alignment challenge–ensuring advanced AI systems are robustly aligned with human intentions & values
Honored to have
@lcastricato
@fdesouza
as cofounders & Seed funding from
Stable
@ClipDropapp
😊
"Stability AI’s First Acquisition Furthers Its Goal Of Delivering The Most Advanced Creative Applications With Its Open Source Generative AI Models"
Read more ➡️
🚀 New Stability AI
@Blender
Addon🚀
Use
#StableDiffusion
inside your favorite 3D software.
Generate textures, generate AI video from your renders, and more. ❤️
#b3d
Woahhhh... wtf?!
"No Modality Left Behind" = transform ANY user input into high-def generative
#GenerativeAI
audio with Make-An-Audio 🎧
📝 Text → 🎵
📸 Image → 🎵
🎥 Video → 🎵
🖌️ Inpainting → 🎵
📸 Take a depth scene screenshot of camera view
🎨 Use Controlnet to generate images with the screenshot
💻 Import the texture and project it from the same camera view
🚀
#UE5
+
#StableDiffusion2
FTW
AI unlocks creativity by flattening learning curves (scripting 🧑💻) and eliminating unnecessarily tedious tasks.
Your own personal troubleshooter & liaison—Asking ChatGPT to write scripts for complex 3D tasks in
@Blender
. Terrain, textures, animations... 3 point lighting!
🚀 We just launched SDXL 0.9 with massive improvements in image & composition detail.
3.5B base model & 6.6B model ensemble pipeline + 2x CLIP models, including the largest OpenCLIP model to date.
Research blog coming soon.
@kylegriffin1
Y'all are all mindlessly jumping on a freak-out bandwagon. "Verified" isn't some preferential endorsed by
@Twitter
status, it's just proof that an account is representing who they claim to be. You *want* well known ppl to be verified so you're not tricked by fake accounts.
🔥BLIP-2🔥 demo is live! Come play with LLMs that can understand images and share your examples!
Project page:
BLIP-2 knows mass–energy equivalence! More examples in the 🧵
@Timcast
These are hilarious, but not reflective of modern AI text-to-image.
If you haven't seen DALLE, I made a list that will probably blow your mind:
Let's be clear here - Using AI does not break laws, it is free - it is not a corporate monopoly, it is not fascism, it is not theft, it is no plagiarism, and it certainly is not ok to wish or imply that you hope that us "tech bros" die in a hail of bullets.
#AIArt
#StableDi
…
The Spotify model: autonomous "squads" for focused projects, group related squads in "tribes" for collaboration, organize "chapters" by skill sets, and "guilds" to foster a culture of trust, innovation, and continuous learning
@Scobleizer
@elonmusk
"Long after severely overfitting, validation accuracy sometimes suddenly begins to increase from chance level toward perfect generalization. We call this phenomenon ‘grokking’."
Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Models
turn the publicly available, state-of-the-art text-to-image LDM Stable Diffusion into an efficient and expressive text-to-video model with resolution up to 1280 x 2048
abs:
After an amazing year at
@StabilityAI
, I'm excited about a new chapter. This past year has been nothing short of amazing, a testament to the extraordinary people who I've had the privilege to work with. I want to extend my heartfelt gratitude to
@EMostaque
and the entire
Medusa can improve the LLM generation efficiency by 2x
Instead of using an additional draft model like speculative decoding, Medusa merely introduces a few additional decoding heads
Just 1 day since
#StableLM
launch, & GitHub stars already skyrocketing 🌟
To prevent Y-axis congestion, GPT-NeoX, Pythia etc left out. Featuring LLaMA, Alpaca, Dolly, & Vicuna below.
Adoption is accelerating, but this is just the start - Buckle up, folks! 🚀
Late, but couldn't help myself.
#DeepFloyd
is great at AI caricatures.
"This new Bing will make Google come out and dance, and I want people to know that we made them dance" —
@satyanadella
#DeepFloydIF
“No, sir, we did not use copyrighted images. We believe Dall-e 2 spent too much time on the internet and has learned to enjoy trolling.”
(Alamy stock image watermark)
#dalle2
#dalle
#gpt3
I disagree. Apple & OpenAI might too.
Apps with no value-add features, just profiting from "AI Chat" keywords, are competition magnets 🧲. Go make a better app (yes, 🫵).
As competition grows 📈, they experiment with positioning and marketing, accelerating the potential to
Now that Twitter knows about
#dallemini
, y'all might as well have fun with these other free generative AI apps.
A few 💎's you might not know about...👇
@mcuban
@elonmusk
@Meta
Twitter → Global consciousness/information. Until we all get
@elonmusk
's Neuralink, written text is the most efficient way to consume information. AR/VR on
@Meta
is unlikely to directly compete with
@Twitter
's strength.
@bowtiedbower
@GuruAnaerobic
Maybe a better way to frame this would be, "one piece of specific advice is difficult to generalize or systematize." If you blindly follow a piece of specific advice... good luck iterating.
Goldman predicts generative AI could increase global GDP by 7% and lift productivity growth by 1.5 percentage points over a 10-year period. Estimate that two-thirds of US occupations are exposed to some level of automation by AI, of which between a 1/4 and 1/2 of their workload
🚨Announcing the release of DeepFloyd IF🚨
Our multimodal AI lab,
@DeepFloydAI
, is publicly releasing their state-of-the-art text-to-image model.
Learn more here →
@AntonNellCrypto
Hey Anton, I’d like to continue maintaining/operating chainlist—would you be willing to hand over ownership instead of letting the domain expire & get snapped up at auction by a phisher?
Massive AI models (like GPT-4) aren't an efficient or scalable approach.
Smaller & more specialized AI models working together or in a decentralized approach is a more viable solution 🌐🤖
Sure giant models aren’t the way but swarms are but like was google ever going to give giant models to billions of users? Would need all the flops in the world 🌎
Google’s core business is aggregation per
@stratechery
innovation etc is cherry on top
In just 3 months,
@SakanaAILabs
(co-founded by my friend
@hardmaru
) lands NTT partnership, the world's 4th largest telecom 🚀
They're building an AI constellation—a distributed net of smol, task-oriented AIs, autonomously cooperating
A big step towards collective intelligence!
We’re excited to be partnering with NTT Group to develop AI in Japan!
Partnership announcement on
@Nikkei
NTT、元Google研究者が設立のサカナAIと提携
NTT partners with Sakana AI, a company by former Google researchers.
I think I'm in love 💚
@pipedream
Adding Python code (serverless) between generic (pre-built) App/API tasks + /tmp access + generous dev tier big enough to experiment = immediately solved EVERY pain point I had with Zapier/IFTTT/Et al.
(Fantastic brand name too)
ControlNet +
@clipdropapp
API's have been integrated!
You can now use much better normal maps to capture the juiciest features with Stable Diffusion.
@EMostaque
my
@StabilityAI
canned response ✉️
📉 “Sry, we've been...🧪💣🚀”
📉 “Oops”
📈 "The
#ai
accidented & sent the
#chatGPT
w/o asking me 1st"
📈 “Indelibly confusing email miscommunications like this are actually valuable training data 😮☝️!”
It's ready! 🚀🚀🚀🚀🚀
Introducing Ellie - your email assistant! 💌
Ellie is powered by
@OpenAI
and will learn your writing style and reply to emails as if you wrote them 🔥
If you want to be an early user, please retweet and comment below and I'll send you an invite code! 🙂
ChatGPT stock portfolio outperforming top UK investment funds
19% of UK adults would consider getting financial advice from ChatGPT
A further 8% said they were already doing so... 🫠
Yet, 38% of Brits still ponder, "What's a ChatGPT?" 🤔
What I would *REALLY* love to have is a private version of ChatGPT that’s been trained on your internal org documents. Have an HR question? Ask Chat. Don’t know how to use some application? Ask Chat. When is the next vacation day? Ask Chat.
Current approaches for alignment, evaluation, and dynamic adaptation of AI models have shortcomings we aim to redefine through cutting-edge research
Check out the great article by
@rachelmetz
@technology