Adam King

Excited to announce Imagen Video, our new text-conditioned video diffusion model that generates 1280x768 24fps HD videos! #ImagenVideo imagen.research.google/video/ Work w/ @wchan212 @Chitwan_Saharia @jaywhang_ @RuiqiGao @agritsenko @dpkingma @poolio @mo_norouzi @fleet_dj @TimSalimans

Happy to announce DreamFusion, our new method for Text-to-3D! dreamfusion3d.github.io We optimize a NeRF from scratch using a pretrained text-to-image diffusion model. No 3D data needed! Joint work w/ the incredible team of @BenMildenhall @ajayj_ @jon_barron #dreamfusion

We’re pleased to introduce Make-A-Video, our latest in #GenerativeAI research! With just a few words, this state-of-the-art AI system generates high-quality videos from text prompts. Have an idea you want to see? Reply w/ your prompt using #MetaAI and we’ll share more results.

@CKtalon @ylecun It does say 175, but the OPT readme says "The use of OPT model weights is subject to the Model License", implying all the models listed there are under that license. See the bottom of the readme: #license" target="_blank" rel="nofollow noopener">github.com/facebookresear…

@AdamDanielKing @ylecun That's for OPT-175, isn't it? It's written in the title.

OPT-66B is available. Unrestricted, Open source.

Introducing Imagen, a new text-to-image synthesis model that can generate high-fidelity, photorealistic images from a deep level of language understanding. Learn more and and check out some examples of #imagen at g.co/research/Imagen

@markchen90 A "robot hand" painting a robot hand painting a robot hand painting

"a robot hand painting a self portrait on a canvas" by dalle-2 (openai.com/dall-e-2/)

@HopefulIdiot @alphonzo20 @BretWeinstein @alexandrosM @SamHarrisOrg @clairlemon Not true. The study tested 47 of the drugs in vitro: "ivermectin was able to completely inhibit the SARS-COV-2 3CLpro enzymatic activity in vitro at the tested doses."

@alphonzo20 @BretWeinstein @alexandrosM @SamHarrisOrg @clairlemon These are computation models. They have to be empirically proven. They haven't.

The west was turned upside down with a broadening and weaponizing of the term “racist”. Under Covid the same thing is happening with “anti-vaxx”. Lumping those with scientific concerns about certain vaccines with those who reject the concept of vaccination itself is a dire error.

@amadeus It's a crypto scam. The htm file uses JavaScript to redirect to a base64-encoded website URL where it tries to convince you to send Bitcoin. I guess that gets around Gmail's spam filters? Here's a similar example: reddit.com/r/cybersecurit…

*curiosity boiling*

@ydeigin @BretWeinstein Your main point seems to be about short half-life. See this study in which blood taken 28 days after treatment with ivermectin reduced mosquito survival with statistical significance: thelancet.com/journals/lanin… Blood taken on day 7 had a large effect:

1/5 @BretWeinstein claims we know how IVM works because it binds the SARS2 spike protein. Is he correct? No. The only evidence we have for IVM binding the spike comes from a computer modeling:

@MickWest Points twice as far as the focused plane blur to circles in the focused plane of half the diameter of the aperture. ATFLIR is 0.33 m wide, so aperture is smaller. If the target is 1 m across, then the blurring would be less than (0.33 / 2 / 1) ~= 1/6 of target size (~1.5 pixels).

@MickWest The reason his focus argument doesn't apply to large scales is that his camera's aperture is much too large compared to his model. If the GOFAST video were taken with a huge camera (~100-meter aperture) then it would have similar blurriness.

My response to F-16 Pilot Chris Lehto's "de-debunking" of GoFast and Gimbal. Focus is mostly on A) Focus for GoFast and B) The effect of rate-of-turn on Gimbal - which I think Chris has too high. I'm open to corrections and discussion. youtube.com/watch?v=fBeqP4…

@MickWest I get the same: Banked 30 degrees, centripetal acceleration is tan(30 deg) * (9.8 m/s^2) = 5.658 m/s^2. Setting that to v^2 / r with v = 350 knots, the turning radius r ~= 5730 m. A complete turn takes 200 seconds ~= (2*r*pi) / (350 knots). 360 deg in 200 s is 1.8 deg/s.

Fact check needed. (Updated!) What's the rate of turn in deg/sec of an F/A-18 at 25,000 feet barometric altitude, doing 241 Knots CAS (350 Knots TAS) BANK ANGLE = 30° I get 1.8°/sec?

@kelseyhightower Does GCP not have many A100s? Haven't been able to get quota for even one (12 A2 CPUs), though I can get a lot of other GPUs. Last tried March 2nd

@VovaZk @harishkgarg Only the generated characters. Pricing is character-based because many users don't know what a token is and the number of tokens depends on what model I'm using, which will be upgraded over time.

@AdamDanielKing @harishkgarg ... and do you charge according to the sum of input+output volume (like GPT-3 does), or?

Having a GPT-3 FOMO? Check out - inferkit.com by @AdamDanielKing

@VovaZk @harishkgarg Yes it's alive and my main thing. I just don't tweet much

@harishkgarg @AdamDanielKing Found it: “We're currently using Megatron-11b, the largest publicly available language model. It was created by Facebook and has 11 billion parameters.” Will definitely sign up. I just wonder why doesn’t @AdamDanielKing seem to be on Twitter anymore? Is the project alive?

@MickWest Rounded edge of the counter has a circle of normal vectors that reflect the light into circles. Parts closer to wall form the inner circles since the light has less time to spread out before hitting the wall. Counter is black but mirror finish can be seen from the soap bottle.

Fun optics puzzle. What is going on here? Sunlight through my bathroom window creating a circular pattern on the wall. When I put my hand in the way, there's a concentric ring of shadow.

@MickWest Do you think the fuzziness of the microphone makes it look blurry and so it appears to be out of focus (close to the camera)?

Explaining why a dead cat appears behind Joe Biden. The video also explains why the other microphone appears both in from of and behind his hand. It's because it was. The boom operator bumped his hand, then moved back a bit. youtube.com/watch?v=w_QQ2x…

@anderssandberg @zoe_NELL_ Actually this is not GPT-2. It's Megatron-11b, which is a model created by Facebook AI using Nvidia's Megatron-LM model parallelization. @namangoyal21 github.com/pytorch/fairse…

@zoe_NELL_ This was just app.inferkit.com/demo - nothing special, no pre-training. This is what GPT-2 thinks out of the box.

@ponjimon Sent you a reply in-app. A caching bug was causing the dialog to continue coming up.

@AdamDanielKing How do I create a Custom Generator? I only get the modal telling me the limits of the free plan, even after I added deployment credits to my account.

I've been making an app called inferkit.com, for creating and using text generators based on neural networks. Fine-tuning a generator on your own dataset requires no ML knowledge. Includes an API