AI can guess 3D objects and how it moves

Hey everyone! Wow, I'm genuinely stuck with what to say here today. I wanted to make a Turkey/Thanksgiving reference, a joke, even. But I just kept drawing a blank. So instead of stuffing anything in here, let's get to it.

😏

🤌 Ben's Picks

• Meta AI presents CICERO — the first AI to achieve human-level performance in Diplomacy, a strategy game. Basically, AI strategically plans world conquest, negotiating with humans and then betraying them. Cool. Cool. Cool. Cool. Cool. AI is just a toy, right?! (link)

• I like pretty much anything Steph Smith does, so it’s no surprise that her podcast interview “The Future Of A.I. Businesses With Steph Smith” is something I recommend listening to/watching. She is always on top of trends and interesting opportunities. Steph is a podcast host at a16z and talks about bringing AI into the workflow (see, there’s a theme with smart people - work WITH the tools). My personal opinion is that using AI for [insert anything here] is basically like a designer using Photoshop. It’s a tool to get a better output than you could without it. (link)

• MagicPony learns a function that can estimate the 3D shape, articulation, viewpoint, texture, and lighting of an articulated animal like a horse, given a single test image. The method is trained purely from in-the-wild single-view images of the object category, with minimal assumptions about the topology of deformation. (link)

🛠️ Cool Tools

• Heyy - Create your own AI-generated avatar model, without code. (link)

• PixelVibe - Find high-quality, curated AI-generated stock images. (link)

• SciSpace - Your AI assistant to discover and understand research papers. (link)

• RecruitingEmails.ai - Make your cold emails a lot warmer with highly personalized, exciting copy powered by AI and GPT-3. (link)

• Cover Letter Writer - Use AI to personalize your letter of motivation based on the company's needs. (link)

• Boolpic - An AI-powered toolkit transforming your pictures in seconds. (link)

🔬 Research

• A new approach to 3D inpainting that addresses the challenges of view consistency and geometric validity. The proposed method uses a perceptual optimization-based approach that leverages learned 2D image inpainters. (link)

• HyperTuning uses a hypermodel to generate task-specific parameters for a fixed downstream model. (link)

• Never Ending Visual-classification Stream (NEVIS'22) is a benchmark consisting of a stream of over 100 visual classification tasks. (link)

• ConCoRD can be used to boost the consistency and accuracy of pre-trained NLP models. (link)

• Exact Diffusion Inversion via Coupled Transformations (EDICT) is a new method for inverting denoising diffusion models (DDMs) to produce an input image from a noise vector. It is more stable than the state-of-the-art approach, which is based on local linearization assumptions. (link)

• SinDiffusion is a method for generating photorealistic and diverse images from a single image. The method is based on two core designs: training with a single model at a single scale, and identifying that a patch-level receptive field of the diffusion network is crucial and effective for capturing the image's patch statistics. (link)

• A new text-to-image benchmark that contains a suite of thirty-two tasks over multiple applications that capture a model's ability to handle different features of a text prompt. (link)

• Videogenic - extracting highlight moments out of videos by converting long-form videos into engaging short-form highlights. (link)

• DiffDreamer - generating long-range novel views. The DiffDreamer framework is unsupervised and is able to synthesize views depicting a long camera trajectory while only training on internet-collected images of nature scenes. (link)

• Real-time synthesizing of talking portraits. (link)

• EVA is the first open-sourced billion-scale vision foundation model that achieves state-of-the-art performance on a broad range of downstream tasks. (link)

🤓 Everything else

• Hugging Face recently added materials to their Depth Estimation task. Depth estimation is the task of predicting the depth of the objects present in an image. (link)

• I put together a thread of the best AI threads I’ve found on Twitter. Yeah, it’s meta. (link)

• A generative AI prototype for 3D models and scenes built in 2 weeks using only 8,000 models. There’s a huge need for 3D generation but there are also problems to overcome. Physna, a 3D search and analysis company, did this research to push the boundaries of what’s possible. (link)

• It looks like folks are moving to their own custom models using Replicate from Astria. (link)

• There’s been a lot of chatter about Galactica, Meta’s scientific search + generation tool. And we all love reading Alberto’s takes don’t we (he wrote about GPT-4 rumours). So here’s his not-so-hot take on Galactica. (link). Personally, I think these tools shouldn’t be taken as fact, but we need to be able to play and push these things when they come out. Luckily Emad, from StabilityAI, is looking to release an open-source version. (link) Never a dull day in AI!

• Categories for AI talk: Causal model abstraction & grounding via category theory - by Taco Cohen (link)

• Discoo Diffusion - Personalized Stable Diffusion with few-shot fine-tuning, create your own Stable Diffusion on a single CPU. (link)

🧑‍💻 Who's hiring in AI

VEED.IO - Simple Online Video Editing. VEED is hiring AI / ML engineers to level up its creative toolkit and make it more magical.

Buildspace - where builders, build! They're looking for an ML/AI instructor to build their new course.

Add your company here | Join the Talent pool

🖼 AI IMAGES OF THE DAY

The Mother F***ing Wizard; Samuel L Jackson as Dumbledore.

🤗 SHARE BENS BITES

Send this with 1 AI-curious friend and receive my AI project tracker database!

or copy/paste this link: https://bensbites.beehiiv.com/subscribe?ref=PLACEHOLDER

👋 SEE YA

⭐️ HOW DID WE DO?

⭐️ REAL REVIEWS