Record-speed AI images

Hey folks, last full day in Milan for me, so I need to cram in a pizza ASAP. I'm loaded up with podcasts and articles for the journey home tomorrow (all AI-related obviously). Food last night was an all-round 8/10 - crispy prawn tacos were a dream.

See ya next week. Let's get to it.

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 33k subscribers? 

Advertise with us | Job board | Join Discord

🤌 Ben's Picks

1/

Qualcomm demo record-speed AI image generation on a mobile

I remember at some point in secondary school (high school for our friends across the pond) I got the “world’s slimmest phone”, and then suddenly things got really small really quickly, and now I realise it really wasn’t that slim 🙁. Anyway, are we starting to see the same with AI? Qualcomm demonstrated using AI image generation on a mobile phone whilst simultaneously hitting record speeds for doing so, and the results look spot on. Being able to generate AI images on your phone would be really cool. One big caveat though: it looks like the demo was on hardware with custom optimization that’s not publicly available… Hmm.

2/

Photorealistic 3D photo rendering using NeRF

Whenever I see these NeRF shots of images it makes me feel really weird but in an excitable and good way! It’s so cool how the camera develops this ability to make it feel like you’re really in the scene, poking and prodding around and viewing 3D objects at all angles. The Luma AI team have taken this technology to the next level, showing volumetric photorealistic 3D rendering in real time. To really see how awesome this is, click the link and check out the demo.

3/

Pinecone is becoming the MongoDB for AI

Writing sentences like that pick header often makes me wonder what new-joiners to the tech world must be thinking about the wacky names we come up with. But we love ‘em. For those who don’t know, the AI language models doing the rounds rely heavily on vector embeddings - sentences converted to numeric vectors. And Pinecone seems to have established itself as the leader in providing storage and retrieval of embeddings, with developer usage doing 20x since ChatGPT hit the scene. That’s huge. But there’s a big, important question: Like MongoDB, will Pinecone maintain product stickiness and stand strong in the face of Big Tech?

4/

Some very funny logo modifications using AI

Time to finish the week on a light-hearted note that’ll send you into a weekend of fun. ControlNet is a new model that allows you to take control of parts of an image, and then let AI work its magic on the rest. This Twitter thread shows someone’s experiments passing famous logos (Starbucks, Burger King etc) through ControlNet and seeing what AI can muster up… Some of the results are hilarious, especially the Starbucks one for me 🤣. Happy Friday all!

🛠️ Cool Tools

• Onu - Turn scripts into internal tools without doing any frontend work. (link)

• Booth - Create pro quality product photography with AI. (link)

• Buildt - AI tool to help developers quickly search and understand large codebases. (link)

• Salient - Personalise your outbound at scale. (link)

• Autofiller AI - Automatically fill forms and applications using information provided by you. (link)

• AudioShake - Take any song and break it into its stems. (link)

• PromptLayer announces PromptRegistry to manage prompts outside of your prod codebase. (link)

• What the cron - Write cron expressions from natural language. (link)

• Test & Start - Validate your startup idea. (link)

• Mymind - The extension for your mind to remember everything. (link)

• GPTFlix, AI for movie reviews, makes its code public. (link)

• Aigur Client - A free and open source MIT library to compose and invoke fully typed generative AI pipelines. (link)

• Summary.legal - AI-powered transcripts and summaries to help you organize and search video calls and meetings. (link)

• Luma Labs now has full volumetric photorealistic NeRF rendering on the web. (link)

• Mini Yohei Autodeck- Generate a basic pitch deck from a startup description in 30 seconds. (link)

• Arcwise AI - GPT copilot for sheets. (link)

🤓 Miscellaneous

• ChatGPT shows the U.S. government needs to step up on AI. (link)

• Nvidia extends its AI ambitions to the cloud. (link, paywall removed here)

• Chinese apps remove ChatGPT as the global AI race heats up. (link)

• Hugging Face CEO on the future of open vs. closed source in AI. (link)

• An interview with Kevin Systrom and Mike Krieger about Artifact. (link)

• Blinded by analogies. (link)

• Scale AI - Why data will power the AI revolution. (link)

• Source.ag raises $23M to raise the bar on raising crops with AI. (link)

• Watch robot in action using diffusion models to augment data. (link)

• AI-created images lose U.S. copyrights in test for new technology. (link)

• How big tech is leaning on EU not to regulate general purpose AIs. (link)

• Runway Studios creative grants - Giving filmmakers everywhere the production support they need to realise their creative vision with AI. (link)

• The 2023 MAD landscape - Machine learning, artificial intelligence & data. (link)

• Qualcomm demos fastest local AI image generation with Stable Diffusion on mobile. (link)

• Currents 082 - Dan Shipper on practical applications of GPT-3. (link)

• Why the future of ML is open source. (link)

• How I broke into a bank account with an AI-generated voice. (link)

• Pinecone - The MongoDB of AI. (link)

• Microsoft has been secretly testing its Bing “Sydney” chatbot for years. (link)

🎓 Learn

• How users can use RLHF in their own assistants with trlX. (link)

• Text-to-Image Diffusion Models - A guide for non-technical readers. (link)

• Techniques for improving the training performance of your PyTorch model without compromising its accuracy. (link)

• All about prompt engineering - Guides, papers, lecture, and resources for prompt engineering. (link)

🔬 Research

• MimicPlay - An imitation learning algorithm to teach robots to perform long-horizon tasks efficiently and robustly. (link)

• Language model crossover - Variation through few-shot prompting. (link)

• Pre-training generalist agents using offline reinforcement learning. (link)

• Aligning text-to-image models using human feedback. (link)

• Diffusion Prior from Adobe - Controlled and conditional text to image generation. (link)

• Region-aware diffusion for zero-shot text-driven image editing. (link)

• Use an encoder to personalise a text-to-image model to new concepts with a single image and 5-15 tuning steps. (link)

• DiffusioNeRF - Regularizing NeRFs with denoising diffusion models. (link)

• Portrait distortion correction with perspective-aware 3D GANs. (link)

• VoxFormer from Nvidia - Sparse voxel transformer for camera-based 3D semantic scene completion. (link)

• Learning neural volumetric representations of dynamic humans in minutes. (link)

• MERF - Memory-efficient radiance fields for real-time view synthesis in unbounded scenes. (link)

• On the robustness of ChatGPT - An adversarial and out-of-distribution perspective. (link)

• Can pre-trained vision and language models answer visual information-seeking questions? (link)

• Teaching CLIP to count to ten from Google. (link)

• Learning visual representations via language-guided sampling. (link)

📰 Unclassifieds

• Dexterity Robots is hiring a Head of Computer Vision and Machine Learning. Learn more >>

• Braintrust is hiring a freelance ML engineer. Learn more >>

• Checkout Ben's Bites job board for more jobs here. Post a job for free if you're looking to hire for AI related roles.

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 33k subscribers? 

Advertise with us | Job board | Join Discord

🖼 AI images of the day

The carpool lane

🤗 Share Ben's Bites

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

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

⭐️ How did we do?

⭐️ REAL REVIEWS