. . AI: Ideogram's Creativity Booster Lands a Whopping $80M (2.29.24)

Today's biggest news about “creativity booster” Ideogram, which has secured a massive $80M in its Series A funding round. We also cover some cool new guides, posts and tools to help improve productivity.

As always, these are the stories the AI community is talking about most. If you find these summaries and thoughts useful, please share them with a friend.

Thanks!

Marshall Kirkpatrick, Editor

Ideogram Secures $80 Million in Series A Funding

First impacted: Artists, Designers
Time to impact: Short

Ideogram announced that it has secured $80 million in Series A funding, led by a16z, with contributions from Redpoint, Pear VC, Index Ventures, and SV Angel. The funds will be used to advance its generative AI product, Ideogram 1.0, aimed at boosting human creativity. The platform has been called “a printer for the imagination,” and that’s certainly provocative: if more creation equals not just cumulative consumption, but opportunities for transformation for the viewer of the creations, then more and different creators could lead to faster iterative development of the human experience. But printers also have a homogenizing effect, so there may be some trade-offs.

Ideogram is led by CEO Mohammad Norouzi, who spent nearly 7 years as a senior staff research scientist at Google Brain. [via @mo_norouzi] Explore more of our coverage of: AI Funding, Generative AI, Creativity Enhancement. Share this story by email

Foundation Model Development Cheatsheet Launched with 250+ Tools

First impacted: AI developers, software engineers
Time to impact: Short

The Foundation Model Development Cheatsheet has been assembled by a team from 12 AI companies and Universities, and includes over 250 resources for developers. The guide offers an array of resources, including data catalogs, search and analysis tools, and evaluation repositories. These are designed to support data sourcing, auditing, and the assessment of environmental impacts.

The UI for navigating these resources is excellent. Incidentally, Gartner says that codifying best practices into easily accessible tools is one of the keys to democratizing best practices. Sharing success stories and mapping the network of tool providers and power users are the other steps in that model. [Foundation Model Development Cheatsheet: Resources and recommendations for best practices in developing and releasing models.] Explore more of our coverage of: Responsible AI Practices, Foundation Model Development, AI Tool Resources. Share this story by email

Diffusion Models Refined to Require Fewer Steps

First impacted: Stable diffusion researchers, AI model developers
Time to impact: Short

Sander Dieleman from DeepMind wrote a detailed post on progress in AI diffusion models which has led to a reduction in the number of sampling steps required. The paper argues that with a powerful enough model providing training data to a new model, the new model can produce powerful results with far fewer iterative steps. Worth checking out if you are in the image and stable diffusion space, but also metaphorically fascinating in many contexts. [The paradox of diffusion distillation] Explore more of our coverage of: AI Diffusion Models, Distillation Method, AI Research. Share this story by email

Copilot++ Launched with Focus on Editing Code

First impacted: Software engineers
Time to impact: Short

The team at Cursor have launched Copilot++ with the intention of optimizing existing code, rather than solely generating new code. Founder Aman Sanger says "...Copilot tries to solve the wrong problem. It tries to solve the problem of writing completely new code. But as software engineers, most of what we're doing is editing existing code." The company says Copilot++ has the ability to fix a variety of errors, all with the aim of streamlining the coding process.

This made me wonder if the same is true for other uses of AI. As a test, we took each story in this newsletter, started a timer for 3 minutes, freeform brainstormed notes about the story, then put both the story and our abstracted thinking into ChatGPT and asked it to evaluate the consistencies in thinking and opportunities to improve our thinking. Several valuable things came out of those recommendations. [Cursor - Copilot++] Explore more of our coverage of: Copilot++, Code Optimization, AI Tools. Share this story by email

Hugging Face Experiences Infrastructure Disruption

First impacted: AI developers, software engineers
Time to impact: Short

Hugging Face experienced technical issues which temporarily disrupted their services and affected related features on other platforms, such as Wand and LM Studio. The issues prompted the Wand app to issue automatic refunds for any failed training runs, highlighting the interconnectedness of these AI platforms. It made our job a little harder working on this newsletter, too. Hugging Face says the problem has been addressed, implemented temporary reduced rate limits, and their services are back up, with a detailed report about the incident to be released in the future. Couldn’t hurt to think of a personal fallback plan for your projects, too, though, just for resiliency’s sake.[via @huggingface] Explore more of our coverage of: Hugging Face, Technical Disruption, AI Platforms. Share this story by email

That’s it! More AI news tomorrow!