The AI Rundown by Lightscape Partners – 01/22/24

Good morning, happy Monday, and welcome back to this week’s AI Rundown.

Last week, Google DeepMind announced their AlphaGeometry program succeeded in the geometry section of the International Mathematical Olympiad, achieving a breakthrough in geometry algorithms and programs. At the World Economic Forum in Davos, sentiment throughout the week highlights a readiness for the AI wheels to hit the pavement, moving from conceptual thinking to practical action. Further, Arizona State University partners with OpenAI to bring generative AI to the university in an unprecedented move.

Before getting started, do you think you can tell if a photo is AI-generated? Test yourself with this NYT quiz. (I got 6 out of 10).

Top AI Developments of the Week

AI program AlphaGeometry from Google DeepMind succeeds in solving problems from the International Mathematical Olympiad. Link.

The International Mathematical Olympiad (IMO) is considered one of the most prestigious mathematics competitions. Each year, pre-university students from around the world compete over the course of two days to win gold, silver, and bronze medals.

• The program successfully solved 25 out of 30 geometry problems from past IMOs.

• The previous program standard for solving geometric problems correctly answered 10 out of 30, and GPT-4 answered 0 out of 30 correctly.

• The success rate is on par with previous gold medal winners.

• The program found a better, undiscovered solution to a problem from 2004.

• Questions from the IMO can even be unsolvable by experts in some cases.

• The program combines a large language model (LLM) approach with specialized deductive algorithms to introduce and utilize auxiliary geometric objects in proofs.

Despite the progress of AlphaGeometry, the International Mathematical Olympiad consists of problems in several mathematical domains, like algebra and number theory, with geometry occupying a third of the problems on the IMO.

Common themes at the World Economic Forum highlight a shift in AI trends to tangible action in 2024. Link.

At the World Economic Forum in Davos, the focus on AI shifted from theoretical discussions back in 2023 to practical applications in business.

• Companies are moving from mere discussions about AI to implementing tangible strategies, driven by board-level pressure.

• The potential of generative AI to improve productivity and enable new business opportunities was a central theme.

• Various tech firms and consultants, like Tata Consultancy Services and Builder.ai, showcased their AI capabilities, aiming to assist businesses in incorporating AI.

• There's a general optimism that AI will significantly enhance productivity and contribute to global economic growth

Arizona State University and OpenAI team-up in first-of-its-kind partnership. Link.

Arizona State University (ASU) announces a partnership with OpenAI to integrate generative AI technology into its educational and administrative processes.

• An open challenge in February invites faculty and staff to suggest best-use scenarios for OpenAI’s technology.

• Plans to provide full access to ChatGPT for Enterprise to ASU from February 2024.

• Development of a personalized AI tutor for students, focusing on STEM subjects and Freshman Composition for writing assistance.

• Creation of AI avatars as “creative buddies” for studying, such as bots for singing or poetry in biology.

The partnership represents a significant step in the mainstream adoption of AI in higher education. ASU's diverse application of AI across various fields highlights the potential for AI to enhance educational experiences beyond just computer science. The partnership could pave the way for more universities to embrace AI, potentially changing the landscape of higher education.

Hardware + Software

Samsung and Google Cloud team up to integrate generative AI into Samsung Galaxy S24 Series smartphones. Link.

Samsung and Google Cloud announced a multi-year agreement to bring generative AI to Samsung smartphone users.

• Starting with the Samsung Galaxy S24 series, Samsung will integrate Gemini Pro and Imagen 2 on Vertex AI to its devices via Google Cloud.

• Users will be able to use summarization features across Samsung applications, like Voice Recorder, Keyboard, and Notes.

• Samsung will be one of the first to test Gemini Ultra, Google’s largest and most complex model.

• Gemini Nano, an on-device LLM, will be available on the S24 series smartphones as well.

This partnership between Samsung and Google Cloud signifies a major advancement in integrating generative AI into mobile devices, offering a wide range of innovative features and enhanced user experiences.

Mark Zuckerberg announced Meta’s artificial general intelligence efforts will be open source. Link.

In an Instagram Reel on Thursday, Meta CEO Mark Zuckerberg announced the company’s development of an open source artificial intelligence (AGI);.

• The initiative combines the efforts of Meta's FAIR and GenAI research teams.

• Meta has started training Llama 3, its next generation of primary generative AI models.

• Zuckerberg foresees the use of Ray-Ban smart glasses as a medium for interacting with AI.

• This move contrasts OpenAI’s closed source model.

Meta's initiative to develop open source AGI, led by Mark Zuckerberg, marks a pivotal moment in AI research, aiming to democratize access to advanced AI technologies. This move not only sparks a potential shift in the AI landscape but also raises important considerations about the ethical and practical aspects of widespread AGI deployment.

Venture

Recraft, an AI graphic design tool, raises $12M to build its foundation model. Link.

The London-based startup announced $12 million in Series A funding.

• The round was co-led by Khosla Ventures and former GitHub CEO Nat Friedman.

• The funding will go toward developing a proprietary graphic design foundation model and enhancing AI-powered design tools for brand consistency and rapid iteration.

• The platform launched 8 months ago and has amassed over 300,000 users.

• The valuation is unknown at this time.

Ethics in AI

A report published by the IMF claims AI will impact 60% jobs and worsen inequality. Link.

The IMF report emphasizes the profound impact AI will have on global employment, highlighting the potential for increased productivity alongside significant job displacement and rising inequality.

• AI predicted to affect 60% of jobs in advanced economies (U.S., UK, EU) and 40% globally.

• AI likely to worsen inequality, impacting job displacement and economic disparity.

• Same proportion at risk of negative impacts, including higher-wage, white-collar jobs.

• The report assess readiness of 125 countries for AI adoption, with Singapore, U.S., and Denmark in the lead in preparedness.

• Readiness factors include: digital infrastructure, human capital and labor market, AI innovation and economic integration, and regulation and ethics.

• AI threatens jobs requiring judgment, creativity, or complex data interpretation.

The report underscores the need for comprehensive policies and readiness, especially among vulnerable demographics and less-prepared economies. As AI continues to evolve, strategic planning and international cooperation are crucial to harness its benefits while mitigating its challenges.

Fairly Trained launches to certify ethically trained gen AI tools. Link.

Fairly Trained, a new non-profit organization, has been established to certify generative AI tools trained on data obtained with explicit consent from creators. The initiative addresses concerns about AI models trained on web-scraped data without the creators' permission.

• Fairly Trained offers a "Licensed Model (L) certification" for AI providers, distinguishing those who obtain creators' consent for training data.

• The non-profit charges a fee for certification, ranging from $150 plus $500 annually for smaller companies to $500 plus $6,000 annually for companies with over $10 million in revenue.

• Founder and CEO Ed Newton-Rex explains the fees are to cover operating expenses for the non-profit.

With several on-going lawsuits between large publishing companies, creators, and others against industry leaders like OpenAI, it’s hard to predict the restrictions that may be placed on web-scraped data training. But, Fairly Trained’s initiative highlights the growing sentiment against the current training model.

Data ‘poisoning’ tool, Nightshade, sets out to sabotage generative AI models using their work for training. Link.

From the University of Chicago’s Glaze Project, Nightshade 1.0 allows artists to disrupt genAI models training on their work.

• The software makes invisible pixel-level changes to images that trick models and corrupt image output.

• The software is out now for Apple Silicon Macs and Windows PC.

Data ‘poisoning’ may become more prevalent as creatives fight against data collection on the public domain.

Thank you for reading the AI Rundown by Lightscape Partners. Stay tuned for the latest updates in the world of Artificial Intelligence!