BS”D

In the news this week:

• 🚨 OpenAI's GPU Shortage

• 🔥 UAE's ChatGPT Competitor

• 🤖 AI's Healthcare Innovation

• 🛠️ $100M Investment in Legal AI

📰 AI in the News

Starved for GPUs

Last week, ChatGPT maker OpenAI sat down with a small group of developers to talk roadmap. It’s no longer online, but a single message came through loud and clear: we’re hitting the GPU shortage, and hard.

(GPUs - graphics processing units - are the computer cards that make everything from crypto to gaming to AI models possible. In the latter case, AI models need GPUs to run - there’s a GPU spinning every time you talk to ChatGPT.)

But while OpenAI has ambitious plans - a GPT-4 that is faster, processes/generates more text, and eventually works with images - all of that is only possible with more GPUs.

They’re not alone. “2nd place” startup Anthropic (who recently raised $450 million from investors including Google) still hasn’t opened their chatbot to the public, in part due to the shortage.

All eyes are on NVIDIA (dominant GPU maker; stock 3x’d to nearly $1T since Oct lows amidst the AI boom) to ramp production, and for all types of challengers, including AMD and new startups, to quickly catch up over the next years.

The UAE Comes Out of Nowhere

There’s a race going on to build an open-source (fully open, free-to-use) AI model that can compete with OpenAI’s GPT-3.5 and GPT-4 models (closed-source, pay to use).

The stakes are fairly high - most notably 1) the ability to run AI models on your laptop and 2) avoiding an “OpenAI tax” that would apply to nearly everyone if OpenAI runs away with it.

In a laborious effort to piece together OpenAI’s code and training data/process, we have a new prince topping the open source leaderboards. Earlier this month, the UAE (yes, that one) suddenly announced Falcon, the new leading open source language model.

Completely open for commercial use, it’s not only a more permissive but also higher-performing model than LLaMA, the previous leader built by Meta.

We’re still quite far from open source catching up to OpenAI, but it’s completely feasible that we get an open source “good enough” model in the next few months. (link)

More Stories

• A Salesforce report found that 51% of marketers are using generative AI at work, with the most popular use case being content and copy creation. (link)

• A top IMF official sees imminent “substantial disruptions in labor markets” from generative AI. (link)

• ETF providers are plowing money into generative AI companies and seeing big inflows in 2023. (link)

• Apple commits to mental health with tracking on watch and iPhone. (link)

• Google Cloud collaborates with Mayo Clinic to transform healthcare with generative AI. (link)

• Helpful article explains the concept of large language models. (link)

💡 Industry Insights

🏥 Healthcare

AI/ML in drug discovery / detection: anti-superbug antibiotics, brain damage, colorectal cancer, multiple sclerosis

AI in the clinic: Should AI come to your doctor's office? OpenAI's co-founder thinks so. Primary/urgent care startup Carbon Health is rolling out hands-free AI charting which 1) transcribes the appointment, 2) combines the transcript, lab results and doctor notes and 3) auto-generates a summary. Time to generate charts: <4 minutes vs 16 minutes manually. Manual charting is tedious and one of the contributing factors of burnout among technicians. 75% less time saves each clinician over ten minutes on average per patient. More than 125 clinics will get access. (link)

AI in prior authorization (PA): McKinsey believes 50-75% of prior authorization can be automated with AI. The data is messy, unstructured and spread across numerous sources - terrible for humans, but potentially perfect for AI. Google also seems to see the same opportunity, announcing their Claims Acceleration Suite earlier this year. (link)

🏙️ Real Estate

Using AI for market analysis: AI/ML algorithms can now analyze millions of data points in seconds to help predict property prices and identify investment opportunities. By examining various factors such as business activities, redevelopment plans, historical trends, news, crime statistics, and business registration records, AI can forecast future market trends and highlight potential investments. We expect a wave of products to emerge in this category, like DealMachine’s AI assistant that provides real estate investors with a competitive edge. (link)

👩‍⚖️ Legal

Spellbook closes $11m in funding: Ask any venture investor for a slam-dunk generative AI use case win, and they’ll probably start with legal. The newly cash-infused Spellbook is helping smaller firms and in-house counsel draft contracts, highlight aggressive/missing terms and offer negotiating positions. The other hot company in the space: Harvey.ai, which is targeting “elite law firms”. (link)

Thomson Reuters leans in on AI: In early May, Thomson Reuters said they’d invest ~$100m/year in generative AI capabilities, including embedding functionality into products like Westlaw and Drafting Assistant. But their corp dev and venture teams are also looking for some market heat - they invested in Spellbook (above), and word is that they’re knocking on some select doors looking for new targets (speculation ours). (link)

📈 Venture Deals of the Week

• BenchSci raised $95 million to help build its AI drug discovery platform. (link)

• Infogrid raised $90 million for its AI-driven startup that collects and analyzes data to monitor air quality, occupancy and energy consumption in buildings. (link)

• FlowX AI raised $35 million for its platform helping enterprises port software into a single location to run applications. (link)

• Orbital Witness, a London-based generative-A.I. platform for real estate lawyers, raised £7.5 million ($9.33 million) in Series A funding. (link)

• Gan AI, an AI platform for creating personalized videos, raised $5.3 million in seed funding. (link)

🛠️ Latest AI Tools

• Rask uses AI to dub videos in different languages while still sounding like the original person. (link)

• Micro1 is an AI vetting tool that quickly assesses technical talent. (link)

• Reimagine is a photo-editing app that preserves and enhances new and old images. (link)

• Siit is an AI assistant that answers employees’ questions based on internal knowledge bases. (link)

• Recraft is an AI generator for vector art, illustrations, and 3D images. (link)

• StonksGPT is an AI search tool for stocks and finance data. (link)

• Rep uses behavioral AI to deliver personalized shopping experiences. (link)

• TaxGPT uses AI to automate tax filing. (link)

• InterviewMe AI helps candidates practice interviewing with an AI. (link)

Share & Subscribe to AI Future

Did someone send you this newsletter? Subscribe Here.
Know someone who’s interested in learning about AI? Copy and paste this link to others: https://tiferes.beehiiv.com/subscribe