"AI" Contact Lenses for Oncologists!

Plus: AI has nothing on Dwight Schrute

April 14, 2024

Hello Tensor Black Fans!

Are “Smart” Contact Lenses the next UI (User Interface) for AI?

Xpanceo thinks so. They landed $40 million to chase the dream. TechCrunch reviewed their beautiful pitch deck this week and caught our eye. We’re excited to see what they’re building.

Level up with these bites today:

  • Software is Eating Cancer

  • Better Drugs through AI Prompts

  • An LLM in 1000 Lines of Code

  • LLMs are worse than Dwight Schrute

  • “Smart Contact Lenses” nabbed $40M

Monoclonal antibodies and personalized mRNA vaccines are a hot topic at this year’s annual meeting of the American Association of Cancer Research.

Why it Matters

Marc Andreessen famously said that “software is eating the world.” Well, apparently that now includes cancer.

The A, T, C, G letters of DNA are the source code of life. AI is helping us learning what those sequences of letters mean, how they produce proteins, and their shape within the body. MABs and are essentially programmable DNA antibodies to boost our immune system. mRNA use DNA sequences found in nature to replicate antigens, activating our normal immune response.

Insilico Medicine announced a new and improved drug by prompting their AI “Chemistry42” with an existing drug, then asking for a new one.

Source: InSilico Medicine

Details

  • PTNP2 is like a supervisor protein that helps control cell growth, but cancer cells can silence it, leading to uncontrolled growth.

  • By inhibiting PTPN2, drugs disrupt the signaling pathways that cancer cells rely on, essentially throwing a wrench into cancer machinery.

  • This disruption can lead to a slowdown or halt in cancer cell growth and may also make the cancer cells more susceptible to other treatments like chemotherapy or immunotherapy.

  • In Silico fed the molecular structure of an existing PTPN2 inhibitor to a generative AI platform and produced a more effective version.

Why it Matters

Generative AI is becoming a tool of science. An AI can be taught the “language” of molecules, then given a prompt in that language, producing new and improved versions. AI radically reduces the search space for new molecules, cutting drug development time in half (we still need to test and trial).

Andrej Karpathy spent his vacation writing an LLM from scratch in a startling 1000 lines of C, then trained it on a single GPU.

Source: Andrej Karpathy

Why it Matters

Thousands of researchers spent decades chasing human-level AI, piling on features to a massive code base that serves thousands of approaches to AI. That’s an enormous pile of technical debt that large companies and researchers just aren’t motivated to fix (it ain’t broke, so…).

Startups and AI geniuses are taking the time to rewrite AI from scratch, heralding a new era of highly efficient, compact algorithms with new chips. Andrej’s llm.c currently runs on a 1996 era computer!

Current LLMs can only accomplish 2-20% of the office tasks we do everyday with a mouse, keyboard and screen (see the code here). Dwight Schrute, your job is safe.

Why it Matters

AI has demonstrated productivity improvements of 40% and higher. Where we really need help is using AI for the daily office grind, with very little integration cost. That’s coming. Researchers in Hong Kong created “OS World” for training and benchmarking AI systems in real world tasks, advancing state of the art.

Xpanceo is betting big on turning us all into cyborgs with smart contact lenses, securing a cool $40 million to make our sci-fi dreams a reality.

Why it Matters

The future interface for AI will be live, streaming, bidirectional video. Video streams will be augmented with realistic imagery, music, sounds, captions and more. Startups are now pushing the frontier of AI interfaces, experimenting with everything from contact lenses, to smart pins, earbuds, and more. See their pitch deck that snagged $40M, reviewed and shared by TechCrunch this week.