S3T Panorama: Emerging Tech

Signals & Data Points

Legal test: can Generative AI products be sued for defamation?

In January 2024 OpenAI filed a motion to dismiss based on the notion that the user knew the AI output was false. The judge refused to dismiss the case. As of Nov 2024 OpenAI was still seeking resolution on the matter.

Shift from generic to specific AI

This shift represents a return to leveraging specialized, proprietary datasets to train AI models that better align with specific use cases and industries. Drivers:

• More widespread recognition of the limitations of models trained on generic public data.
• Public foundation models like ChatGPT or Gemini excel at general tasks, while proprietary data allows organizations to develop models with deeper domain expertise and precision - and overcoming the limitations of generic, internet-based training data.
• These opportunities have been neglected/underprioritized by some over the past 18 months thanks to the over-hyped expectations around Generative AI trained on public data.

Examples below.

Evo - AI led discovery of new genomes

AI + Crisper = Evo, the first biological foundation model trained on DNA at scale. Evo's large context window allows it to analyze entire genomes. Is generating entirely new CRISPR systems unknown in nature, and expected open up new treatments for a range of diseases.

AI-Imitation Lea Surgery: Robots Learn Precision from Human Surgeons' Videos

Imitation learning continues to advance: Johns Hopkins and Standard researchers used imitation learning to successfully train a da Vinci surgical robot to perform surgical procedures: manipulating a needle; lifting body tissue, and suturing at unprecedented levels of precision.

The method employs imitation learning combined with a machine learning architecture similar to ChatGPT, adapted to interpret robotic kinematics instead of text. By training on video data from da Vinci surgical robots worldwide, the model learns to predict and execute relative robotic movements needed for precise surgical tasks, addressing challenges of imprecision in existing systems.

This approach is a step forward as it demonstrates how leveraging vast archives of human-performed surgical data can enable robots to generalize to new environments and perform with human-level skill.

The energy, processing and infrastructure needs of AI are serving as an economic driver for multiple other industries.

• AI boosting telecom - telecom player Lumens gets a new lease on life.
• Big tech firms are investing in nuclear energy to meet escalating energy demands of AI
• FlawlessAI is transforming film editing and localization - the tools allow edits to dialogue, or lip synching actor dialogue to speak in different languages.

Key conversation: How the deregulation agenda will impact emerging tech

The incoming Trump administration's aggressive deregulation agenda with its GOP House & Senate majority will have significant impacts on emerging tech:

• AI - "expect a change in direction toward less regulation...less compulsory AI safety measures" says Nick Reese Emerging Tech Policy, DHS
• Cyber security may get mixed treatment pulled between the desire for less regulation vs. the undeniable need to strengthen protections for core infrastructure.
• Use of social platforms to spread misinformation. The concept of free speech needs a refreshed understanding in light of technology platforms that give owners and privileged participants bigger megaphones, the ability to prioritize self serving narratives, and decide what the conversation will be. One fascinating twist: The Onion buys InfoWars in a bankruptcy auction with the approval of the Sandy Hook families who sued Alex Jones/InfoWars into bankruptcy.
• Less accountability for spyware companies. Recommended reading: Carnegie Endowment Explainer on the spyware industry and why it thrives. See also this dataset: Global inventory of commercial spyware and digital forensics technology and US State Department Sept 2024 note on Countering Misuse of Commercial Spyware.