It’s time for some Friday Big Ideas.
TechCrunch talks about the challenges of reviewing AI systems and why it is crucial to attempt reviews despite the difficulties. The rapid pace of AI model releases, the complexity of these models as platforms with multiple services, and the lack of transparency from companies make comprehensive evaluation impossible. However, qualitative analysis serves as a real-world counterweight to industry hype. The article highlights the limitations of synthetic benchmarks and the emergence of unknown qualities in AI models. TechCrunch aims to provide subjective judgments based on hands-on experience and a general sense of the capabilities of AI models through a systematic testing approach, and the article walks through their approach.
MIT Technology Review dives into how Adobe’s approach to generative AI, exemplified by its image-generating model Firefly, stands out as an example of building responsible tech without scraping copyrighted data from the internet. By training Firefly on content with explicit licenses and compensating creators, Adobe aims to recognize and fairly compensate human labor. This approach protects creators’ rights, helps combat misinformation, and promotes content moderation. The integration of generative AI tools into existing products, such as Photoshop, has been well-received and has attracted new customers to Adobe.
Markus Zim of Workato writes that Generative AI has the potential to revolutionize business processes by allowing companies to run entire knowledge processes through natural language prompts. These processes, dynamically created by AI brains, will integrate with applications, data sources, and human experts to generate insights and increase efficiency. Human validation steps remain crucial in this paradigm, as LLMs still require human supervision. Well-architected data flows, and end-to-end process orchestrations will be essential for successful implementation, offering immense promise for businesses.
FIDO Alliance CEO Andrew Shikiar believes that our dependence on passwords will be gone in the next several years, as passkeys are gaining traction as more big companies show support. Passkeys use public key cryptography and possess-based authentication, making them more secure against phishing attacks. The FIDO Alliance aims to have half of the top 1000 sites supporting passkeys by 2026. While passwords won’t be eliminated, passkeys are expected to become the dominant form of sign-in in the near future.
And I wanted to go off beat with a moving article in the New York Times. Madeline de Figueiredo writes in Modern Love about her experiences cloning her dead husband’s voice with AI for his 27th birthday. We talk a lot about the business side of tech, and with this I didn’t want to lose sight of the human side. She writes of her longing for a genuine conversation with her husband and reflects on the limitations of technology in dealing with grief.
Why do we care?
I included the TechCrunch perspective because I suspect many providers are pondering how to evaluate models themselves. I appreciated the transparency.
I’m going to linger on the human side of technology. Want better business outcomes? Factor in the emotional component of technology implementation. There are some great business concepts here… but loyal customers are also driven by emotion.