The Generative AI Reality Check

Welcome to the first 2024 edition of Tomorrow Bytes! As we embark on a new year, we are evolving our newsletter to bring you more in-depth reporting and analysis on the rapid developments in AI. This week, we unpack critical industry headlines - from revolutionary medical discoveries to legal battles defining AI's future. We will go deeper on pivotal stories moving the field, including OpenAI's copyright lawsuit and emerging techniques to optimize ChatGPT. Other highlights showcase AI's swelling momentum and complex implications for governance worldwide. From policy debates to personal implications, our featured articles track generative AI's accelerating pace alongside the global high-stakes decisions facing consumers, executives, and regulators. Read on for the highlights in this new Tuesday edition as we monitor the space more closely in 2024 and beyond!

📌 Byte-sized Breakdowns

• Apple is on the brink of their turn in the AI revolution for iPhone users, developing cutting-edge on-device AI technologies capable of creating animatable avatars and running large language models, enhancing user interaction and intelligence on mobile devices​

• Revolutionizing affordable housing, Simply Homes raises $22 million to transform blighted neighborhoods into havens for low-income families, the elderly, and disabled individuals.

• AI health coaches, utilizing wearable data, are set to revolutionize personal health management by offering customized, data-driven health advice, enhancing individual health outcomes through proactive and personalized interventions.

• Southern California is now home to the world's first fully autonomous, AI-powered restaurant, CaliExpress by Flippy, where robots take charge of cooking burgers and deep-frying French fries, showcasing an innovative integration of artificial intelligence in the culinary industry​

• A groundbreaking AI model has been developed to predict future waves of COVID-19 variants, capable of detecting approximately 73% of variants that may cause significant outbreaks in a country within three months, showcasing an advanced step in managing pandemic responses through technology

• The iconic Steamboat Willie version of Mickey Mouse, having entered the public domain, has sparked a surge in AI-generated art, inciting both creative expression and a deeper exploration of copyright boundaries, reflecting a significant shift in the intersection of popular culture, digital technology, and intellectual property rights.

🔦 Spotlight Signals

AI Is Telling Bedtime Stories to Your Kids Now

Artificial intelligence can now tell tales featuring your kids’ favorite characters. It’s copyright chaos—and a major headache for parents and guardians. Generative AI allows parents to create personalized Bluey stories, delighting kids but appropriating IP. While personalized tales resonate, plots are repetitive and dull. Stricter constraints and oversight are needed to ensure safety, quality, attribution, and diversity. AI’s bedtime offerings remain legally questionable, ethically murky, and creatively lackluster. However, the technology could enhance story time if harnessed responsibly. [Dive In]

AI Unlocks First New Antibiotic in Decades

AI has unlocked the first new antibiotic class in decades, aided by machine learning that screened 12 million candidates and identified a potent treatment for drug-resistant superbugs, specifically MRSA. Initial tests have demonstrated a tenfold reduction in bacterial levels in infected mice, showcasing AI's potential to accelerate medical discoveries. This achievement, while promising, also highlights the importance of model transparency and raises concerns about equitable access and overuse risks as AI continues to shape drug development. [Dive In]

Sam's OpenAI Secures $100B Valuation

OpenAI seeks over $100 billion valuation in new funding, cementing its status as the world's second-most valuable private startup. The AI darling aims to capitalize on the ChatGPT frenzy and transform interactions. However, leadership turmoil last month caused investor uncertainty, which is now seemingly settled with Sam Altman as CEO. As rivals rush to compete, OpenAI chips away at the next ventures in chips, products, and services. With sights on the future, its valuation mirrors optimism despite ethical quandaries. Rapid progress prompts open-ended funding as institutions jockey for AI leadership. [Dive In]

How Not to Be Stupid About AI

Meta's chief AI scientist, Yann LeCun, scoffs at dystopian scenarios but agrees machines will eventually surpass human intelligence. He advocates open-sourcing models to speed progress and prevent corporate control. LeCun sees chatbots and generative AI transforming how we interact digitally. But despite rapid advances, he argues current techniques remain inefficient compared to biological learning. LeCun is pioneering "objective-driven AI" to instill safety. He believes future AI assistants will amplify our abilities if properly constrained. Music and art may one day match human creativity technically but lack deeper resonance. For LeCun, AI brings abundant potential benefits if guided prudently. [Dive In]

💼 Business Bytes

Brace for Impact: Generative AI as the Next Internet-Scale Change

Few technologies in history can claim generative AI’s potential for market transformation. Early text, image, video, and audio generation benchmarks already showcase versatility exceeding human creatives on select metrics. Yet, as models now transition from research to revenue generation, hard questions on infrastructure needs, job impacts, bias mitigation, misuse prevention, and environmental sustainability abound.

We are merely in the first innings of a multi-year game. Much like the commercial internet era, present use cases likely pale against the unfolding breadth. However, winners will balance rapid experimentation with robust governance.

Those who instill such resilience without losing agility are poised to dominate. Business models remain fluid, but the foundations implemented today will determine who outlasts the hype cycle. Ultimately, generative AI’s next chapter needs protagonists who innovate responsibly - the binary choice between progress and ethics is one we can no longer afford.

The six questions dictating AI’s future explored in this piece….

1. Will we ever mitigate the bias problem? This question addresses the pervasive issue of bias in AI, particularly the biases inherent in real-world data used for training AI models. The challenge is to find ways to mitigate these biases, which are deeply embedded in the data and, thus in the AI models themselves.

2. How will AI change the way we apply copyright? The legal and ethical implications of AI in relation to copyright law are being questioned. This includes the impact of generative AI on the works of artists, writers, and coders and how copyright laws need to evolve in response to AI-generated content.

3. How will it change our jobs? This question explores the impact of generative AI on the workforce, including the potential risks to both white-collar and blue-collar jobs and how AI might transform various professional roles.

4. What misinformation will it make possible? The potential of generative AI to create convincing fake content (deepfakes) and spread misinformation at an unprecedented scale raises concerns about its impact on public opinion, politics, and social trust.

5. Will we come to grips with its costs? The human and environmental costs of developing and deploying generative AI are significant. This includes the ethical implications of the labor used in training and refining AI models, as well as the environmental impact of the energy-intensive processes involved in AI development.

6. Will doomerism continue to dominate policymaking? The question here is whether the fear of AI’s potential negative consequences will overshadow its benefits and lead to overly cautious or restrictive policies that could stifle innovation and beneficial uses of AI.

[Dive In]

Tomorrow Bytes’ Take…

1. Generative AI's Broad Impact: Like the early internet, generative AI creates a significant shift in technology and society, with widespread excitement and uncertainty about its applications and implications.

2. Potential for Market Disruption: The technology is still in its early stages, akin to the dot-com era, suggesting that while many startups may fail, there's potential for significant successes and innovations.

3. Unpredictable Misuse and Risks: The true potential and risks of generative AI will only be understood through widespread use, paralleling the internet's evolution, bringing revolutionary benefits and serious challenges like cyberbullying and misinformation.

4. Bias and Ethical Concerns: AI's inherent biases, reflecting real-world data, present ongoing ethical and operational challenges, with efforts underway to mitigate these through various techniques.

5. Copyright and Legal Challenges: The technology is pushing the boundaries of copyright law, leading to lawsuits and new legal precedents that will shape the industry.

6. Workplace Transformation: Generative AI is poised to transform jobs across sectors, automating tasks, creating new roles, and changing existing ones.

7. Misinformation Risks: There's a growing concern about using generative AI to spread misinformation, with potential political and social implications.

8. Environmental and Human Costs: The development and deployment of generative AI come with significant human and environmental costs, calling for more sustainable and ethical practices.

9. Policy and Regulation Challenges: The technology is sparking debates around regulation and policy, with different approaches emerging globally.

10. Uncertainty About AI’s Ultimate Role: Despite its rapid growth and hype, there's still uncertainty about generative AI's most effective and transformative uses.

☕️ Personal Productivity

Figma Goes Beyond Design, Sets Sights on AI-Powered Workspace

Once relegated as productivity software for design professionals, Figma’s rapid ascent has defined expectations of workplace collaboration tools. Powered by the visionary introduction of the AI-enabled FigJam whiteboarding environment, the company has achieved breakout appeal across functions. As non-designers now comprise over 60% of users, FigJam’s versatile co-creation feature set reimagines meetings as creative workshops for the whole organization.

In particular, the integration of assistive AI transcends transactional efficiency. Sophisticated yet playful suggestions during brainstorming phases unlock team creativity on-demand. Meanwhile, automated administrative workflows like note transcription, summary generation, and action item identification remove friction. As Figma further obscures barriers between design thinking and broader business strategy, its influence increasingly intersects with frontrunners pioneering the future of work, IP giants, and niche disruptors alike.

While the collapsed acquisition talks with Adobe initiated reflection around Figma’s ceilings as an indie player, rapid success post-deal has affirmed the conviction in its autonomy. As enterprises seek inspiration for digital transformation within and beyond IT, expectations now center on Figma, inaugurating a new generation of collaborative productivity solutions mainstreamed through AI augmentation. Yet, as competition mounts, sustained differentiation rests on nurturing an open culture that stays attuned to evolving user needs. For Figma, the next bold bets start and end with its community.

[Dive In]

Tomorrow Bytes’ Take…

1. Figma's Focus on AI-Enhanced Meetings: Figma is innovating in artificial intelligence to improve the efficiency and creativity of meetings. This indicates a strategic pivot towards leveraging AI for design and enhancing overall workplace productivity and collaboration.

2. Expansion Beyond Designers: Figma’s user base now includes many non-designers, highlighting its evolution from a niche design tool to a more versatile collaboration platform.

   • Adobe’s withdrawn acquisition offer was valued at $20 billion, with a $1 billion breakup fee.

   • Two-thirds of Figma’s user base are now non-designers.

   • Figma's valuation skyrocketed after the launch of FigJam in 2021.

3. FigJam AI's Playful and Practical Features: Incorporating playful elements like icebreakers and brainstorming exercises alongside practical applications like summarizing notes and suggesting next steps demonstrates a balance between engaging user experience and functional utility.

4. Importance of Human-AI Collaboration: The need for human oversight in AI-generated outputs, as indicated by Code and Theory’s approach of using AI for 80% of the work, underscores the ongoing importance of human judgment in the AI-assisted workplace.

5. Figma's Strategic Independence: Despite the withdrawal of Adobe’s acquisition offer, Figma continues to pursue its AI innovation independently, suggesting confidence in its standalone growth and potential competition with Adobe.

🎮 Platform Plays

All Eyes on OpenAI as GPT Store Launch Signals Strategic Shift

OpenAI’s upcoming launch of the GPT Store platform signifies an ambitious pivot towards an integrated AI ecosystem model. By simplifying the development and distribution of AI applications built on Generative Pre-trained Transformer (GPT) models, the company now cedes greater control to external developers in exchange for reach and innovation velocity.

Our latest research suggests that platform-based approaches will dominate the next wave of AI commercialization. Fostering openness and collaboration between user communities taps into more creative applications as barriers to experimentation lower. Further, the aggregation of use cases provides proprietary data on niche demands per industry, allowing more tailored product development.

However, successfully governing decentralized ecosystems proves non-trivial across SaaS and mobile application marketplaces. Beyond brand security, quality control, and policy enforcement, sustainable monetization depends on correctly calibrating commission structures to balance developer incentives against profitability.

Additionally, the false start of the GPT Store in Q4 2022 led to temporary leadership turmoil, underscoring the importance of stability in steering nascent platforms. As competitors race to launch alternative AI marketplaces, achieving escape velocity requires technological robustness and maintaining strategic clarity despite external shocks.

Those that quickly establish network effects through strong core platform-model execution can lock in dominant positions. For OpenAI, the GPT Store signifies their opening gambit to own the next generation of AI application infrastructure. Though ambitions run high, the playbook remains a work in progress across the industry.

[Dive In]

Tomorrow Bytes’ Take…

1. Strategic Shift to Platform Model: OpenAI's transition from a mere AI model provider to a platform with the GPT Store launch signals a strategic pivot. This shift underlines OpenAI's ambition to create AI technologies and foster an ecosystem where developers can contribute and benefit.

2. Simplifying AI Development: The GPT Store democratizes AI app creation by allowing developers to build GPTs without extensive coding experience. This ease of development could significantly lower the barrier to AI innovation and broaden the developer base, leading to more diverse and creative AI applications.

   • Currently, over 50,000 public GPTs have been created, and the count doubles every 18 days.

   • The community speculates that GPTs will skyrocket past 20,000 once the OpenAI GPTs Store officially launches.

3. Leadership Impact and Stability: The delayed launch, linked to CEO Sam Altman's temporary ousting, reiterates the importance of stable and consistent leadership in guiding a tech company's direction and maintaining investor and employee confidence.

4. Policy and Compliance Emphasis: The requirement for developers to comply with updated usage policies and GPT brand guidelines before listing their GPTs in the store emphasizes OpenAI's commitment to maintaining quality and ethical standards in its ecosystem.

5. Uncertainty in Monetization Strategy: The lack of a clear plan for monetizing the GPTs in the store reflects ongoing uncertainties in the AI market regarding sustainable business models. This could be a critical factor in shaping the store's long-term success and the financial viability for developers.

Predictions

1. Discoverability will be more important than monetization - the biggest challenge right now is finding relevant GPTs, not making money from them. Better discovery tools in the store will increase usage.

2. The tools most easily monetized will be coding and no-code tools - people are more likely to pay for something that lets them do something entirely new versus just improving on something they can already do.

3. Early adopters will be students looking for study aids and digital marketers who rely heavily on words/copy.

4. GPTs will be much more relevant for professionals than for personal use - they embody common workflows that professionals repeat often.

5. The store's presence will unlock creativity as people try different things to make money, leading to an explosion of weird and novel use cases, some of which may stick.

6. Specialized vertical GPTs will emerge - We'll see industry-specific GPTs tailored for domains like law, medicine, engineering, etc., that incorporate niche knowledge.

7. Free alternatives will appear - Services like Bard may introduce free tiers to compete for market share.

8. Novel applications will emerge - Unexpected and innovative use cases will likely arise that surprise even OpenAI.

🤖 Model Marvels

The Future of ChatGPT is Structured Data: Mastering Inputs and Outputs

The rapid proliferation of generative AI across industries has ushered in urgent questions on optimization for enterprises. While much commentary focuses on content moderation, prompt formulation methodology remains poorly documented yet pivotal. Through systematic testing, it becomes evident that several key levers can amplify both the quality and alignment of model outputs.

Specifically, exploiting the ChatGPT API alongside structured data frameworks unlocks transformative performance enhancement. Commercial users actively employing Pydantic and similar tools to enforce JSON schemas note dramatic improvements in response relevance, coherence, and security. Furthermore, intricate system prompts stretching over 15 lines help concretize the persona and domain space for the model, reducing hallucinated or ungrounded outputs by over 37 percent.

The combined techniques ultimately showcase generative AI’s flexibility for customer-facing content creation and as a novel tool integrated directly into development stacks. Like traditional SaaS embedded offerings, ChatGPT can interface bi-directionally with external functions and data streams once configured through protocols like function calling. Consequently, the addressable use cases expand exponentially across domains as diverse as data analysis, system troubleshooting, and even low-code application scaffolding.

Through methodological fine-tuning, enterprises now stand to tap generative AI for a multiplicity of workflows ranging from interactive documentation to automated testing suites. Yet recipes cannot remain proprietary; the industry stands to benefit profoundly from open and detailed best practices on prompt engineering. If transformations at scale are to be realized, such collaboration remains non-negotiable.

[Dive In]

Tomorrow Bytes’ Take…

1. The ChatGPT API, especially the paid version, significantly enhances control over output, allowing for more sophisticated results and customization.

   • The API allows for the maintenance of specific JSON schemas, significantly impacting the model's output and data handling capabilities.

2. System prompts enable users to define the "persona" of ChatGPT, effectively influencing its text generation behavior and output quality.

   • System prompts can be intricate, with some containing over 20 lines to ensure desired outputs.

3. Implementing structured data with ChatGPT, particularly using JSON schemas and tools like Pydantic, simplifies input/output processes and supports more complex data handling​​.

4. ChatGPT's function calling feature represents a significant advancement, allowing for more reliable connections between the model's capabilities and external tools or APIs.

   • ChatGPT's function calling responded to the popularity of libraries like LangChain and AutoGPT, which popularized similar flows.

5. Using structured data and system prompts together can lead to more precise, efficient, and secure outputs, minimizing risks like prompt injection attacks.

   • The combination of structured data techniques can significantly optimize ChatGPT's responses, as seen in examples like the palindrome detection problem​​

6. Applying Pydantic and similar tools in generating JSON schemes demonstrates the growing intersection of AI and traditional software development practices​ and a response to its robustness in parsing and validation.

🎓 Research Revelations

Beyond the Human Eye: AI Reveals Invisible Detail in Raphael's Art

Forging an unlikely union between fine art and artificial intelligence (AI), researchers have developed a pioneering machine learning algorithm that authenticates works attributed to High Renaissance stalwart Raphael with startling 98 percent accuracy. This specialized model meticulously inspects color palette, shading, and brushstroke style, unearthing previously undetected technical consistencies across the master’s portfolio.

The implications are profound. Museums and private collectors alike are struggling under the swelling weight of fraudulent artwork flooding global markets, with estimations that counterfeit comprise up to 20 percent of circulating fine art. The associated loss in value is steep, siphoning as much as $6 billion annually. Further exacerbating matters is the highly specialized skill set required to authenticate pieces mid-evaluation, which happens to be shared by an exceedingly rare breed of historians, connoisseurs, and conservationists rapidly reaching retirement age.

Enter AI as a timely solution to detect forgeries and verify authenticity to an unrivaled degree. While no technological solution can entirely replace hard-won human expertise built over lifetimes of scrutinizing technique and provenance documentation, an early partnership between data scientists and art experts promises to strengthen attribution processes substantially.

With 98 percent accuracy distinguishing authentic brushwork coupled with the ability to process thousands of paintings annually, this form of AI works collaboratively alongside human authenticators, bolstering efficiency and providing a second layer of invaluable insight. The future of provenance may indeed be joint custody between human and machine.

[Dive In]

Tomorrow Bytes’ Take…

1. AI's capability to discern minute details beyond human perception is groundbreaking in fields like art history, showcasing its potential to unearth hidden aspects of historical artifacts.

2. Integrating deep feature analysis and machine learning illustrates AI's versatility in various disciplines, including art authentication, signifying a blend of technology with traditional academic fields.

   • The AI model was a modified version of Microsoft's ResNet50 architecture and a traditional machine learning technique, a Support Vector Machine.

3. This AI application underlines the importance of collaboration between technology and human expertise, highlighting AI as a tool for enhancement rather than replacement.

🚧 Responsible Reflections

AI's Ethical Quandary in Deepfake Dilemma

As concerns rise over AI-generated fake nudes, experts are calling for regulations and heightened public awareness. This movement signals a crucial juncture in addressing the ethical implications of AI's capabilities, particularly in safeguarding privacy and preventing digital manipulation. [Dive In]

Tomorrow Bytes’ Take…

1. Prevalence of AI-Generated Fake Nudes: The use of AI to create fake nude images of individuals, especially women, and students, is increasingly common, with incidents reported in schools and public figures being targeted.

   • In 2023, 143,000 new AI-generated porn videos were added across popular websites, surpassing the total from 2016 to 2022.

   • Fake nudes have increased by 290% since 2018 on top websites hosting AI-generated porn photos.

   • 26,800 U.S. victims of “sextortion” campaigns in 2023 through September, marking a 149% rise since 2019

   • There was a 54% increase in deepfake pornographic videos uploaded in the first nine months of 2023 compared to all of 2022.

2. Lack of Federal Regulation: There is no federal law in the United States specifically addressing the creation and distribution of pornographic deepfakes, leaving a legal gap in handling these cases.

3. Impact on Victims: The psychological and social impact on victims of these deepfakes is profound, yet there's limited understanding or research on the full extent of this impact.

   • 90-95% of deep fake videos involve non-consensual pornography of women

4. Complexity in Legislation: Creating effective federal legislation against pornographic deepfakes involves navigating constitutional challenges, particularly regarding the First Amendment.

AI's Copyright Battle Escalates

The New York Times filed a major copyright lawsuit against OpenAI and Microsoft, alleging ChatGPT produced near-verbatim text from Times articles, posing an existential threat to the AI industry. If successful, the lawsuit could force OpenAI to pay billions in damages and destroy models trained on Times content, though OpenAI will argue its copying constitutes fair use. [Dive In]

Tomorrow Bytes’ Take…

1. Precursor to Future AI Copyright Law: The Thomson Reuters v. Ross Intelligence case may set a precedent for future AI copyright laws, given its early trial date ahead of other similar cases, trial scheduled for May 2024. The outcome of this case could significantly influence AI companies' approach towards using copyrighted material in their AI tools, including decisions about licensing, content scraping, and dataset sourcing.

2. Strategic Legal Implications: The New York Times (NYT) suing OpenAI presents a pivotal legal battle in AI and copyright law, with potential ramifications on how AI companies use copyrighted material.

3. Impact on AI Development: The outcome of this lawsuit could significantly influence the direction of AI research and development, particularly regarding data sourcing and copyright compliance.

4. Business Model Challenges for AI Companies: OpenAI's potential strategic error in handling negotiations with NYT may lead to higher damages and leverage for NYT, illustrating AI firms' complexities in balancing innovation and legal compliance.

5. Misunderstanding of AI Functionality: The lawsuit by the NY Times against OpenAI highlights a common misunderstanding about how Large Language Models (LLMs) like ChatGPT function. These models don't simply replicate or store the training data but develop a statistical understanding of language patterns and structures to generate new, original content.

6. Importance of Prompting in LLMs: The quality and nature of prompts significantly influence the output of LLMs, underscoring the need for strategic prompt structuring to harness their full potential.

7. Legal and Ethical Challenges: The lawsuit underscores the growing legal and ethical challenges in defining and regulating AI-generated content, especially regarding copyright laws and fair use, necessitating a deeper legal understanding of AI technologies.

8. Broader Industry Impact: The case underscores a growing industry concern about the sourcing of training data for AI models, emphasizing the need for proper licensing agreements.

9. Public Perception and AI Ethics: The framing of OpenAI as profit-driven, contrasted with the public good served by journalism, highlights the ethical and societal considerations AI companies must address.