Welcome to the last 2024 edition of the Daily AI Show Newsletter.

In this issue:

• The 2024 AI Predictions We Got Right—and Wrong

• o3 and the Future of AI: Incremental or Transformative?

• What’s Stopping You? Overcoming Barriers to AI Adoption

• Making Room For AI In 2025: The Tools That Don’t Make The Cut

Plus, we discuss the dangers of personalization with AI, Oreos getting optimized, Ilya talks about the “post pre-training” phase of AI developments, building giant mech robots, and our favorite “good news” stories from 2024.

It’s Sunday morning

The countdown to 2025 is on, and AI is already predicting how many resolutions we’ll break by February. 😟

Let’s get to it and Happy New Year!
The DAS Crew - Andy, Beth, Brian, Eran, Jyunmi, and Karl

The 2024 AI Predictions We Got Right—and Wrong

We made some predictions about AI’s future in November of last year. Expectations ranged from Microsoft Copilot driving enterprise AI adoption to Apple delivering competitive AI products, toNVIDIA facing serious challenges in the hardware space. A year later, the reality is a mix of progress, missed opportunities, and surprises.

Here’s how we did:

PREDICTION: Microsoft Copilot Will Deliver Strong ROI

Verdict: Wrong.

Businesses struggled with Copilot’s clunky interface and lack of meaningful integration with Office 365. While heavily marketed, its actual functionality fell far short of expectations. Companies that adopted it reported frustration with its high costs and limited utility. Far from dominating, Copilot left many businesses looking for alternatives like ChatGPT.

PREDICTION: Anthropic and Google Will Compete Significantly with OpenAI

Verdict: Partially Right

Both Anthropic’s Claude and Google’s Gemini made strides in 2024. Claude gained prominence in coding and creative writing, while Gemini delivered native multimodal capabilities and benchmark performance. However, neither managed to pull significant market share of users away from OpenAI, which held its lead with innovations like Canvas and o1.

PREDICTION: Real-World ROI Will Drive Business AI Adoption

Verdict: Partially Right

Reports highlighted measurable ROI for early adopters, particularly in sales and customer service. However, the broader adoption of AI remained slow, with only 6% of enterprises fully integrating AI into their operations. Despite clear benefits, many businesses hesitated to invest, citing cost and complexity.

PREDICTION: Apple Will Introduce Competitive AI Products

Verdict: Wrong

Apple Intelligence delivered underwhelming updates and failed to make Siri or its broader AI ecosystem competitive with offerings from OpenAI, Google, or Anthropic. Apple’s limited progress in AI left users disappointed and skeptical of its future potential in this space. If Apple is the Tortoise vs OpenAI’s Hare, then we have to wait longer for them to impress.

NVIDIA Will Face Significant Competition in AI Hardware

Verdict: Right

Companies like AMD, Qualcomm and Google introduced competitive hardware in 2024, Amazon’s Trillium chips made headlines, and major partnerships emerged to develop alternatives for AI data centers. Startups like Cerebras show promise for leapfrogging GPUs for AI training and inference with new technology. While NVIDIA maintained its dominance, the emergence of new players signaled growing competition in the AI hardware market. However, these competitors have yet to take significant market share.

PREDICTION: Fragmentation in AI Platforms Will Create Opportunities for Consultants

Verdict: Wrong

While consultants remain valuable, the prediction that fragmentation would drive their necessity didn’t fully materialize. OpenAI’s ecosystem continued to consolidate with tools like Canvas, reducing the need for external expertise. Businesses gravitated toward streamlined solutions rather than embracing multiple platforms.

o3 and the Future of AI: Incremental or Transformative?

OpenAI’s announcement of o3 has reignited debates about whether we’re on the brink of Artificial General Intelligence (AGI). The model demonstrates impressive capabilities, including reasoning through complex problems and leveraging multimodal inputs.

Yet, there’s no consensus on whether this marks the leap to AGI or simply another step in the progression of smarter, more capable AI systems. The broader conversation reveals a shift in focus from theoretical milestones like AGI to the tangible ways AI can transform industries and workflows today.

While o3’s ability to self-assess, adapt, and execute tasks like writing and running Python programs that support o3’s pursuit of objectives suggest significant advancements, practical limitations such as high compute costs and the absence of general-purpose memory highlight how far there is yet to go for AGI.

For businesses and individuals, the real value lies in understanding how these advancements translate into real-world applications that boost productivity, enhance creativity, and solve complex problems.

WHY IT MATTERS

Practical Applications Today: Whether or not o3 qualifies as AGI, its capabilities represent a leap forward in reasoning, task automation, and multimodal interactions, offering immediate benefits for businesses.

The Cost Barrier: High computational costs limit accessibility to o3, making it critical for companies to weigh the trade-offs between adopting cutting-edge AI models vs using competent recent models at lower cost.

Rethinking Workflows: Tools like o3 signal a future where AI takes on more of the planning, execution, and iteration processes, freeing humans to focus on strategy and innovation.

Broader Impact Beyond Labels: The AGI label may be less important than the real-world outcomes these tools can deliver, from revolutionizing customer support to enhancing research workflows.

Incremental vs. Transformative Change: While o3 isn’t yet the all-encompassing AGI some imagine, its advancements demonstrate the potential for iterative improvements to reshape industries in meaningful ways.

What’s Stopping You?
Overcoming Barriers to AI Adoption

AI adoption remains surprisingly low, with only a small percentage of individuals and businesses fully integrating AI into their workflows. Common roadblocks include unfamiliarity with AI tools, concerns about data security, and misconceptions about the effort required to implement them. For many, the initial hurdle is gaining understanding of where AI fits into their daily tasks or workflows. Instead of seeing AI as a burden, it helps to view it as a tool for simplifying repetitive tasks, generating creative ideas, and enhancing productivity.

AI is now more accessible than ever, offering solutions for professionals and individuals alike. From improving automation with tools like Zapier to performing strategic business analyses with advanced models, AI has reached a level of sophistication that can revolutionize how work gets done.

The key lies in understanding the practical entry points, regularly exploring what’s newly available, and making AI usage part of your daily routine.

WHY IT MATTERS

Time-Saving Potential: AI can streamline repetitive tasks like data entry, content summarization, or simple automations, freeing up time for more strategic work.

Skill Development Through Practice: Regular use of AI tools builds confidence and expertise, making it easier to expand applications in personal and professional contexts.

Integrating Strategic Capabilities: Advanced models can perform complex analyses and generate actionable insights, helping businesses identify opportunities and optimize marketplace decisions.

Organizational Efficiency: For teams, internal hackathons and collaborative learning sessions can uncover innovative ways to use AI and encourage widespread adoption.

Accessible Entry Points: Even beginners can explore AI by starting with simple tasks like summarizing content, creating schedules, or automating basic workflows.

Making Room For AI In 2025:
The Tools That Don’t Make The Cut

It is time to clear out the tools, workflows, and platforms that no longer serve us and make room for the ones that will drive efficiency and innovation in the year ahead. This metaphorical housecleaning highlights trends in AI adoption, where users are choosing streamlined, high-functionality solutions over fragmented or redundant tools.

For many, 2024 saw a move away from earlier-generation AI tools that felt like “wrappers” around base models. Writing assistants like Jasper and HyperWrite, once essential for generating content, have been replaced by more advanced offerings from Claude and ChatGPT, which now include native customization, better integrations, and improved usability.

Similarly, platforms like Perplexity have surpassed earlier plug-in options for real-time web search with enhanced AI integration and accessibility. The focus is shifting to tools that provide versatility and simplicity without sacrificing power.

WHY IT MATTERS

Efficiency Over Redundancy: Users are gravitating toward tools that eliminate the need for multiple platforms, prioritizing integrated solutions with robust functionality.

Adapting to Advanced AI Models: Frontier models like Claude and Gemini are making older, simpler tools obsolete, driving users to reevaluate their tech stacks.

Streamlining Creativity: From writing and coding to image generation, professionals are opting for platforms that centralize workflows, reducing friction and saving time.

Making Space for Innovation: Letting go of outdated tools creates room for experimentation with emerging technologies and more efficient alternatives.

The Value of Good Enough: Many users prefer tools that are functional and accessible, even if they’re not the absolute best, balancing convenience with effectiveness.

Just Jokes

Did you know?

Mondelez International, the company behind Oreo and Chips Ahoy, is using AI to develop new snack recipes and bring them to market faster. By leveraging machine learning tools, Mondelez can specify desired product attributes like flavor, aroma, and appearance, while also considering ingredient costs, environmental impact, and nutritional value.

This AI-driven approach has already contributed to the creation of 70 new products, including Gluten Free Golden Oreos, and has accelerated product development by four to five times.

Ilya Sutskever Talks About Post Pre-Training Period of Scaling

Eran shared a YouTube video talking about Ilya’s recent talk at the neural Information Processing Systems conference in Vancouver. He asserted that we are at a pivotal moment in AI development. He outlined key advancements and challenges in the field, emphasizing the shift beyond pre-training and "peak data" to new paradigms like synthetic data generation and advanced inference-time compute. The talk also explored the future of AI reasoning, predicting the emergence of genuine, human-like reasoning in AI systems, which will lead to unpredictability and transformative capabilities.

Karl Wants to Be A Gundam Pilot

Japanese startup Tsubame Industries has unveiled the Archax, a 15-foot-tall, four-wheeled robot inspired by the Gundam series. Priced at $3 million, the Archax can transform between a vehicle mode and a standing "robot" mode, accommodating a pilot within its cockpit. It was showcased at the Japan Mobility Show, highlighting Japan's enthusiasm for large-scale robots. Tsubame Industries envisions future applications for the Archax, including robot fighting leagues and potential lunar missions.

Karl said, “Forget AGI, I want to live out my childhood fantasy of being a Gundam Pilot.

Beth’s response . . . “We're going to need a bigger show. 🦈”

This Week’s Conundrum
A difficult problem or question that doesn't have a clear or easy solution.

The AI Personalization Identity Paradox::

As AI personalizes every aspect of our lives—how we learn, communicate, shop, and even think—it begins to mirror and reinforce our preferences and identities in profound ways. Over time, our interactions with personalized systems may subtly reshape who we are, creating an iterative loop where our choices and preferences are not entirely our own but heavily influenced by the AI’s design. What have we learned from the impact of social media algorithms on the personalized “inputs” to our worldview?

This raises a difficult question: At what point does personalization stop serving us and start defining us?

The conundrum: If AI systems reflect and reinforce our identities over time, are we truly autonomous in our choices, or are we shaped by a feedback loop of tailored experiences?

Should we celebrate this as a form of self-discovery enabled by technology, or should we fear it as a quiet erosion of free will and individuality?

On our Christmas episode this year, we decided to focus on some of our favorite good news stories from 2024.

Machine Psychology Introduced to Enhance AI Alignment

A researcher from Linköping University proposed "machine psychology," a framework combining psychological principles with AI to enable systems to operate with limited data and computational power. This approach aims to solve alignment issues, paving the way for safer and more efficient AI models.

AI Improves NICU Monitoring for Infants

AI systems trained on video data now monitor neurological changes in neonatal intensive care units (NICUs). This technology provides real-time alerts to doctors, helping detect potential issues earlier and improve care outcomes for vulnerable newborns.

Robotic AI Scientists Automate the Scientific Method

Tetsuwan Scientific is developing robotic AI systems capable of autonomously executing the full scientific method, from hypothesis formation to experimentation and analysis. These robots promise to accelerate research in fields like medicine and environmental science.

AI Prevents Train:Elephant Collisions in India

AI-powered systems integrated with drones and sensors in India have successfully averted over 800 train-elephant collisions. This technology helps preserve wildlife while ensuring safer railway operations.

Waymo Demonstrates Superior Driving Safety

Waymo’s autonomous vehicles reduce property damage by 88% and bodily injuries by 92% compared to human drivers, reinforcing their position as a safer alternative to traditional driving methods.

Liquid AI Neural Models Mimic Worm Neural Networks

Liquid AI, inspired by the simple neural system of the C. elegans worm, is developing efficient models that require less computational power. These systems are ideal for edge computing, making AI more accessible for consumer and industrial applications.

NVIDIA’s ‘Jetson Nano Super’ Empowers Small Projects

NVIDIA released the Jetson Nano Super, a $249 device that brings powerful AI computing to hobbyists and small businesses. This edge AI tool enables innovations in robotics, automation, and localized AI systems without cloud reliance.

MIT Advances Neural Networks with Logic Gates

MIT researchers have developed a way to run neural networks directly on hardware using logic gates. This approach significantly reduces energy consumption by skipping burdensome matrix multiplication, enabling more efficient AI compute operations on edge devices.

AI Creates Inclusive Educational Materials

AI is being used to create more inclusive educational content, such as subtitles, transcripts, and audio versions. These tools not only aid individuals with disabilities but also enhance learning experiences for everyone.

AI Cooperation Study Highlights Anthropic's Claude 3.5

A recent study comparing AI models found Anthropic’s Claude 3.5 excelled in cooperation-focused tasks, outperforming GPT-4 and Google’s Gemini Flash by maintaining long-term collaborative strategies. This reflects the effectiveness of Anthropic’s constitutional AI approach.

Google Launches AI for Good Platform

Google unveiled an "AI for Good" platform to showcase projects in climate, healthcare, education, and more. Highlighted initiatives include AlphaFold, which has revolutionized protein folding and drug discovery.

Did You Miss A Show Last Week?

Enjoy the replays on YouTube or take us with you in podcast form on Apple Podcasts or Spotify.