The Cal Newport Show
Ep. 367: What if AI Doesn’t Get Much Better Than This?
· Summarized
Key takeaways
- The hype around AI has cooled after GPT-5, revealing that current models are more incremental than revolutionary, and the future likely involves steady, specialized improvements rather than a sudden leap to superintelligence.
- Much of the economic disruption attributed to AI is exaggerated; layoffs and job changes are primarily driven by economic cycles, not AI replacement, and media often conflates usage of AI tools with job automation.
- The original scaling law, which suggested that bigger models would continually outperform smaller ones, has plateaued, leading the industry to shift focus towards post-training techniques that optimize existing models.
- Expect a more gradual, fragmented, and less dramatic AI evolution over the coming years, emphasizing bespoke systems and incremental innovations rather than a singular breakthrough.
- It remains important to regulate, ethically examine, and thoughtfully develop AI, but the doomsday or utopian narratives are unlikely to match the reality in the near future.
Summary
The Reality of AI Progress and Disappointment
This section explores the recent shift in perceptions about AI's capabilities following the underwhelming release of GPT-5. Once thought to be on the brink of superintelligence and revolutionary impacts, the reality appears more incremental. Prominent AI executives like Sam Altman and Mark Zuckerberg fueled the hype, comparing their work to the Manhattan Project and the race to build superintelligence. However, GPT-5's limited improvements and the stalled scaling law indicated that the industry might not be progressing toward the anticipated 'artificial general intelligence.' This caused skepticism among industry experts and the public, leading to a reassessment of AI's near-term potential.
Current Impacts and Misconceptions about AI’s Economic Effects
This section addresses the misconception that AI is already causing widespread unemployment and economic disruption. Based on industry analysis, including insights from tech analyst Ed Zitron, the actual impact is modest. Media headlines often conflate the use of AI tools with job displacement, but data suggests many layoffs are due to economic cycles, not AI replacement. AI tools are mainly being used for small efficiencies, not large-scale automation. The industry's revenue and investment reveal that AI's economic footprint remains relatively small for now, with most claims of imminent upheaval being hyperbolic.
The Shift from Scaling Law to Post-Training Techniques in AI Development
This part details the technological evolution in AI from the dominant focus on scaling models (bigger data, bigger models) to alternative methods like post-training refinement, reinforcement learning, and synthetic data generation. The scaling law—proof that larger models performed exponentially better—led to massive investments and giant models like GPT-3 and GPT-4. But around 2024, this approach plateaued; larger models yielded diminishing returns. Industry leaders shifted focus to post-training techniques that fine-tune pre-trained models, improve performance on specific tasks, and enhance capabilities without exponentially increasing model size. This shift explains why newer models show incremental gains rather than revolutionary leaps.
The Future of AI: Realistic Expectations and Gradual Progression
Cal Newport concludes that AI is unlikely to deliver the dramatic disruptions forecasted in recent years. The era of rapid, universal AI-driven automation appears to be over for now. Instead, progress will be steady and fragmented across various specialized systems, gradually improving with bespoke, incremental innovations. Experts like Gary Marcus warn that current technologies are at their limits and that true artificial general intelligence requires breakthroughs beyond scaling. Newport emphasizes the importance of regulation, ethics, and careful technological development, suggesting that while AI will remain powerful, the hyperbolic predictions of immediate catastrophe or utopia are unrealistic for the foreseeable future.
Listener Questions and Additional Topics
The podcast features listener questions on diverse topics, including career decisions in AI and computer science, managing overhead and communication post-project, organizing personal libraries, and personal habits like disconnecting from constant connectivity. Newport advises focusing on impactful work, setting boundaries to reduce overhead, and pursuing projects aligned with genuine interests. The discussion also touches on Ed Sheeran’s approach to digital minimalism and the benefits of selective communication. Overall, the questions highlight a desire for practical ways to live more intentionally and reduce distraction in a world saturated with technological noise.
Sponsors
- Grammarly · Grammarly is the essential AI communication assistant that boosts productivity, helping you with tone detection, rewriting, and generating variations of your text. Use it to improve your writing efficiency and clarity.
- Cozy Earth · Cozy Earth offers ultra-comfortable, temperature-regulating sheets, apparel, and bath products. Their bamboo bedding and clothing are designed for comfort, durability, and quality. They offer a 100-night sleep trial and a 10-year warranty.
- Caldera Lab · Caldera Lab provides high-performance men's skin care products including serums and moisturizers, designed to improve skin health, hydration, and youthful appearance. Use code DEEP for 20% off your first order.
- ShipStation · ShipStation helps eCommerce businesses streamline their shipping with a central dashboard, automation, and discounts up to 88% on shipping rates across major carriers. Try it free with no credit card required by visiting shipstation.com/deep.
Show notes
In recent years, it’s been hard not to react to the possibilities of generative AI without a mixture of euphoria or dread. But after OpenAI’s lackluster GPT-5 launch, a new, almost heretical-seeming question has emerged: what if progress on AI is stalled well short of the wild predictions we were promised? In today’s episode, Cal draws from reporting on his recent New Yorker article to go deep into this question. What is going on with AI? How did we get here? What does it mean for our personal quest to live deeper lives? He then answers listener questions and ends by discussing his recent brush with literary acclaim.
Below are the questions covered in today's episode (with their timestamps). Get your questions answered by Cal! Here’s the link: bit.ly/3U3sTvo
Video from today’s episode: youtube.com/calnewportmedia
Deep Dive: What if AI Doesn’t Get Much Better Than This? [1:00]
Will AI leave me unemployed in 10 years? [1:06:26]
How should I structure the next 10 years as a recently retired college professor? [1:13:58]
I just moved. How should I arrange my book collection? [1:15:36]
CALL: Overhead tax [1:17:47]
CAL REACTS: Ed Sheehan and the Booker Prize [1:27:55]
Links:
Buy Cal’s latest book, “Slow Productivity” at www.calnewport.com/slowGet a signed copy of Cal’s “Slow Productivity” at peoplesbooktakoma.com/event/cal-newport/Cal’s monthly book directory: bramses.notion.site/059db2641def4a88988b4d2cee4657ba?newyorker.com/culture/open-questions/what-if-ai-doesnt-get-much-better-than-thisyoutube.com/watch?v=zju51INmW7Uyoutube.com/watch?v=Dtdue31z-X8youtube.com/watch?v=0SXCIfFK5r8youtube.com/shorts/dYZmGHOLNRUyoutube.com/watch?v=k82RwXqZHY8youtube.com/watch?v=qhnJDDX2hhUyoutube.com/watch?v=qbIk7-JPB2c&t=3syoutube.com/watch?v=k82RwXqZHY8youtube.com/shorts/JCImeOUVTJE
Thanks to our Sponsors:
grammarly.com/podcastcozyearth.com/deepcalderalab.com/deepshipstation.com/deep
Thanks to Jesse Miller for production, Jay Kerstens for the intro music, and Mark Miles for mastering.
Learn more about your ad choices. Visit podcastchoices.com/adchoices
Follow this show on AudioBrevity and new episodes get summarized for you. Create a free account