A great summary of AI cross currents. You touch on it towards the end but I think AI progress may be slower for the simple reason many people DO NOT WANT IT. You note job satisfaction going down among scientists using it, and also that scheming is going up. Suppose a new technology is A. a direct threat to your job, B. a direct threat to your sense of meaning, and C. possibly a direct threat to your life (until scheming is fixed-and I've heard of no way to address that any more than they've fixed hallucinations). That's a trifecta of powerful motivations not fully appreciated by the tech bros building AI. IMO, the more they push to get it everywhere, the more quiet anti-AI backlash will gain steam. A large majority of US people are already suspicious of AI-with good cause. If AI progress is measured by benchmarks in labs (which are routinely gamed, and a great reason not to take AI labs at their word), then AI progress is accelerating. If you measure it by how many AI integrations actually work in businesses that try them, the success rate is dismal. And I'd argue the trifecta of strongly opposing motivations plays a significant part in those failures. https://www.cio.com/article/3617614/cios-lack-of-success-metrics-dooms-many-ai-projects.html
I'm 71, and spent my early career assuring leaders that computers would provide a positive ROI, eventually. One almost had to wait for senior management to die off.
But very often the benchmarks measured the wrong things; before long people realized that computerization was necessary to even be in the conversation, even if it was an expense.
Lastly, AGI progress is NOT sensitive to approval or popularity (e.g. Trump, and his cabinet of billionaires).
This was a really insightful perspective on the current state of AI. The idea that progress hasn’t slowed but has instead become harder to interpret really resonates. A lot of what users see day-to-day can feel incremental, but under the hood there seem to be significant advances especially in reasoning, coding, and research capabilities.
It also feels like we’re entering a phase where the bottleneck is less about model capability and more about how effectively we integrate AI into real-world systems. The gap between what AI can do in controlled environments and what organizations can actually deploy at scale is still quite large.
That’s why approaches that combine AI with structured data management and governance are becoming more relevant. For example, https://www.solix.com/products/enterprise-ai/ focuses on bringing more control and usability into enterprise AI adoption, which seems critical if we want these advancements to translate into measurable outcomes.
Overall, it doesn’t feel like AI is hitting a wallit feels more like we’re moving into a more complex and less visible stage of progress.
This was a really thought-provoking take on whether AI is actually slowing down or just evolving in less visible ways. It feels like a lot of the “AI is hitting a wall” narrative comes from what everyday users see, while the real progress is happening behind the scenes in areas like reasoning, coding, and complex problem-solving.
What’s interesting is that this gap between perceived capability and actual capability can create both underestimation and overhype at the same time. In practice, the challenge for organizations isn’t just about how fast AI is improving, but how effectively they can integrate it with real data, governance, and workflows.
That’s where more structured approaches to enterprise AI are becoming important. Platforms like https://www.solix.com/products/enterprise-ai/ focus on combining AI with data management and governance, which seems critical if we want these advances to translate into real-world impact.
Overall, it feels less like AI is hitting a wall and more like it’s entering a phase where progress is harder to interpret—but potentially more meaningful.
Interesting perspective on whether AI is hitting a wall or simply evolving at a different pace than expected. The point about real-world adoption versus lab benchmarks is especially important, since many organizations still struggle to translate AI capabilities into consistent business value.
In several enterprise environments, the challenge is not just model performance but integrating AI into existing data systems, governance frameworks, and workflows. Without that alignment, even promising AI initiatives can fail to deliver meaningful outcomes at scale. There are also emerging approaches around enterprise AI platforms that focus on combining governance, data lifecycle management, and AI enablement in a more structured way, as discussed here: https://www.solix.com/products/enterprise-ai/
Curious whether the slowdown perception is more about limitations in the technology itself, or the complexity of integrating AI effectively within real-world enterprise environments?
A great summary of AI cross currents. You touch on it towards the end but I think AI progress may be slower for the simple reason many people DO NOT WANT IT. You note job satisfaction going down among scientists using it, and also that scheming is going up. Suppose a new technology is A. a direct threat to your job, B. a direct threat to your sense of meaning, and C. possibly a direct threat to your life (until scheming is fixed-and I've heard of no way to address that any more than they've fixed hallucinations). That's a trifecta of powerful motivations not fully appreciated by the tech bros building AI. IMO, the more they push to get it everywhere, the more quiet anti-AI backlash will gain steam. A large majority of US people are already suspicious of AI-with good cause. If AI progress is measured by benchmarks in labs (which are routinely gamed, and a great reason not to take AI labs at their word), then AI progress is accelerating. If you measure it by how many AI integrations actually work in businesses that try them, the success rate is dismal. And I'd argue the trifecta of strongly opposing motivations plays a significant part in those failures. https://www.cio.com/article/3617614/cios-lack-of-success-metrics-dooms-many-ai-projects.html
I'm 71, and spent my early career assuring leaders that computers would provide a positive ROI, eventually. One almost had to wait for senior management to die off.
But very often the benchmarks measured the wrong things; before long people realized that computerization was necessary to even be in the conversation, even if it was an expense.
Lastly, AGI progress is NOT sensitive to approval or popularity (e.g. Trump, and his cabinet of billionaires).
This was a really insightful perspective on the current state of AI. The idea that progress hasn’t slowed but has instead become harder to interpret really resonates. A lot of what users see day-to-day can feel incremental, but under the hood there seem to be significant advances especially in reasoning, coding, and research capabilities.
It also feels like we’re entering a phase where the bottleneck is less about model capability and more about how effectively we integrate AI into real-world systems. The gap between what AI can do in controlled environments and what organizations can actually deploy at scale is still quite large.
That’s why approaches that combine AI with structured data management and governance are becoming more relevant. For example, https://www.solix.com/products/enterprise-ai/ focuses on bringing more control and usability into enterprise AI adoption, which seems critical if we want these advancements to translate into measurable outcomes.
Overall, it doesn’t feel like AI is hitting a wallit feels more like we’re moving into a more complex and less visible stage of progress.
This was a really thought-provoking take on whether AI is actually slowing down or just evolving in less visible ways. It feels like a lot of the “AI is hitting a wall” narrative comes from what everyday users see, while the real progress is happening behind the scenes in areas like reasoning, coding, and complex problem-solving.
What’s interesting is that this gap between perceived capability and actual capability can create both underestimation and overhype at the same time. In practice, the challenge for organizations isn’t just about how fast AI is improving, but how effectively they can integrate it with real data, governance, and workflows.
That’s where more structured approaches to enterprise AI are becoming important. Platforms like https://www.solix.com/products/enterprise-ai/ focus on combining AI with data management and governance, which seems critical if we want these advances to translate into real-world impact.
Overall, it feels less like AI is hitting a wall and more like it’s entering a phase where progress is harder to interpret—but potentially more meaningful.
Interesting perspective on whether AI is hitting a wall or simply evolving at a different pace than expected. The point about real-world adoption versus lab benchmarks is especially important, since many organizations still struggle to translate AI capabilities into consistent business value.
In several enterprise environments, the challenge is not just model performance but integrating AI into existing data systems, governance frameworks, and workflows. Without that alignment, even promising AI initiatives can fail to deliver meaningful outcomes at scale. There are also emerging approaches around enterprise AI platforms that focus on combining governance, data lifecycle management, and AI enablement in a more structured way, as discussed here: https://www.solix.com/products/enterprise-ai/
Curious whether the slowdown perception is more about limitations in the technology itself, or the complexity of integrating AI effectively within real-world enterprise environments?
Do you also think there's a larger gap between where China appears to be and where they are?