Concerns about LLM hallucinations in mission-critical areas often miss the bigger picture: the processes they replace are often manual, performed by humans, and riddled with errors or 'hallucinations' of their own. The focus should be on managing risks, not chasing perfection.
tag > ML
-
-
The “Intelligence Trap”
The “Intelligence Trap” is a concept in psychology that suggests that highly intelligent people are more susceptible to cognitive biases and flawed thinking than less intelligent people. For example, it can be argued that Stephen Hawking’s Warning on Contacting Aliens may be an example of the so-called intelligence trap, as some evidence from the field of physics suggests.
-
First, AI took over private homework tutoring. Then it gradually replaced specialized teachers. Soon, the entire school system is in deep crisis. The time for radical curriculum reform is now.
-
Connected Graphs are a great way to get absolutely confused about a reality permeated by non-local fields
-
After a decade working on HCI / ML latent space exploration, I’ve noticed a big common misconception: Many believe longer exploration improves user outcomes. I argue the opposite—short, efficient interactions serve users better and should be our goal.
-
Still using outdated Neural Network Activation Functions like Sigmoids and ReLU? Check out these powerful alternatives which will radically boost the performance of your models.
-
The computational science community still has a long way to go to realize that the latent space of ideas shaped by evolution is vast & distributed. The West, stuck in a local minimum, is erasing all Indigenous tribes—some of the last sources of truly unique insights.
-
The ability to take any poorly written or even negatively toned text and tell an LLM to "Re-write this text in a positive and 'can do' style" has profound implications for business & culture we’re only beginning to grasp.
-
The ability to take any poorly written or even negatively toned text and tell an LLM to "Re-write this text in a positive and 'can do' style" has profound implications for business & culture we’re only beginning to grasp.
-
For nearly a decade, I've been waiting for AI tools that can explore ideas seamlessly across different levels of abstraction—switching perspectives at will. Yet, beyond basic LLMs, this hasn't hit the mainstream. When will we see it? Who's working on this?
-
Despite the rapid advancements in AI-generated music, it's intriguing that nothing has come close yet to crafting an iconic, genre-defining sound like classic instruments such as the TB-303.
