Christian Bäuchl
@lifeofthemind.bsky.social
psychologist, postdoc at TU Dresden
working on cognitive neuroscience of learning, memory and decision making
working on cognitive neuroscience of learning, memory and decision making
Personal experience: LLMs helped me with learning CS and math. This works especially well when trying to understand core concepts, where the training set of an LLM is huge. They are basically big knowledge processing & regurgitation machines. Faulty? yes, therefore not to be relied on exclusively
February 10, 2026 at 9:56 AM
Personal experience: LLMs helped me with learning CS and math. This works especially well when trying to understand core concepts, where the training set of an LLM is huge. They are basically big knowledge processing & regurgitation machines. Faulty? yes, therefore not to be relied on exclusively
... they are very useful for coding and learning STEM subjects (they can help foster understanding of basic concepts).
Despite their many drawbacks, I believe transformer-applications (LLMs, agentic-AI) will be a net-positive on the long run.
Despite their many drawbacks, I believe transformer-applications (LLMs, agentic-AI) will be a net-positive on the long run.
February 10, 2026 at 9:14 AM
... they are very useful for coding and learning STEM subjects (they can help foster understanding of basic concepts).
Despite their many drawbacks, I believe transformer-applications (LLMs, agentic-AI) will be a net-positive on the long run.
Despite their many drawbacks, I believe transformer-applications (LLMs, agentic-AI) will be a net-positive on the long run.
Even that that term is too generic (no pun intended). This would also include things like Variational Autencoders which are used for data compression and biological representation learning.
But even if were just talking about transformer-architecture based LLMs: I disagree with you on this one...
But even if were just talking about transformer-architecture based LLMs: I disagree with you on this one...
February 10, 2026 at 9:14 AM
Even that that term is too generic (no pun intended). This would also include things like Variational Autencoders which are used for data compression and biological representation learning.
But even if were just talking about transformer-architecture based LLMs: I disagree with you on this one...
But even if were just talking about transformer-architecture based LLMs: I disagree with you on this one...
Now you're taking it too far (or maybe that's just ragebait?).
It's like saying that you cannot be friends with people who get excited about web science because social media is ruining democracy.
It's also frustrating how people use the very broad umbrella term 'AI' as a stand-in for LLMs.
It's like saying that you cannot be friends with people who get excited about web science because social media is ruining democracy.
It's also frustrating how people use the very broad umbrella term 'AI' as a stand-in for LLMs.
February 10, 2026 at 8:16 AM
Now you're taking it too far (or maybe that's just ragebait?).
It's like saying that you cannot be friends with people who get excited about web science because social media is ruining democracy.
It's also frustrating how people use the very broad umbrella term 'AI' as a stand-in for LLMs.
It's like saying that you cannot be friends with people who get excited about web science because social media is ruining democracy.
It's also frustrating how people use the very broad umbrella term 'AI' as a stand-in for LLMs.