Brad Wyble
banner
Brad Wyble
@bwyble.bsky.social
Academic, cognitive & vision scientist, computational modeller, cofounder @neuromatch Academy, He/His. This is a personal account.
People will always be set in their ways, that's just a universal constant, but it's not a reason to dismiss other issues.
January 21, 2026 at 1:59 AM
Switching languages is important for better coding practices but also very difficult for the reasons you suggest. I'm arguing that LLMs will make it even harder.
January 20, 2026 at 8:17 PM
switching to new languages, even ones that are clearly better at data management, security, typing, efficiency, etc, could therefore increase substantially. I don't think that's good.
January 20, 2026 at 8:15 PM
Because languages become adopted when people adopt them. But if the majority of coders in the future rely on LLMs, they will be discouraged from using languages that there is insufficient training data for, because the LLMs won't be able to structure code very well for them. The resistance to >
January 20, 2026 at 8:14 PM
This seems 100% true.
January 20, 2026 at 12:45 PM
Try to have a first complete draft done at least a month in advance so that you have time for serious revisions.
January 20, 2026 at 12:44 PM
People won't use Claude for helping to code a language it hasn't been trained on, so where does that interaction come from?

As for incentive for new coding, imagine if we'd stopped at BASIC or COBOL?
January 19, 2026 at 11:46 PM
Modularity/locality does not require discrete, gradient-free separation.
January 19, 2026 at 6:58 PM
I don't think anyone is expecting a truly discrete boundary like we see in artificial systems. But there seem to be some parts of the cortex that are at least substantially distinct from their immediate neighbors.
January 19, 2026 at 6:16 PM
But again, this negate the many other kinds of processing that are clearly non-modular/non-local.

What I struggle with here is the idea that there is a right approach. The brain has multiple styles of computation running in parallel.
January 19, 2026 at 5:17 PM
There are others though. Dorsal lesions allow people to identify objects but not grasp them. Ventral lesions allow people to appropriately interact with objects without being able to recognize them. These seem reasonably specific from a functional perspective.
January 19, 2026 at 5:08 PM
True but some of these specific deficits are pretty crisp (from my arm's length understanding). E.g. aphasia is a very profound phenomenon and speaks to a significant clustering of speech generation function.
January 19, 2026 at 4:54 PM
Agreed, we're using a pretty weak definition of modularity in this thread.
January 19, 2026 at 4:17 PM
The interesting evidence for modularity is selective forms of agnosia, as suggested above.
January 19, 2026 at 2:51 PM
Reposted by Brad Wyble
Of course. And, every single time, I tell the story of the Google code starting with an NSF GRFP, just to remind the numbskull dipshits that the NSF has paid for itself a bazillion times over. Not that it matters, it's just screaming into the void.
January 15, 2026 at 10:45 PM
Reposted by Brad Wyble
There are only so many hours in the day, and only so many sycophants willing to work for government salaries

Every hour DOJ wastes going through hiring, onboarding, federal court admission, learning how to not fuck up, etc is an hour they can't spent doing repugnant shit

bsky.app/profile/ohzo...
Why is quitting good? Don't they just get replaced with sycophants?
January 14, 2026 at 2:20 AM
Is that illegal?
January 14, 2026 at 1:35 AM