Carson McNeil
carsonmcneil.bsky.social
Carson McNeil
@carsonmcneil.bsky.social
Researcher in pathology, neuroscience, and machine vision. Stoked about the brain, Buddhism, mental health, and Mysticism.

https://carsonmcneil.me

I write long-form stuff too:
carsonogenic.substack.com
Absolutely. I'm curious whether there is a good cultural scaffolding to make the interaction safer. "Shoggoth with a mask" is better than "my personal assistant" or "a dumb program". But it's still a little nerd-core.
August 4, 2025 at 6:30 PM
I think ideally the AI chat would have a personality that is not just "conversation partner of some level", but "tutor", willing to occasionally break the flow of conversation to explain a construct or correct bad grammar, and then return to where the conversation paused.
March 29, 2025 at 8:20 PM
2. I think it would really maintain your advantage if you operationalize and track this type of "quality" thing, even if it's not possible to get it to 100% today.

Good luck! I'll probably check in and use it every so often! It feels closer to what I want than anything else I've found so far.
March 29, 2025 at 8:17 PM
I understand these are extremely difficult ML problems, but
1. They are possible to solve, as a human tutor is generally pretty good at this sort of thing.
March 29, 2025 at 8:17 PM
- Quality of voice to text: I would like to see more "did you mean...?" and correction of pronunciation based on context as opposed to just mishearing me (or I mispronounce) and the conversation veers off on a tangent. It feels hard to correct or "rewind" a conversation.
March 29, 2025 at 8:17 PM
- I would love to see better ability to blend English and mandarin, the way an intro tutor would with you. I'm sure it's debated but I don't think 100% pure immersive is as good as some translation and Chinglish when starting.
March 29, 2025 at 8:17 PM
Oh hey! Yea, I like the format of the AI chat agents. Free-form voice chat with pinyin included is great. I feel most of what could improve is not additional features, but intelligence of existing features:
March 29, 2025 at 8:17 PM
I can tell you from exploring so far that Langotalk is much much better than Talkpal, which is practically unusable. However, it still leaves much to be desired. The bones of Langotalk are good, I just wish its agent was smarter and had fewer mistakes/misunderstandings.
March 18, 2025 at 5:30 PM
I really want you to be a contestant on Jetlag.
March 16, 2025 at 9:50 PM
My roommate @exoconnor.bsky.social would be great for this role, and is available rn! Matches everything + some neuro background. Texting you.
February 21, 2025 at 4:32 AM
It seems likely that some other party will buy the cars and do that.
January 15, 2025 at 3:44 PM