Gary Lupyan
@glupyan.bsky.social
Reposted by Gary Lupyan
I have been writing critically about twin studies of behavior for many years. Here is a brief overview article I published online last year.
www.madinamerica.com/2024/09/the-...
www.madinamerica.com/2024/09/the-...
November 22, 2025 at 4:31 PM
I have been writing critically about twin studies of behavior for many years. Here is a brief overview article I published online last year.
www.madinamerica.com/2024/09/the-...
www.madinamerica.com/2024/09/the-...
Have you guys done any comparisons with how relatively effective people are in of debating arguing w conspiracy believers?
November 22, 2025 at 5:29 PM
Have you guys done any comparisons with how relatively effective people are in of debating arguing w conspiracy believers?
There should be a recording here: talkbank.org/psyling/E*3/
The E*3 presentation series is dedicated to exploring the emergence
of structure and process in language and cognition. Emergentism,
broadly defined, examines the role of the physical, neural, and social
environment in providing mechanistic constraints that shape both
temporary and semi-permanent structures on various structural levels,
eventually leading to linguistic and cognitive expertise. E*3 also
emphasizes the role of ecology, the environment at various levels
(current, over time, or at the time of learning), as a factor in this
process. We also invite presentations that challenge specific
emergentist proposals, as well as presentations that formulate
constraints through alternative and complementary formal models.
talkbank.org
November 21, 2025 at 7:38 PM
There should be a recording here: talkbank.org/psyling/E*3/
I applaud the attempt for sure. arc-agi is very clever. But I think the approach mischaracterizes *human* performance. There’s a great deal of prior knowledge that ppl require. And rather than showing “extreme generalization”, people just show that they share many of Chollet’s conceptual priors.
November 20, 2025 at 3:44 AM
I applaud the attempt for sure. arc-agi is very clever. But I think the approach mischaracterizes *human* performance. There’s a great deal of prior knowledge that ppl require. And rather than showing “extreme generalization”, people just show that they share many of Chollet’s conceptual priors.
Funny that... Also interesting that I saw Chollet give a presentation with this slide just last Spring... I guess lots of goalpost shifting on all sides? 🤷♂️
November 20, 2025 at 3:13 AM
Funny that... Also interesting that I saw Chollet give a presentation with this slide just last Spring... I guess lots of goalpost shifting on all sides? 🤷♂️
How have these kinds of reviews become normalized?!
November 20, 2025 at 2:43 AM
How have these kinds of reviews become normalized?!
I’ve used this a few times before as an example of the pivot from word knowledge to world knowledge. Those non fine tuned models were nearly impossible to actually converse with! (Which is very interesting)
November 19, 2025 at 12:38 AM
I’ve used this a few times before as an example of the pivot from word knowledge to world knowledge. Those non fine tuned models were nearly impossible to actually converse with! (Which is very interesting)
It’s true… we’ve got hotdogs so figured out now!
November 19, 2025 at 12:37 AM
It’s true… we’ve got hotdogs so figured out now!
btw, is this from 2020 or 2021?
November 19, 2025 at 12:07 AM
btw, is this from 2020 or 2021?
@chazfirestone.bsky.social Your hotdog tweet is featured! :)
November 18, 2025 at 11:38 PM
@chazfirestone.bsky.social Your hotdog tweet is featured! :)
What a fun read! My thoughta are closest to Michael’s I‘be been meaning to read Anil’s BBS paper - hoping help me understand what‘s driving him to this view that the substrate inherently matters.
November 12, 2025 at 6:02 AM
What a fun read! My thoughta are closest to Michael’s I‘be been meaning to read Anil’s BBS paper - hoping help me understand what‘s driving him to this view that the substrate inherently matters.
Browsing through the doordash infographics is something else.. They can't even interpret their own data!
November 12, 2025 at 4:07 AM
Browsing through the doordash infographics is something else.. They can't even interpret their own data!
I didn't mean that it's easy in an absolute sense (or easy to 'solve' from kid amount of data), just that it doesn't seem that LLMs get a huge advantage from being trained on text vs. speech. @begus.bsky.social - any thoughts on this?
November 11, 2025 at 2:32 AM
I didn't mean that it's easy in an absolute sense (or easy to 'solve' from kid amount of data), just that it doesn't seem that LLMs get a huge advantage from being trained on text vs. speech. @begus.bsky.social - any thoughts on this?
Yes, that's my understanding as well. A lot of applied work uses text just because it's so effective, but my understanding is that it's possible to train these sytems w/o text ai.meta.com/blog/textles... and that there is no in-principle roadblock
ai.meta.com
November 10, 2025 at 6:28 PM
Yes, that's my understanding as well. A lot of applied work uses text just because it's so effective, but my understanding is that it's possible to train these sytems w/o text ai.meta.com/blog/textles... and that there is no in-principle roadblock
Learning to tokenize is not hard (eg one can start w raw audio) but a bigger issue is that LLMs benefit from an already existing language (and it’s vocabulary) while we had to invent it. Of course babies born into a linguistic world have that same benefit :)
November 10, 2025 at 4:36 PM
Learning to tokenize is not hard (eg one can start w raw audio) but a bigger issue is that LLMs benefit from an already existing language (and it’s vocabulary) while we had to invent it. Of course babies born into a linguistic world have that same benefit :)