We found that motion features were encoded in a temporally ordered sequence: orientation and spatial frequency emerged within 120 ms after stimulus onset, while temporal frequency and direction followed at later latencies.

[2/3]

A nice shift in perceived colour between central and peripheral vision. The fixated disc looks purple while the others look blue.

The effect presumably comes from the absence of S-cones in the fovea.

From Hinnerk Schulz-Hildebrandt:
arxiv.org/pdf/2509.115...

The model can also fit continuous confidence ratings without binning these ratings.

Matlab code here: github.com/mamassian/cn...

(4/4)

Critically, the model includes a parameter that reflects whether confidence relies on the same information that was used for perceptual decisions, or some new information. We describe two ways to estimate this Confidence Boost parameter.

(3/4)

The CNCB (Confidence-Noise Confidence-Boost) model can be seen as an alternative to the popular meta-d’ and M-ratio analyses. It can take more than two stimulus strengths and is a bit more tolerant to sensory biases.

(2/4)

If you work with perceptual confidence judgments, you may be interested in our CNCB model of confidence ratings. Joint work with Vincent de Gardelle.

Uncorrected proofs here:
dx.plos.org/10.1371/jour...

(🧵 1/4)

Why is it so difficult for Large Language Models to properly read time?

arxiv.org/abs/2502.05092

The dynamic version is even more disturbing...

Congratulations!

Richard Gregory probably helped to popularize the cultural hypothesis in his nice 1966 book “Eye and Brain”.

But since then, isn’t the dominant theory related to a misuse of the low spatial frequency content of the image (its blurred part)?

See e.g. doi.org/10.1068/p150...

I agree that the coming and going of 3D perceptions makes this animation interesting!

A nice development on the “stereokinetic effect” of Cesare Musatti a century ago, and more recently of the work of Zili Liu (UCLA):
doi.org/10.1016/j.vi...

original post here:
bsky.app/profile/jcpo...

Interestingly, another model, “GPT-4o”, is also biased like humans, but it can generate some python code to analyse the image pixel by pixel to correct its error

Overall, I agree with Tomer that it is not clear why we would want these models to fall for illusions the way humans do.

4/4

The model also “knows” that the effect is orientation dependent

www.sciencedirect.com/science/arti...

3/4

As predicted, it is also reporting the illusion when the lines are un-equal (“illusion-illusion”), and it explicitly refers to the vertical-horizontal illusion.

2/4

Very interesting analysis!

I have tried the vertical-horizontal illusion in its classic configuration. The model “Claude” is reporting the human-biased percept when the lines are physically equal (“illusion”)

1/4

Cast shadows are indeed so powerful!

Here is another related picture from the following chapter that Dan Kersten and I wrote:
hal.science/hal-03984071/

It was nice to see that many people interested in time perception at the first FAsT (French Association for Timing) meeting in Saclay last week!
fast2024.sciencesconf.org

Happy to share the results of interdisciplinary discussions between scientists and artists on colour (in French) — Heureux de partager les résultats de discussions interdisciplinaires entre scientifiques et artistes sur la couleur (en français)
doi.org/10.3917/herm...