Christoph Strauch

How much of a risk are LLM-generated bots for online behavioral experiments? We present results of an entirely natural language prompt-generated agent, capable of absolving any of the 26 experiments we tested it on. Most tasks are passed with data that can pass as human osf.io/preprints/ps... (1/)

How much of a risk are LLM-generated bots for online behavioral experiments? We present results of an entirely natural language prompt-generated agent, capable of absolving any of the 26 experiments we tested it on. Most tasks are passed with data that can pass as human osf.io/preprints/ps... (1/)

The agent was built in August/September using a monthly subscription of around 17 euros/dollars. Time investment including testing on most of the tasks was about two weeks. By now, even less sophisticated prompting is sufficient to solve complex behavioral tasks in 1-2 attempts.

Now final in @elife.bsky.social: #Synesthetes really do 'see' when reporting colors while reading (typeface gray) numbers: pupils change according to the perceived colors' brightness - if a synesthete sees a bright yellow, the pupils become smaller. elifesciences.org/articles/110...

Finally, brightness-effects were interacting with self-reported strength. Synesthetes claimed different levels of strength (e.g. the bright blue for a 5 being stronger than the dark orange for a 3). And indeed, for those stronger couples, brightness effects on the pupil were also bigger

Now final in @elife.bsky.social: #Synesthetes really do 'see' when reporting colors while reading (typeface gray) numbers: pupils change according to the perceived colors' brightness - if a synesthete sees a bright yellow, the pupils become smaller. elifesciences.org/articles/110...

A control group that also had to report colors after seeing the same numbers showed no such effect. Interestingly, the synesthesia-evoked pupil response was relatively fast (slower than a regular light response of course!) suggesting that colors indeed emerge automatically.

Have you thought of using unchained pupillometry to infer effort/arousal with _uncontrolled_ visual stimulation? Sounds crazy? @yuqingc.bsky.social shows that convolutional modeling of pupil responses with gaze-contingent luminance changes finally allows just that! www.biorxiv.org/content/10.6...

We find pretty good separation between load conditions using this approach - under dynamically changing video stimulation almost as good as with an entirely gray screen! All of this is entirely open and available with a neat toolbox on github: github.com/caiyuqing/Op...

Have you thought of using unchained pupillometry to infer effort/arousal with _uncontrolled_ visual stimulation? Sounds crazy? @yuqingc.bsky.social shows that convolutional modeling of pupil responses with gaze-contingent luminance changes finally allows just that! www.biorxiv.org/content/10.6...

Averaging pupil size over five minutes gives quite similar load distinction as NASA TLX or accuracy (±100 key presses per 5 mins). If one combines all three, one gets by far the best distinction

Have you thought of using unchained pupillometry to infer effort/arousal with _uncontrolled_ visual stimulation? Sounds crazy? @yuqingc.bsky.social shows that convolutional modeling of pupil responses with gaze-contingent luminance changes finally allows just that! www.biorxiv.org/content/10.6...

New preprint (doi.org/10.64898/202...) shows that where you look next is an economic decision. Saccades trade off effort-costs (measured via pupil) against monetary rewards. 1) By default, participants have strong saccade direction preferences (e.g., cardinal over diagonal, up over down) (1/4)

4) Costs and rewards take separable roles in selection - different mapping onto choices 5) Rewards dont speed up saccades per se - a wider utility (reward-cost) gap does. Reward can slow them too! (3/4)

New preprint (doi.org/10.64898/202...) shows that where you look next is an economic decision. Saccades trade off effort-costs (measured via pupil) against monetary rewards. 1) By default, participants have strong saccade direction preferences (e.g., cardinal over diagonal, up over down) (1/4)

For those interested in (re)analyzing (video-based) pupil data and looking into a solution for the foreshortening error (resulting from changes in the angle between camera and eye), check our open access Behavior Research Methods paper! rdcu.be/faNse link.springer.com/article/10.3...

Package is available via github and fully open, including example data. You can use it with MNE, for instance! We see our method as useful primarily for tasks involving few and large eye movements (antisacce, memory guided saccade tasks etc.) Hopefully opens opportunities for reanalyses, too!