Stian Reimers
@steer-rhinos.bsky.social
Behavioural scientist and/or professor of psychology, interested in decision making in its many forms. Currently also director of digital innovation at City St George's School of Health and Medical Sciences.
A festive project here, bringing together animation, MIDI input, USB relays to contol low-voltage devices, and smart device management, all run from a browser with JavaScript. The owl tells the students what Chrismas melody to play and when they get it right, the piano gets more festive...
December 23, 2024 at 2:20 PM
A festive project here, bringing together animation, MIDI input, USB relays to contol low-voltage devices, and smart device management, all run from a browser with JavaScript. The owl tells the students what Chrismas melody to play and when they get it right, the piano gets more festive...
Making most of my first video post to show this magnificence from tonight's Taunton carnival. Something I grew up with, but I still get that sense of awe as one of these comes around the corner.
October 19, 2024 at 9:40 PM
Making most of my first video post to show this magnificence from tonight's Taunton carnival. Something I grew up with, but I still get that sense of awe as one of these comes around the corner.
Waking up this morning approaching Penzance on the Night Riviera sleeper. Really is a wonderful way to travel, albeit one entirely unsuited to sleeping.
October 19, 2024 at 6:36 AM
Waking up this morning approaching Penzance on the Night Riviera sleeper. Really is a wonderful way to travel, albeit one entirely unsuited to sleeping.
We ran it a couple of weeks ago with 70 students as part of an onboarding session. Feedback, both on usefulness and realism, was positive, so we'll be updating and running again in the winter with the next intake. Can give more practical details or dm a link to have a go if useful...
October 17, 2024 at 9:02 PM
We ran it a couple of weeks ago with 70 students as part of an onboarding session. Feedback, both on usefulness and realism, was positive, so we'll be updating and running again in the winter with the next intake. Can give more practical details or dm a link to have a go if useful...
In AI roleplay news, I've built a web app for law students to prepare for consultations with clients in our law clinic. Exciting developments: (a) a point-and-click interface for selecting cases; (b) using openAI's whisper API so users can speak/listen rather than type. Built with GPT-o1 support.
September 29, 2024 at 8:27 PM
In AI roleplay news, I've built a web app for law students to prepare for consultations with clients in our law clinic. Exciting developments: (a) a point-and-click interface for selecting cases; (b) using openAI's whisper API so users can speak/listen rather than type. Built with GPT-o1 support.
Today has been induction week escape room day: Students learn about the university to crack combination boxes, decode phone numbers, and scan barcodes to open the treasure chest. All modelling collaborative autonomous learning. Six hours and 200 - all successful - students later, ready for a pint.
September 26, 2024 at 4:15 PM
Today has been induction week escape room day: Students learn about the university to crack combination boxes, decode phone numbers, and scan barcodes to open the treasure chest. All modelling collaborative autonomous learning. Six hours and 200 - all successful - students later, ready for a pint.
Excitement: I now have a large language model comparable to ChatGPT running on my own PC. As in, without an internet connection. It's a capable model, which opens up use cases in universities where you wouldn't want to pass sensitive data outside the organisation.
September 16, 2024 at 9:17 PM
Excitement: I now have a large language model comparable to ChatGPT running on my own PC. As in, without an internet connection. It's a capable model, which opens up use cases in universities where you wouldn't want to pass sensitive data outside the organisation.
Nice to wake up and see a piece about our BICYCLE project in the most read articles on the BBC website. We're looking at development effects of lockdown on children's cognitive and linguistic development. Also there's actually a photo of one of the tasks I developed! www.bbc.co.uk/news/article...
September 1, 2024 at 7:41 AM
Nice to wake up and see a piece about our BICYCLE project in the most read articles on the BBC website. We're looking at development effects of lockdown on children's cognitive and linguistic development. Also there's actually a photo of one of the tasks I developed! www.bbc.co.uk/news/article...
The most impressed I've been with GPT-4 is building browser-based cognitive tasks. I just prompted "...use HTML, css, and JavaScript to implement a Corsi visuo-spatial short-term memory task.", and it's produced a working version (after 1 x feedback). At no point did I explain what a Corsi task was.
November 1, 2023 at 4:00 PM
The most impressed I've been with GPT-4 is building browser-based cognitive tasks. I just prompted "...use HTML, css, and JavaScript to implement a Corsi visuo-spatial short-term memory task.", and it's produced a working version (after 1 x feedback). At no point did I explain what a Corsi task was.
As a churchgoing child I loved how the year was marked with the different liturgical colours: The excitement in advent as the altar cloth changed from green to purple meaning Christmas was coming. I miss that. Anyway, here I am on Nov 1 switching over my coffee syrups from Pumpkin Spice to Cinnamon.
November 1, 2023 at 10:04 AM
As a churchgoing child I loved how the year was marked with the different liturgical colours: The excitement in advent as the altar cloth changed from green to purple meaning Christmas was coming. I miss that. Anyway, here I am on Nov 1 switching over my coffee syrups from Pumpkin Spice to Cinnamon.
It’s maddeningly inconsistent – if you ask for a small change to an existing output, it generates a completely new output. I asked it to convert the hot air balloon to monochrome. Result: A different hot air balloon (in monochrome).
October 30, 2023 at 7:15 PM
It’s maddeningly inconsistent – if you ask for a small change to an existing output, it generates a completely new output. I asked it to convert the hot air balloon to monochrome. Result: A different hot air balloon (in monochrome).
It’s frustrating with complex stimuli – the GPT part captures requirements well, but often the image doesn’t match, and GPT can't see what it’s generated. Here it says “On the left, the joyful lion; in the center, the cheerful zebra; on the right, the lion and zebra in a warm hug”. Which it ain’t.
October 30, 2023 at 7:14 PM
It’s frustrating with complex stimuli – the GPT part captures requirements well, but often the image doesn’t match, and GPT can't see what it’s generated. Here it says “On the left, the joyful lion; in the center, the cheerful zebra; on the right, the lion and zebra in a warm hug”. Which it ain’t.
An advantage over midjourney is that it has the whole conversation history, so it’s quick to iterate. I needed a picture of a chocolate cockroach (illustrating an experiment fwiw); output was too gruesomely realistic so I said “make it more chocolate and less cockroach”, which gave what I needed:
October 30, 2023 at 7:12 PM
An advantage over midjourney is that it has the whole conversation history, so it’s quick to iterate. I needed a picture of a chocolate cockroach (illustrating an experiment fwiw); output was too gruesomely realistic so I said “make it more chocolate and less cockroach”, which gave what I needed:
I couldn’t get it to do transparent backgrounds – it does a pastiche of a transparency checkerboard, except it’s actually a checkerboard in the image, with glitches, so is worse than a white background, which it can do. So needed to remove background elsewhere.
October 30, 2023 at 7:09 PM
I couldn’t get it to do transparent backgrounds – it does a pastiche of a transparency checkerboard, except it’s actually a checkerboard in the image, with glitches, so is worse than a white background, which it can do. So needed to remove background elsewhere.
It can be hard to get consistency of styling if you want to create a set of similar stimuli. One way is to have them all created in the same image and then separate them downstream. (Another is to specify a very precise style but I need to explore that further.)
October 30, 2023 at 7:08 PM
It can be hard to get consistency of styling if you want to create a set of similar stimuli. One way is to have them all created in the same image and then separate them downstream. (Another is to specify a very precise style but I need to explore that further.)
Lessons from a foray into using ChatGPT with DALL.E 3 plugin for academic work:
It's great for generating simple stimuli for experiments. I needed a rocket to move up the screen as children complete a cognitive task. This is better than I'd have made and took 30 seconds to create.
It's great for generating simple stimuli for experiments. I needed a rocket to move up the screen as children complete a cognitive task. This is better than I'd have made and took 30 seconds to create.
October 30, 2023 at 7:06 PM
Lessons from a foray into using ChatGPT with DALL.E 3 plugin for academic work:
It's great for generating simple stimuli for experiments. I needed a rocket to move up the screen as children complete a cognitive task. This is better than I'd have made and took 30 seconds to create.
It's great for generating simple stimuli for experiments. I needed a rocket to move up the screen as children complete a cognitive task. This is better than I'd have made and took 30 seconds to create.
Finding GPT-4 useful for teaching activities. I saved my first week's lecture as a pdf, used the AI PDF plugin in ChatGPT, gave the URL for my pdf and asked for 20 multiple choice questions. About 18 of those were great - plausible alternatives, and even showed which slide the question came from.
October 8, 2023 at 11:05 AM
Finding GPT-4 useful for teaching activities. I saved my first week's lecture as a pdf, used the AI PDF plugin in ChatGPT, gave the URL for my pdf and asked for 20 multiple choice questions. About 18 of those were great - plausible alternatives, and even showed which slide the question came from.