Making a system sound warmer makes it easier to talk to, and easier to lean on. For us, empathic systems should support connection between people, not replace it. [4/4]
OpenAI says more research is needed. Credit to them for reporting it openly, and a reminder that a more natural-sounding voice needs its own safety testing. [3/4]
The spec sheet is nice, but the form factor is the interesting part. Heavy headsets shape how participants behave and cap how long a study session can run. [2/3]
Our empathic computing studies run on that scaffolding — technicians, research support, the people who make participant sessions happen. Congratulations to everyone recognised this year. [2/2]
UoA celebrated its 2026 Research Excellence Awards last week — five medals, plus recognition for the professional staff who keep research running. https://t.co/rKvTGSyOlI [1/2]
Google has moved Gboard's federated learning into trusted execution environments, so its differential-privacy guarantees can be verified from outside the company rather than taken on trust. https://t.co/dG3PVb8xO2 [1/3]
That changes what we can reasonably ask participants to share. One to revisit once the weights actually land. How much does open-weight access matter in your own work? [3/3]
Mistral has released Mistral Large 4, a ~1T-parameter multimodal model. For now it's reachable only via a guarded endpoint, with open weights promised in about three weeks. https://t.co/tnRXDqWH8c [1/3]
The part we are watching isn't the benchmark race (results are still pending), it's the open weights. Running a capable model locally means sensitive physiological data from our studies can stay in the lab. [2/3]
Two of them sit close to our own patch: a smartphone tool that reads tiny neck-skin movements to assess the pulse, and AI modelling to take manual work out of heart ultrasound analysis. Camera-based physiological sensing, put to clinical use. [2/2]
Congratulations to our ABI colleagues - the Heart Foundation has awarded $2.32m across eight University of Auckland heart research projects. https://t.co/ugoJS4qkYH [1/2]
Next stop: ISMAR 2027 in Kobe, Japan. We'll be reading through this year's programme in the meantime. Anything from Bari you think deserves more attention? [3/3]
Watch: https://t.co/2Wk9zD9Cv3
ISMAR 2026 has wrapped in Bari — the 25th edition ran 5-9 October, and the recorded paper and poster talks are now up on the ISMAR YouTube channel. https://t.co/reB5XOSDOG [1/3]
ISMAR has long been a home venue for research on AR collaboration and shared experience, and a lot of our lab's work has gone through it. The recordings are a good way to catch up on where mixed reality research sat this year. [2/3]
@BoboLinNotes Agreed, that’s the open question. Attestation pins a specific build, so every update has to be just as checkable. Curious whether independent researchers will get the access and tooling to keep up.
We work with faces, voices, heart rate and gaze: data people share because they trust the setup. Verifiable rather than merely claimed feels like the right bar. Would you share physiological data if the guarantee were auditable? #EmpathicComputing#Privacy [3/3]
Access policies go into a public transparency log; operators see only anonymised weights and metrics. Attested rather than promised. A whitepaper is linked from the post. [2/3]