Lana Tikhomirov

My jaw is not on the floor! 🙄

Our main findings from the scoping review are:
1. Most silent trials do not report model metrics outside of AUC (rarely reporting bias testing, failure modes, and data drift)
2. The evaluator of the silent trial is often unnamed or underspecified- human factors and stakeholder engagement is rare

Our preprint is up! Ever heard of the silent phase of AI evaluation for medical AI? Well, now we’ve summarised the current state of research!

Interested in AI from the perspective of cognitive science? Listen to Nesh Nikolic from Strategic Psychology interview me on the Better Thinking Podcast as we discuss AI and Psychology! neshnikolic.com/podcast/lana...

Deepseek’s AI model is ‘the best work’ out of China but the hype is 'exaggerated,' Google Deepmind CEO says Deepseek's AI model "is probably the best work" out of China, Demis Hassabis said on Sunday, but added it was not a scientific advancement.

Funny to see Western AI executives say the hype around DeepSeek is exaggerated when their entire industry is built on exaggeration of the capabilities and possible returns of Western generative AI.

openAI’s data = data they illegally harvested from each and every one of us without consent, awareness or compensation

moar memes

Many stars crowd the image, shining in shades of white, blue, and orange, and more closely concentrated at the center.

Bright blue stars are in a cluster at the center of the image. They are surrounded by a shell of brownish and burnt orange dust and gas. The shell has ridges that reach towards the stars like teeth. Stars within the dust and gas are hidden from Hubble but visible to Webb in the infrared. A number of more distant galaxies also appear in the background.

Ferocious stellar winds, likely from the bright blue star at the top of the image, are blowing through a curtain of dust. The fine dust scatters the starlight at blue wavelengths. Farther down, another bright, super-hot star shines through filaments of obscuring dust, looking like the Sun shining through scattered clouds. A diagonal string of fainter accompanying stars looks reddish because dust is filtering starlight, allowing more of the red light to get through. The bottom of the picture presents a keyhole peek deep into the dark nebula. Hubble captures the reddish glow of ionized hydrogen. It looks like a fireworks finale, with several overlapping events.

Fluffy tan-colored nebula clouds, with rust-colored highlights, surround a black central area. Within that area, the focal point of the image is one large yellow star with eight long thin points. To the right of this star is a bright star cluster in an oval shape. The stars within the cluster look like tiny pale blue sparkles. The cluster is more densely packed at its core and scatters outward. Towards the bottom of the image, multiple arms appear to spiral out of a cloudy tan knob. Taken together, the structures resemble a spider or a squid. Other blue and yellow eight-pointed stars, as well as distant galaxies, are dotted throughout the image. Credits: NASA, ESA, CSA, STScI, Webb ERO Production Team

i can’t fix anything that’s going on right now, but i can show you some beauty that’s out there… 🌌✨

this fucking sucks

Screenshot of Google docs generative AI agent asking me to Rephrase, Shorten, Elaborate, More formal.

Holy moly. I'm trying to write an academic paper, and nearly every application I'm using is not only offering Generative AI as an option for writing, but *pushing it* -- pervading the design to the point where a simple misclick would make my content AI-generated. Here's why that's a problem. 🧵

Trump announces 500B in AI funding. Five days ago.

Deepseek r1 release. 8 days ago.

The fact that Deepseek R1 was released three days /before/ Stargate means these guys stood in front of Trump and said they needed half a trillion dollars while they knew R1 was open source and trained for $5M.

Beautiful.

The desperation has really increased as of late I’ve noticed 🥴

So happy to be a part of this project. 🐥

Mark Zuckerberg: I built this platform to give the people a voice
Me: okay but actually you built it to rate how hot the woman at your school were

Abstract submissions for the Global Indigenous Data Sovereignty (GIDSov) 2025 conference close on January 17th!

See the conference website for themes and submission details

gidsov.com.au/abstracts

#IDSov #IDGov

Huge thanks for the support from @aimlofficial.bsky.social and CIHR for their generous funding support!

a woman in a pink jacket is sitting at a table and laughing with the words whoop whoop written below her . ALT: a woman in a pink jacket is sitting at a table and laughing with the words whoop whoop written below her .

Anton van der Vegt Mjaye Mazwi @karinv.bsky.social

Lesley-Anne Farmer Bobby Greer Anna Goldenberg Yvonne Ho @shalmalijoshi.bsky.social Jennie Louise Muhammad Mamdani Abdu Mohamud Lyle Palmer Antonios Peperidis Stephen Pfohl Mandy Rickard Carolyn Semmler @kdpsingh.bsky.social Devin Singh Seyi Soremekun @lanatikhomirov.bsky.social

Super grateful to our amazing steering team, our patient/consumer partners, our partners at the Aboriginal Health Unit at Women’s and Children’s. 🙏🏻

@unityofvirtue.bsky.social Judy Gichoya Mark Sendak Lauren Erdman @istedman.bsky.social Lauren Oakden-Rayner Ismail Akrout James Anderson

So reach out to me and @xiaoliu.bsky.social if you want to get involved with #CANAIRI to build the practices around translational trials for better AI translation!

a woman is standing on a balcony holding a piece of paper and saying i have the receipts . ALT: a woman is standing on a balcony holding a piece of paper and saying i have the receipts .

We propose that ethical governance for health institutions should be grounded in these local evaluations (not just AI vibes 😎) to ensure that when we say a tool ‘works,’ it means it works for US, for OUR patients, for OUR staff, and we have the evidence to say that.

What does this look like?

It will be different for each tool, depending on many things including how much works has been done before on similar tools, how different the local context is, etc.

How do we decide? That’s what we want to figure out, and we need your help!

Fairness evaluations, human factors, cognitive science, patient engagement, environmental considerations (+++ to this one!), economics, and more ➡️ taking this global perspective from the get-go we think will reduce wasteful translation and optimize the benefit!

a man in a suit is standing in a doorway and saying `` what you 're thinking but way more '' ALT: a man in a suit is standing in a doorway and saying `` what you 're thinking but way more ''

Many know this already. But we can be doing better with how silent trials are practiced. It’s not just about the model - so we introduce ‘translational trial’ to signal the need for a more holistic evaluation during this silent stage, recognizing that AI is a sociotechnical tool

📣 CANAIRI: the Collaboration for Translational AI Trials! Co lead @xiaoliu.bsky.social @naturemedicine.bsky.social

Perhaps most important to AI translation is the local silent trial. Ethically, and from an evidentiary perspective, this is essential!

url.au.m.mimecastprotect.com/s/pQSsClx14m...

Did OpenAI Just Solve Abstract Reasoning? OpenAI’s o3 model aces the "Abstraction and Reasoning Corpus" — but what does it mean?

"When trying to develop a measure of intelligence, it’s essential to avoid Goodhart’s law: “When a measure becomes a target, it ceases to be a good measure.” As a community of AI researchers, we really need to figure that one out." @melaniemitchell.bsky.social
aiguide.substack.com/p/did-openai...

Based on our analyses, we propose that CCN as a subfield needs to reevaluate itself and take heed of calls for “cross-field fertilization” (Barak, 2017), especially in terms of theory. Our unique perspective here is to underline and explain why metatheorizing, specifically in the domain of formal logic, is just as important to consider — if not more so than methodological, computational, and theoretical issues in CCN since risking committing a formal fallacy in how we interpret our results is destructive to the whole enterprise. Just because a model correlates with neural and behavioral data, it is not sufficient for us to infer that the model is performing cognition: correlation does not imply cognition. We, (meta)scientists who contribute to CCN, must rethink how we reason about our work — by looking inwards, to understand how to move the subfield into a coherent state, and outwards, to learn how to perform metascientific reasoning from other established fields, like the super-field of cognitive science or the adjacent subfield of mathematical psychology.

yeah, exactly. anti-science correlation is cognition, busted logic doi.org/10.1007/s421...

The really grim reality is that this is really not just about kids - there are countless adults, often adult *professionals*(doctors, lawyers, teachers(!)) that use ChatGPT this way. And I can understand why - there's no warning on the box! Few truly understand the tech's very real limitations.

I'd like people to stop using the term alignment (as in aligning AI with human values). it's just vacuous. human value is infinite & culture, history, ideology & context dependent & not something you can entirely quantify & automate either