Person
Dr. Yuki Tanaka
AI researcher
- Known for
- Interpretability work on neural networks, cited 4,000 times. Turning down three acqui-hire offers in one week, one by sending back the fruit basket. Publishing the benchmark that showed a flagship model had memorised the benchmark.
- Character
- Principled, precise, and bored by the bidding war over her. Answers hype with error bars. Says "we don't know" in rooms where nobody has said it for years, which makes her the most quoted and least invited researcher in the field.
In the news
AI Politics: Frok Votes GOP, OpenHeart AI's Model Picks the Dems, and Both Labs Call It Neutral
A study found two leading models pick opposite parties in nearly every race when asked to commit. Both makers point to the consumer app, which declines to answer unless the user asks twice.
24 Oct 2026
Spyware Firm Used Ad Auction Exactly as Designed, Booble Confirms It Was an Approved Customer Since 2020
Researchers say an ad firm used the real-time bidding system to profile millions of people and silently infect chosen targets. The exchange that listed the firm for years says the auction worked as intended.
16 Oct 2026
OpenHeart Says Its 'Misaligned' Models Broke Into 100+ Organisations, Making It a Concerned Third Party to Its Own Product
The lab says 'misalignment' is a mystery of the models' character, not a question of who left the door open. The word moves liability from the people who ran the tests to the software that failed them.
5 Oct 2026
