🎓 Assistant Professor @UniLeiden. Affiliated Researcher @LeidenDH. PhD @UofT_PolSci. 👨🔧Techno-optimist.
Leiden, the Netherlands
Joined September 2009
- Tweets8.7K
- Following3K
- Followers3.7K
- Likes20.1K
Steven Denney retweeted
Thanks for the plug
@analisereal ! Let me know if you give librepaper.org a shot, @jt_kerwin , I'm really keen to get bug reports, feature requests, or design idea.
Steven Denney retweeted
With still roughly $1.54T in bitcoin available to steal worldwide, and with the progress of AI to make it possible to steal more efficiently, the North Korean economy has still vast potential for growth
Steven Denney retweeted
Grant applications have become a lottery. A 60% growth in applications to one of the world’s most prestigious research-grant programs is resulting in an exceptionally low success rate
I think this is one of the major problems of AI--it's creating tsunami of grant applications and turning the decision process into chaos.
Grant decisions already felt fairly random and this seems to make the system worse.
nature.com/articles/d41586-0…
Steven Denney retweeted
AI could push back against public irrationality like conspiracy theories, fake news, & medical quackery. Whereas social media optimize for sheer engagement (clicks, likes, & reposts), AI is sold as a useful product that aids decision-making in the real world. For all the problems of LLMs, they're less likely to send users down a conspiratorial rabbit hole. This may restore the influence of epistemically responsible expertise. So argues the philosopher Dan Williams @danwilliamsphil open.substack.com/pub/persua…
This is the way.
“It’s the only class I took at Harvard where people would call each other out for not having done the reading.”
Harvard students are lining up to take Hum 10:
“Applications (essays required) shot up 40% this fall with 180 incoming freshmen vying for 90 spots.”
“The class is definitely not for padding the GPA. Students read a hefty book a week and write regular essays. There is no pass/fail option. Digital devices are verboten, and attendance is taken in every lecture (pink slips for those who arrive late).”
“In their applications and in interviews, students expressed the urge to be forced to tackle classics they otherwise wouldn’t read, the need to combat frayed attention spans and the desire for an antidote to The Machine. Several mentioned with disdain that people who aren’t in the class wondered why they didn’t ‘just Chat the books,’ i.e., use AI.”
Steven Denney retweeted
AI-manipulated images and videos making North Korean artillery appear to be just outside a Starbucks in Gimpo, Gyeonggi Province, have been circulating on social media, according to a report by NK News.
koreaherald.com/article/1091…
Steven Denney retweeted
Artificial Analysis finally moves to closed benchmarks (goodbye GDPval-AA), and finds that open weights models are falling further behind than they had previously measured. Take with a grain of salt, but matches the general vibes that we haven't seen a Fable-class open model.
Open weights models fall further behind in v5. The leader, Kimi K3.1 Pro (weights not yet released), scores 46, 12 points behind the leader, compared with a 3-point gap in v4.3. Of models with released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) score highest at 32, and Claude Haiku 5.5 (max) outscores all of them.
The gap is widest on HLE-Diamond and Terminal-Bench-Science, where Kimi K3.1 Pro scores 37% and 34%, compared with 68% for Claude Opus 5.5 (max) and 63% for GPT-6 Astra (max).
Steven Denney retweeted
Jobs are modeled as bundles of tasks. But as jobs change, how do we keep those task lists up to date?
Workers often experience new responsibilities and tools before they reach their job descriptions.
We start by listening to them. 🧵
Steven Denney retweeted
New Anthropic research: How AI agents shape experts' judgment calls
When experts delegate work to an agent, do they still make the key decisions? We tested this in a randomized trial with 50 professional data analysts.
Steven Denney retweeted
For many young people whose brains have been thoroughly rotted by TikTok, Snapchat, and the like--it is quite hard, yes. And we as educators must ask them to do the hard thing rather than give them the customer what they think they want.
Steven Denney retweeted
Out today in Nature Human Behaviour: our review of large language models in computational social science, with @xu_fengli, @yongli and colleagues at @Tsinghua_Uni, @hkust & @MSFTResearch Asia. A thread on what we found, including where the field is fooling itself.
Read @ doi.org/10.1038/s41562-026-0…
Steven Denney retweeted
People in rich countries are unhappy. In the US they're truly miserable.
Sentiment fell more in the US since 2019 than any other country.
US sentiment at the 0th percentile of pre-COVID history, others generally around 25th percentile.
And our economy doing better than most!
Steven Denney retweeted
It’s a shame that we’re not making better use of AI’s amazing fact-checking and argument-checking abilities.
Platforms or third parties should build better tools that identify BS immediately and saliently.
Steven Denney retweeted
Everything at the end will look like a Nature/PNAS paper: a quick 6-pager for humans, and a humongous 150-page supplement with all possible experiments and analyses (mainly for the machines to read and review). Together with an automatic verification/reproducibility package that ensures that whatever the paper says is verifiable.
And at the end everything will seem like before, but underneath everything will have changed.
Steven Denney retweeted
We benchmarked 17 open OCR models on OmniParseBench, our new OCR benchmark that builds on AllenAI's olmOCR-bench.
Steven Denney retweeted
Everyone: "Vibe coding is causing all these security bugs"
Reality: PHP, Node, Linux kernel, GitLab. All hand-written code by humans, some of it 20 years old.
AI isn't creating the slop. It's finding it 🫠
Stop with the sensationalism. It’s not “over”. Good grief.
Steven Denney retweeted
This paper is what I have been yelling about for a few years now: the true danger in empirical social sciences is more often "researchers degrees of freedom" in design, data curation, measurement, choice of method.
The big threat is not outright fraud or p-hacking (although these are problems) or even biases against publishing null results. Its those hundreds of small choices researchers make when preparing their papers -- none of which are sins of omission or commission. They're good faith human decisions with no malice.
Yet, two equally good faith moves can generate different results. This apparently forthcoming paper in Journal of Economic Literature asked 146 research teams to estimate the same causal effect using the same underlying data. In unconstrained setups, they arrived at massively different results. Even in more constrained setups (with more directives), they showed differences. All of which came from picking the design and curating the data (cleaning). Huge!
I would reformulate the problem as unexamined researcher degrees of freedom, rather than methodological freedom itself! Indeed, what we need is a strong contestability (and good faith) to point out what could be happening.
