Sichu Lu
banner
sichulu.bsky.social
Sichu Lu
@sichulu.bsky.social
e-tsundoku, supplementary info: nlab fan account, arxiv surveyor, pubmed enjoyer, two culture bridger, vacuous high gossiper, dearth of any domain expertise, reluctant g theorist, gpu poor
Pinned
pgp fingerprint: EC170B8FB69F48E6C14B6BF95BE5E2C1CD617547
you are in the wrong dating market employing the wrong types of mating calls .. and unfortunately all your swiping left and right mistakes have failed to cancel out
December 14, 2025 at 4:55 AM
Reposted by Sichu Lu
I kinda of am annoyed at the constant reinvention (deep irony but whatever)
December 14, 2025 at 3:37 AM
Reposted by Sichu Lu
A lot of so called ai thinkers who haven't even read the sequences and definitely did not read years of lesswrong posts. Agree or disagree about doomerism like very little of the current conversation about economics is actually new
December 14, 2025 at 3:36 AM
stolen but amazing tweet what other things are like this? @norvid-studies.bsky.social
December 13, 2025 at 2:22 PM
Reposted by Sichu Lu
It is a lot harder to put lies in an LLM than in a newspaper.
December 12, 2025 at 9:21 PM
December 12, 2025 at 11:47 PM
i have a lot of questions
December 11, 2025 at 4:10 PM
December 11, 2025 at 3:46 PM
Reposted by Sichu Lu
placed myselfs with 95% confidence interval.
December 10, 2025 at 9:39 PM
December 11, 2025 at 3:23 PM
Reposted by Sichu Lu
Electron microscopy is @nature.com's Method of the Year! 🧠🔬🏆

@natmethods.nature.com shares why MICrONS and FlyWire are proof that #ElectronMicroscopy is helping us see the brain and its incredible complexity at a new level: https://www.nature.com/articles/s41592-025-02988-6

🧠📈 #studyBRAIN
December 9, 2025 at 8:14 PM
Reposted by Sichu Lu
are better-performing face recognition models more human-like? turns out: NO

in terms of how we see/treat different faces as similar/different to each other, there seems to be tradeoff: better models are LESS human-like

so they already work in some 'alien' ways...

osf.io/preprints/ps...

🧠📈🧠🤖🧠💻
OSF
osf.io
December 10, 2025 at 12:34 AM
i should make an alt account that is just wikipedia
The Ones Who Retreat To Wikipedia
December 10, 2025 at 2:56 AM
Reposted by Sichu Lu
New ChatGPT safety filters really don’t like targeted therapy delivery by cell or tissue type, even when the specific targeting mechanism / cell type / payload are all pretty different
December 2, 2025 at 3:03 AM
www.alignmentforum.org
December 8, 2025 at 9:53 PM
and now for some science slop www.nature.com/articles/s41...
December 8, 2025 at 8:39 PM
December 8, 2025 at 6:55 PM
burned again by trusting op was not strawmanning their cite award goes to me
December 8, 2025 at 6:20 PM
Reposted by Sichu Lu
Instagram users may overestimate the extent to which they are addicted to the platform, according to research published in Scientific Reports. The findings suggest that for most social media users, excessive use is driven by habit rather than genuine addiction:

#PsychSciSky #Psychology
Overestimates of social media addiction are common but costly - Scientific Reports
Overuse of social media is commonly termed a behavioral addiction. However, there is reason to believe that users overestimate their addiction; instead, frequent use largely builds habits to automatically activate, scroll, post, and react on social media. The present research (N = 1204) explored the implications of overusing the addiction label. In Study 1, a national quota sample of Instagram users (N = 380) overestimated their addiction to the app, and those who perceived themselves as more addicted (but not more habitual) experienced less ability to control their use. We show that the perception of addiction likely arises from popular media’s frequent labeling of social media as addictive (vs. habit forming). Study 2 (N = 824) demonstrated experimentally that framing frequent Instagram use as an addiction has deleterious consequences for user self-efficacy, including reducing perceived control over social media use and increasing self-blame for overuse. In addition, misperceiving excessive social media use as addictive potentially diverts users from effective strategies that could be used to curb overuse habits.
spklr.io
December 1, 2025 at 10:57 AM
Reposted by Sichu Lu
California is famous for its redwoods and sequoias, but it also has some incredible oaks, and the woodpeckers know.
Trypophobia alert 🫣

This is an Acorn Woodpecker (Melanerpes formicivorus) from Central America & Western United States, filling his acorn larder

If Trypophobia is new to you, it is an aversion to the sight of repetitive patterns or clusters of small holes or bumps
December 7, 2025 at 6:11 AM
Reposted by Sichu Lu
Proud to launch an international multidisciplinary project to look deeper into causality and evidence assessments regarding uses of psychedelics in medicine www.ipeawg.org
🧪 🧠📈 #philsci
IPEA-WG – Independent Psychedelic Evidence Assessment Working Group | Discover Psychedelic Insights Today
Explore evidence-based policies on psychedelics and mental health, featuring scientific research, expert insights, and multidisciplinary perspectives from IPEA-WG.
www.ipeawg.org
December 6, 2025 at 12:43 PM
Reposted by Sichu Lu
That Time I Got Reincarnated in a Chatbot
December 6, 2025 at 9:15 PM
Reposted by Sichu Lu
These kicks look uh pretty lethal
December 6, 2025 at 9:29 PM
Reposted by Sichu Lu
I think LLMs dynamically generating game dialogue is going to be trickier than people assume not because LLMs are bad at it, but because game dialogue having four options mapped to four buttons is just so much faster and easier than, what, typing your full response every time? microphone input?
December 6, 2025 at 10:39 PM
Reposted by Sichu Lu
Another year almost over and I still haven’t made any progress on my book project. Sigh. #birds
December 7, 2025 at 12:29 AM