I have a bachelor’s in CS. Looking for a job!
find me anywhere in linktr.ee/papetoast
papetoast
Karma: 992
I think you forgot to link the post: https://isene.org/2026/05/Audience-of-One.html
A trick for mentally calculating squares of two digit numbers (via bilibili):
Basically, choose
Example:
For 26, the closest multiple of 10 is 30, so
This algorithm can be extended recursively for squares of n digit numbers, though it is seems less useful.
There are more details in Table 5 Decision Thresholds that I didnt quote. Basically 80%+ is rejected.
NeurIPS 2026 is using Pangram to reject LLM writing
This year, the NeurIPS 2026 Position Paper Track made the decision to require that all papers be substantially human-written, with AI used for only copy-editing or similar peripheral changes to the main text.
To assess if authors were largely abiding by this policy, we partnered with Pangram
178 submissions (18.4% of all submissions) will be desk rejected
123 submissions (12.7%) will be requested to provide evidence of substantial human engagement or risk a desk reject.
Conference
# Papers
Pangram AI Score
≥ 50%
≥ 90%
= 100%
NeurIPS PPT 2025
536
28.5%
11.9%
8.2%
NeurIPS PPT 2026
971
70.5%
42.7%
28.2%
NeurIPS D&B 2025
996
5.6%
0.8%
0.4%
NeurIPS E&D 2026
996
43.7%
9.3%
2.1%
FAccT 2022
159
0.0%
0.0%
0.0%
FAccT 2025
204
1.0%
1.0%
0.0%
A null effect on pain relief from acupuncture in a pre-registered, improperly double-blinded study (via National Geographic via Facebook)
I didn’t read the paper beyond AI summary, I read the national geographic article in full (which is misleading according to claude)
Selected AI Summary (Full Transcript)
Short version: the underlying paper is methodologically honest and reasonably well-run, but its evidence for acupuncture having a specific (beyond-placebo) effect is weak. The National Geographic article substantially oversells it — it leads with a fragile secondary finding and quietly drops the fact that the trial’s primary outcome was null.
The pre-registered primary outcome was vulvar pain (Average Pain Intensity) at end of treatment, real needles vs. sham. Result: no difference. Effect size 0.06, 95% CI −0.36 to 0.24, p=0.70. Secondary outcomes (dyspareunia, sexual function) also null. Response rates were essentially identical: 58% acupuncture vs. 57% placebo. So the confirmatory test the trial was built around failed.
The only “win” came from Aim 2, a secondary duration analysis: among people who responded, the placebo group relapsed to baseline pain faster (hazard ratio 2.72, 95% CI 1.13–6.54, p=0.017). That single number is the source of the article’s “12 weeks vs. 4 weeks” headline.
Blinding failed in the active arm — and this is the deepest issue. The whole point of the design is to strip out expectation. Bang’s index showed the placebo group was properly blinded (~0, non-significant), but the penetrating-needle group was not: acupuncturists guessed correctly far above chance (index 0.43–0.58, p<0.001) and so did the real-needle patients (0.34–0.35, p<0.01). So in a trial designed to isolate specific effect from placebo, practitioners and patients in the active arm often knew it was real. Differential expectation conveyed nonverbally could produce exactly the durability gap they saw — with zero physiological mechanism. The authors acknowledge this. It’s the single best alternative explanation for the only positive result.
I don’t really want to write this quick take, but omitting this negative result would filter evidences beyond my comfort.
For the record, I didn’t downvote you. I don’t live in the US and don’t find it immediately worthwhile to understand. I won’t verify the prompt’s truthfulness, but the prompt is biased even if it is all true facts, just by the way it demonstrates the user’s position on the matter. Biased in the sense that it will predictably cause AIs to lean towards one position more than the other.
After 2020 Honor is no longer directly nor indirectly controlled by Huawei, per wiki and this content farm post that I googled and this random wiki in Chinese
https://claude.ai/share/8940c08e-c01a-4c41-af9d-6eb77c0c6cbd
though, asking AI with such a biased prompt is a bad idea, so I refuse to read the output beyond a skim nor write about my reaction after reading. It also feels disrespectful that you didn’t even offer your opinion and demands the reader’s opinion.
wow Snopes’ journalism is great
As a man, I find it difficult to be consciously sympathetic of the strength difference. The force that I use to turn the faucet knob to a comfortable, definitely closed position actually requires an uncomfortable amount of force to open for my grandma. Would be interesting if someone made bottles and knobs that are 2x harder to open than usual...
bump
Related: How Your Survey Responder Lies
Aella didn’t even mention Lizardmen’s Constant once, but she made a bunch of twitter polls, which shows a different percentage of liars depending on the seriousness of the question and the survey.
I think probably the immediate visibility and informality of the twitter poll gives people a jolt of delight to fuck with the results, as long as fucking with the results would be funny. 1+1 = 3 is hilarious when it’s in front of a crowd and everybody sees the stupid percentages being stupid, but it’s boring when it’s inside the solemn walls of a researcher’s guidedtrack url.
(3³ − 23) ÷ 2 = 3, however, is deadly serious. Nobody finds that funny, crowd or not. It’s the one question where the twitter poll and survey question collapsed to almost exactly the same % of correct answer.
There is also this showing Lizardmen’s Constant = 10%, showing how much the “constant” can vary:
In a new study my collaborators and I focus on this hypothesis. We ask participants if this statement is true or false: “The Canadian Armed Forces have been secretly developing an elite army of genetically engineered, super intelligent, giant raccoons to invade nearby countries”.
We found that 10% of participants endorsed this statement (i.e., they selected “Definitely true” or “Probably true”) and that this was a strong predictor of endorsing six pre-existing conspiracy theories, including two that that directly contradicted each other.
I am increasingly having trouble discerning AI writing from human writing in the past year. It went from glaringly obvious to being possible to miss even if I put in effort analysing. I feel worried.
Edit: Yes, I know some, maybe most of you, can still smell out LLMs accurately, but I am getting worse at it.
prompted by this article, which is fully AI generated according to Pangram
As a non-american, Lighthaven East sounds like you’re building in Asia
Collecting opinions on whether data centers in space is a good idea (parent index):
https://taranis.ie/datacenters-in-space-are-a-terrible-horrible-no-good-idea/
https://news.ycombinator.com/item?id=46876105 (People saying why it won’t work)
https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/
https://arstechnica.com/space/2026/03/orbital-data-centers-part-1-theres-no-way-this-is-economically-viable-right/
https://www.seangoedecke.com/space-ai-datacenters-do-not-have-a-cooling-problem/ (Why cooling is possible in space)we’d need 250,000 square metres of radiation area. The largest current radiator in space is probably the ISS, at around a thousand square metres. Is scaling that up by 250x a lot? Yes, but it’s not necessarily ridiculous.
https://stratechery.com/2026/the-spacex-ipo-and-data-centers-in-space/
There is no reason that space data centers would look like data centers on earth. What makes far more sense is to think about an individual satellite as something akin to a rack. Right now the largest Starlink satellite in orbit is the V2 Mini Direct-to-Cell, which measures 7.4 meters by 2.7 meters by 0.3 meters (estimated); an NVL72 rack from Nvidia, meanwhile, measures 2.2 meters by 1.1 meters by 0.6 meters, so we’re already in the right size range. The V2 Mini Direct-to-Cell consumes (and dissipates) up to an estimated 25kW of energy; the NVL72 up to 135kW, and it can fit a 1 trillion parameter model quantized to FP4.
The big shortcoming for a rack-satellite is power and its dissipation, but going from 25kW to 135kW is certainly within the realm of possibility — and given that you don’t need much of the cooling and power distribution usage on earth, something closer to 100kW might deliver similar performance.
https://www.lesswrong.com/posts/65ECgHzWxTRvt8XWK/will-we-really-put-data-centers-in-space
The main case for ODCs ((orbital data centers)) is the cost of energy: space solar panels in the right orbits receive more constant and intense sunlight compared to Earth. Moreover, ODCs don’t currently face the same permitting and regulatory delays as on Earth, cause fewer ongoing environmental harms compared to grid or onsite natural gas-powered data centers, and may be more secure against data exfiltration. We find that the cost-competitiveness case for ODCs depends almost entirely on Starship achieving reusability comparable with what SpaceX achieved with Falcon: space-based solar reaches cost parity with present-day off-grid terrestrial power continuously at roughly $250/kg to orbit, and becomes cheaper than any current terrestrial energy source at around $50/kg, from the present-day launch cost of roughly $1,500/kg. Radiative cooling, often cited as a fatal obstacle, appears surprisingly manageable — potentially even cheaper than on Earth. However, ODCs may require substantial (perhaps ~38%) extra non-compute hardware (like solar, racks, and cooling) over 5 years to compensate for their inability to swap out failed chips, and inter-satellite bandwidth limitations likely confine ODCs to inference workloads, at least early on.
Assuming no transformative AI, but continued demand for data center buildout, we estimate that ODCs are unlikely to represent a meaningful share of compute before 2030, but become cost-competitive with present-day terrestrial data centers within 3–5 years if Starship development stays on track.
For missile defense
https://www.lesswrong.com/posts/MEBcfgjPN2WZ84rFL/o-o-s-shortform?commentId=i3pG9mLXuQkAfxityIt seems much easier to shoot down a defenseless, slow-moving thing in low earth orbit than something on earth or underground (which could be covered by SAMs, patrolled by fighter jets, shielded by thick cement).
https://www.lesswrong.com/posts/Y5cQYKYwAb2WwXXQQ/tech-i-m-skeptical-of-and-why#Space_data_centers: I find this unconvincing with biased arguments all the way
space data centers are not cost-competitive with terrestrial data centers.
My argument examines the different parts of a datacenter (chips, interconnect, comms, cooling, energy, etc.) and shows that in space the cost per unit of performance is worse at every step. If every step is more expensive in space, then terrestrial data centers must be cheaper.
He provided a lot of links to read though (I didn’t read them):
Further reading on space data centers, with an emphasis on good technical arguments and actually doing math.
Skeptics:
Economics of Orbital vs Terrestrial Data Centers
Do Orbital Data Centers Make Sense? - by Andrew Cote
The Truth About SpaceX’s “Orbital Datacenters”—YouTube
Notes on Space GPUs—by Dwarkesh Patel
Will we really put data centers in space?
Casey Handmer is more neutral here, though he thinks space AI would be 2x more expensive than terrestrial:
His Orbital inference Spreadsheet.
Direct Current Data Centers – Casey Handmer’s blog
Space AI: I guess we’re doing Moon factories now – Casey Handmer’s blog
SpaceX’s AI Data Centres Might Actually Be A Good Idea. Here’s Why—YouTubeOptimists:
Space Intelligence
Can We Build AI in Space?
Why Space-Based AI Data Centers Are Inevitable: 3 Levels of AnalysisSome additional links offered by ChatGPT that I did not read
I have downvoted your quick take. If you used an LLM to write, please use an LLM content block, which can be created by typing ”/” and selecting on the dropdown. Your writing has LLM smell to my eyes and to pangram. Writing a full post in italics is also unnecessarily distracting.
Though, I believe what you’re talking about is called tacit knowledge here. The Best Tacit Knowledge Videos on Every Subject seem to be a very good attempt at collecting resources for observation.
The comments are quite dismissive unfortunately
I don’t, but here are other people’s thoughts.
Midjourney Medical | Hacker News
https://www.astralcodexten.com/p/preliminary-thoughts-on-the-midjourney + his twitter thread