jampekka

10 ms·

91.
by jampekka 4mo ago
Saying condoms are an easy fix to STDs is a bit like saying eating less is an easy fix for obesity though.
92.
by jampekka 4mo ago
We don't really know that about human written text either.
93.
by jampekka 4mo ago
Indeed. Well written, clear, informative and to the point.
94.
by jampekka 4mo ago
My bet is on obesity. "Overweight and obesity significantly increase colorectal cancer risk: a meta-analysis of 66 studies revealing a 25–57% elevation in risk" https://pmc.ncbi.nlm.nih.gov/articles/PMC1218149
95.
by jampekka 4mo ago
Diet and exercise (to lesser extent) are the mechanism of obesity. The other factors may affect diet and exercise. A massive other factor for the latter is driving.
96.
by jampekka 4mo ago
Obesity indeed is a massive elephant in the room in public health discussions. And even in TFA "ultra-processed foods" are put first, which is a) just a silly category, and b) effects from poor quality nutrition are mainly via obe
97.
by jampekka 4mo ago
(Three out of) four experiments is anecdotal for sure, but the result meshes with more established instruction following benchmarking (although DeepSeek V4 pro does not top these): https://artificialanalysis.ai/evaluations&#
98.
by jampekka 4mo ago
> Contrary to belief, renewables, or generally speaking DC, makes things this stability problem worse. Is there such belief? My feel is that anybody to whom electrical grid stability has even crossed their minds know this.
99.
by jampekka 4mo ago
Also it was an asthma prevention study, not cognitive functioning one, adding even more "researcher degrees of freedom". Doing such side studies is fine in itself, but selling such shakey results as "This study suggests that
100.
by jampekka 4mo ago
> It's quite strange that it's very easy to detect AI in writing. Or you detect only the easy to detect AI writing?
101.
by jampekka 4mo ago
The idea does smell a bit like a rationalization for policy that was extremely convenient for stockholders and a disaster for workers.
102.
by jampekka 4mo ago
They do. I asked CharGPT for 327 x 48 and it used the "ChatGPT Instruments" calculator. Previously it used to run Python scripts, and may still do for more complex calculations.
103.
by jampekka 4mo ago
The US gun culture resembles nothing like "a militia officered by men chosen from among themselves, fighting for their common liberties, and united and conducted by governments possessing their affections and confidence." This was
104.
by jampekka 4mo ago
Why is this linked through the fishy looking link shortener?
105.
by jampekka 4mo ago
> Now I work mostly with PhDs who were at the top of every academic environment they've ever been in. And yet I can see their thinking skills rapidly declining as well; many of them can no longer brainstorm, code, think deeply, or w
106.
by jampekka 5mo ago
The benchmarks are a bit of a disaster? It's at about DeepSeek V3.2 level, but with about 50% more parameters. Loses handily to the also smaller GLM-5.1, and even worse to the similarly sized Kimi K2.6.
107.
by jampekka 5mo ago
I have a ThinkPad P1 (Gen 2) and I run Linux, but I'd hesitate to recommend at least the Nvidia dGPU versions. The battery life is dismal, even with the dGPU disabled in software (can't be HW disabled). It also runs really hot and
108.
by jampekka 5mo ago
API param passthrough will probably help with many of the cases. Things like sampling params and constrained decoding and returning logits tend to be very finicky with the translated params. But the return value translation also makes debug
109.
by jampekka 5mo ago
Would it be possible to get "raw" access to the provider APIs, but still keep the consolidated billing? The unified API is great when it works, but it often causes hassle with more exotic use cases and new API features.
110.
by jampekka 5mo ago
The main friction reduction, for me at least, is the consolidated billing that avoids extra bureaucracy in corporate environments. The API-translation/abstraction tends to cause more problems than it solves. I’d prefer something that c
111.
by jampekka 5mo ago
Game sales have skyrocketed though. Inflation corrected revenue has increased something like 5x in the past 30 years. https://www.visualcapitalist.com/video-game-industry-revenue...
112.
by jampekka 5mo ago
That's the usual null hypothesis for these kinds of tests.
113.
by jampekka 5mo ago
The methods could be better described in the paper, but my understanding is that they did 10 runs for each question for each prompt and took an average of those, so the compared values are not binary. You could do a sign test, but you'
114.
by jampekka 5mo ago
The instability started well before vibecoding, in around 2018-2019, shortly after the Microsoft acquisition. https://damrnelson.github.io/github-historical-uptime/ https://news.ycombinator.com/item?id=
115.
by jampekka 5mo ago
Arch Wiki's core content is instructions of what commands to run right in the terminal.
116.
by jampekka 5mo ago
Models can answer "I don't know". Hallucination benchmarks, including this, give the models the option to "not attempt". It's just that the metric linked doesn't take into account the rate of correct answe
117.
by jampekka 5mo ago
This counts only incorrect answers though. A model can get 0% hallucination rate just by refusing to answer all questions.
118.
by jampekka 5mo ago
The models still hallucinate bad when called via APIs, especially if web search is not enabled. Gemini hallucinates quite frequently even with the app and search enabled. More recent (e.g. ChatGPT 5.x and Deepseek v4) prompts/harnesses
119.
by jampekka 5mo ago
Not all authoritarian regimes are fascist. Pinochet and Videla regimes were more military juntas and bureaucratic-authoritarian states.
120.
by jampekka 5mo ago
> Sure it’s not as dry, but I’ve been bitten in this regard because css framework and templates are so intransparent, preventing me from simply changing padding or margin. How does this happen? You can always override css values. Either

More ›