saturnV's recent activity
-
Comment on The youth need your help in ~life
-
Comment on The youth need your help in ~life
saturnV Link ParentI think assigning that strong of a causation to this is entirely unjustified. It may be helpful, but to say necessary is IMO too bold a claim to make. Also, there is only so much that Biden is...I think assigning that strong of a causation to this is entirely unjustified. It may be helpful, but to say necessary is IMO too bold a claim to make. Also, there is only so much that Biden is going to do in an election year given he doesn't want to alienate moderate voters, especially as he is already now pressing the Israeli government harder than past administrations have. About 50% of US money given to israel is either for defensive weaponry or for humanitarian aid, which the US gov would be unwilling to drop, and Israel is perfectly capable of continuing the war without US support given how advanced their military industry is, and willing due to high internal support for it.
-
Comment on Telegram creator on Elon Musk, resisting FBI attacks, and getting mugged in California in ~tech
saturnV LinkFor a more balanced overview of Telegram I liked this article tl;dr: it's making “hundreds of millions of dollars” in revenues, with 900mn monthly active users, up from 500mn at the beginning of...For a more balanced overview of Telegram I liked this article
tl;dr: it's making “hundreds of millions of dollars” in revenues, with 900mn monthly active users, up from 500mn at the beginning of 2021, $30bn-plus valuations
Only 50 full-time employees, which is why so much dodgy content remains on it, alongside his "free speech advocate" attitude
According to Durov, he fled Russia a year later after refusing to share the data of certain Ukrainian users of VK with Russia’s security agency. Durov has said he sold his shares in VK to Kremlin-friendly oligarchs for $300mn under duress.
Critics have suggested that the Kremlin may have links to or leverage over Telegram, a claim that Durov dismissed as “inaccurate”. -
Comment on ChatGPT provides false information about people, and OpenAI can’t correct it in ~tech
saturnV (edited )Link ParentYes, in fact openAI have done something very similar to the first half of what you suggest (https://openai.com/research/language-models-can-explain-neurons-in-language-models). This whole field...Yes, in fact openAI have done something very similar to the first half of what you suggest (https://openai.com/research/language-models-can-explain-neurons-in-language-models). This whole field goes under the name of "mechanistic interpretability" if you're interested in further research. Currently the hardest part is that all labels are quite broad and vague, and it is hard to do precise "surgery" on models. Also, as models get larger, this approach gets more impractical, e.g. GPT-4 is rumoured to have 1.7 trillion parameters
-
Comment on What programming/technical projects have you been working on? in ~comp
saturnV (edited )Link ParentI did set 1 of cryptopals pretty quickly but then got distracted with other things and never got bothered enough to resume it. If you're looking for a comprehensive source of n-gram frequencies...I did set 1 of cryptopals pretty quickly but then got distracted with other things and never got bothered enough to resume it. If you're looking for a comprehensive source of n-gram frequencies without having to parse all the data I found peter norvig's blogpost helpful. Seeing how/if frequencies change over time would be interesting, because it's well known that different languages have different unigram frequencies, so you'd expect some sort of change over time as well. I think the issue with google books is that the data for older books are pretty low-quality, lots of OCR mistakes and low sample size, but better than nothing
edit: If you want a good source to start learning about how to approach the ciphers, practical cryptography is a good shout
-
Comment on What programming/technical projects have you been working on? in ~comp
saturnV Link ParentSounds interesting, I did something similar when I was at school with the cipher challenge which is mostly archaic ciphers, but I've been recommended cryptopals as something to learn "real" encryptionSounds interesting, I did something similar when I was at school with the cipher challenge which is mostly archaic ciphers, but I've been recommended cryptopals as something to learn "real" encryption
-
Comment on What programming/technical projects have you been working on? in ~comp
saturnV Link ParentAutohotkey is the tool that most people use on windows for this sort of stuff, on macos I know there is Keyboard Maestro and on Linux Espanso, but I don't know if they are as powerful, sounds...Autohotkey is the tool that most people use on windows for this sort of stuff, on macos I know there is Keyboard Maestro and on Linux Espanso, but I don't know if they are as powerful, sounds pretty crazy what you're doing with it! Personally I would generally delegate most of the logic to python (because that's what I'm used to) as soon as it gets moderately complex, helps keep it fairly tool-agnostic. For scraping websites sometimes beautifulsoup or selenium can be less hacky. Obviously the tool you have is better than the one you don't though ;)
-
Comment on UK asylum seekers will be deported to Rwanda in ~news
saturnV Link ParentWell more specifically it was unlawful until they just passed a law effectively overruling the court and forcing them to treat it as a safe country (ridiculous but that's what you get with...Well more specifically it was unlawful until they just passed a law effectively overruling the court and forcing them to treat it as a safe country (ridiculous but that's what you get with parliamentary sovereignty). The biggest potential challenge left is now from the ECHR.
-
Comment on Looking for free or cheap places to learn some SQL and XML in ~tech
saturnV Linkhttps://sqlbolt.com/ is a nice free way of learning simple SQLhttps://sqlbolt.com/ is a nice free way of learning simple SQL
-
Comment on New Brexit checks will cause food shortages in UK, importers warn in ~food
saturnV LinkFeels a bit weird that this is in ~food, even if it is technically correct. Should this be moved to ~news or ~misc, or is tagging it with politics mean that it's fine here? -
Comment on Where will people commune in a godless America? in ~humanities
saturnV Link ParentI think it is overly dismissive of theologians and critical scholars of the Bible to say that religious people are unwilling to examine their beliefs in depth. I do understand that the academic...I think it is overly dismissive of theologians and critical scholars of the Bible to say that religious people are unwilling to examine their beliefs in depth. I do understand that the academic version of religion is quite different to how it is actually practiced, but still think it's worth mentioning.
Also, I do think it is possible to be a Christian without believing in much of the Bible being literal, one can believe that Jesus' miracles were just teachings like his parables, and that religion is just a way of transmitting morality and culture through non-literal stories. -
Comment on Where will people commune in a godless America? in ~humanities
saturnV Link ParentI'd be interested to know how you justify belief in scripture and a specific religion over another from a rational point of view. I have found arguments for an Aristotelian prime mover or even a...I'd be interested to know how you justify belief in scripture and a specific religion over another from a rational point of view. I have found arguments for an Aristotelian prime mover or even a pandeist kind of thing plausible, but never heard convincing reasons why one specific religion with specific doctrines determined by humans should be followed, short of religious experience literally telling you what to follow. For instance, looking at Christianity, one can see an evolution of beliefs throughout the bible, which one would not expect if God was guiding the writing of scripture.
-
Comment on New Brexit checks will cause food shortages in UK, importers warn in ~food
saturnV (edited )Link ParentLabour have said that they are going to attempt to build closer relations with the EU, basically creating a roadmap to eventually rejoin the EU. Thing is, the EU are not going to let the UK back...Labour have said that they are going to attempt to build closer relations with the EU, basically creating a roadmap to eventually rejoin the EU. Thing is, the EU are not going to let the UK back in on the same terms, because the UK is currently a lot less powerful than it was 5 decades ago, and also because they want to discourage right-wing populists from making it look like a risk-free move for them to do their own exit, and making those kinds of concessions would be immensely politically damaging for UK politicians to make so soon after Brexit.
-
Blind internet users struggle with error-prone AI aids
7 votes -
Comment on Favorite hobby / subculture YouTube channels? in ~hobbies
saturnV Link Parentjust so everybody knows, bald and bankrupt is/was a sex tourist and has credible accusations of lots of dodgy stuff against himjust so everybody knows, bald and bankrupt is/was a sex tourist and has credible accusations of lots of dodgy stuff against him
-
Comment on Favorite hobby / subculture YouTube channels? in ~hobbies
saturnV Linkjan Misali does mostly linguistics, but also some misc nerd stuff like rythym games or base-12 counting ACollierAstro does good videos about astrophysics, but can be broader Dan McClellan does...jan Misali does mostly linguistics, but also some misc nerd stuff like rythym games or base-12 counting
ACollierAstro does good videos about astrophysics, but can be broader
Dan McClellan does interesting stuff about Academic biblical study, fun to watch him debunk biblical literalists and those who claim historicity where it doesn't exist (also does data over dogma podcast)
Alex O'connor does philosophy videos, often related to atheism but not always (also has a podcast)
Another Roof,numberphile,mathologer all do good in-depth maths content
Jenny Draper does interesting videos about british (mostly london) history -
Rulers and Power | Mary Beard and David Mitchell
5 votes -
Comment on Research paper compares LLM responses based on politeness of requests and finds quality difference in ~tech
saturnV LinkIMO, this is the most interesting part from the paper, not the rest which has been talked about quite a lot already in LLM spacesmodels trained in a specific
language are susceptible to the politeness of that
language. This phenomenon suggests that cultural
background should be considered during the devel-
opment and corpus collection of LLMs.IMO, this is the most interesting part from the paper, not the rest which has been talked about quite a lot already in LLM spaces
-
Comment on Research paper compares LLM responses based on politeness of requests and finds quality difference in ~tech
saturnV Link ParentSo, yeah, roughly correct for GPT-4, though the paper described quite high variance between models, so can't necessarily extrapolate very far also from the conclusion:GPT-4’s scores are variable but relatively stable. The highest score is achieved at level 4, and the lowest one is at level 3. Although the score at level 1 is not extremely low, the heatmap indicates that it is significantly lower than those at more polite levels.
So, yeah, roughly correct for GPT-4, though the paper described quite high variance between models, so can't necessarily extrapolate very far
also from the conclusion:
However, highly respectful prompts do not always lead to better results. In most conditions, moderate politeness is better, but the standard of moderation varies by languages and LLMs
-
Comment on Research paper compares LLM responses based on politeness of requests and finds quality difference in ~tech
saturnV LinkThe models’ ROUGE-L and BERTScore scores
consistently maintain stability, irrespective of the
politeness level of the prompts, which infers that
the models can correctly summarize the article con-
tent in the summarization tasks. However, the
models manifest substantial variation in length cor-
related to the politeness level. A progressive reduc-
tion in the generation length is evident as the po-
liteness level descends from high to lower scales.
Conversely, a surge is noted in the length of the
outputs of GPT-3.5 and Llama2-70B under the ex-
ceedingly impolite prompts.
even historically: women's suffrage, end of prohibition, lowering the voting age to 18, passing of ADA, establishment of EPA,FDA,OSHA. These definitely all created genuine long-lasting change, and I would argue did so in a way that align with progressive values. The deaths that caused these to happen were not of a shocking kind but of a mundane one, so aren't really similar to BLM/Vietnam war/AIDS deaths