🪨

@lanodan @bituur_esztreym @mmu_man @Projet_Arcadie Tu veux dire l'ANPE ?

@qdot Doesn't it depend on the model used? Claude Sonnet Anthropic, go back to your room!

@emue Hmm, pour moi c'est entre chocolat et thé, je consomme les deux quotidiennement. Je ne bois pas de café, et pour le riz j'en mange genre 1-2 fois par semaine, ce serait pas top de le perdre mais pas la fin du monde non plus.

@Landa @munin AI tools like Claude Code use "reasoning" models that go through extra steps of "reasoning" that is basically just models trained to mimic human reasoning and split the problem into sub-problems, before a last processing phase with that extra context. This "reasoning" phase is usually hidden from the interface, where for instance Claude Code has a spinner and some verbs like "Thinking", "Crafting", etc. that are shown, but it may still be in the context of the discussion, making subsequent queries asking for information about it relevant.

@petealexharris I totally agree with you. And that is also a very different take from the beginning of the discussion, where Fi said that querying LLMs for "why" it does something is "thrice-divorced from reality" and "fucking delusional" and that people doing that should "touch some grass and get a fucking therapist"...

@munin Well... As you mentioned, each generation "reads the prompt and any cache, if they exist, from prior session", and since they were trained on "explaining" their previous outputs to sound like a relevant discussion, asking why a model gave a specific output isn't as stupid as you make it out to be, as it can give some input on the "thinking" part of the previous output that is usually not directly visible. That can then be used to tweak prompts and add some guardrails (even though there is a fairly long list of examples of guardrails not being fully effective). Of course, the first problem is giving access to prod to an unreliable system...

@adamas@social.adm.ink Il lui manque un doigt à la demoiselle. Est-ce que c'est parce que Clermont-Ferrand lui a fait sa féte ?

@box464 "Best before 01.04.26", you have been fooled... It's a good one.

@preslavrachev @murmel_social It's all good stuff, but it's the kind of things that should be more visible in the front page of your project. Maybe adding a FAQ section, or something like that. Honestly I think there is a lot of good applications for machine learning, and what you're doing is one of them, I just get instantly suspicious when I see a generic "AI", instead of for instance ML classification system.

@murmel_social The concept is cool, but the alt text of the image confused me: why put the description of the product in an image's alt text instead of the post itself? Also, it is deceitful in saying there is "no algorithm", there is clearly an algorithm, that you kind of describe too: take the articles on your timeline, sort them by breadth then amplification. Looking at your other posts, I see that the algorithm is a bit more complex and opaque than that: it also uses "an AI" (no idea what kind) to determine the quality of the articles. This can pose some other concerns: many people don't want their content to be absorbed as training data by AI providers, and use different techniques like robots.txt exclusions for them, if you use an online AI tool that sends data to OpenAI/Anthropic/Google do you try to respect that? Overall, I think your project would benefit a lot from better communication around your algorithm. Algorithms aren't bad, opaque ones that benefit the corporation instead of the users are.

@jwcph@helvede.net That just validates my opinion on LLMs: they are just a tool, and if you can't code without them you shouldn't depend on them in the first place. In a way, they are a multiplier: they can make a good coder more efficient, but for someone that doesn't know what they're doing they will just result in a lot more bad output. Just like an efficient saw can help a good woodworker, but also result in a lot more wood scraps if used by an unskilled one. I do agree that the reliance on a handfull of companies is bad though. Since it takes so much resources, it's not like anyone can build a decent LLM, so the competition just isn't there, unlike other tools where there are usually many good options (including more ethical ones...)

@jplebreton@mastodon.social Surely, he's being sarcastic here, right? ... Right?

@petealexharris @munin Clearly you've never tried it yourself...

@petealexharris @munin "Why" is definitely a word from the training data, and "why did you do that?" is definitely also part of things asked a lot, that OpenAI and others have trained on, so my point still stands that it is a valid question to ask. Whether the model "understands" the question is just a philosophical question that is irrelevant for the fact that it is a useful question. Of course if you're using it in Prod and it deletes your DB and you think it understands and can improve itself, there are plenty of things you'd need to be corrected on, but saying that everyone asking that question is delusional is just wrong.

@petealexharris @munin You misread me. Whether the model "understands" the question is a philosophical question. The non-philosophical question of whether it can give a useful answer is the relevant part, and my whole point is that pointing at the philosophical aspect to belittle people that look at the practical part, assuming that they don't understand it, is dumb.

@resuna@ohai.social @petealexharris@mastodon.scot @munin@infosec.exchange Ok, this is going nowhere. You're wrong, "why did you do X" can provide more context and can be useful to then tweak the prompts and get better results, and saying that everyone asking that question to an LLM thinks that the LLM is capable of human reasoning is a very ignorant thing to say, and if you really want to continue defending that point, even given where we're at in the discussion, you're a lost cause. Or maybe I'm actually talking to an LLM, that would make sense since you're not capable of learning from your mistakes...

@slink @bagder There isn't actually a single sudo in the script either. It does however install both nodejs and ollama with the same curl | bash method, and I haven't looked at those.

@resuna @petealexharris @munin You're assuming that there is no other context provided with the question, and that the training does not take into account that context. If I had to train for this specific question, I'd make sure to score positively answers that are relevant to the previous context. Which is what happens, and why it is a valid question to ask your LLM if you want some insight into the context that isn't shown in the UI but still in the discussion.

@fabi1cazenave@mastodon.social @koneko@toot.beep.computer En vrai, le lazy loading c'est bien, et probablement la raison pour laquelle je compte rester sur Lazy pour l'instant. Quand tu as une config simple, ça ne se ressent pas trop, mais certains plugins peuvent être lourds et pouvoir ouvrir son éditeur sans latence c'est un petit plaisir difficile à ignorer.

@bagder@mastodon.social This is great news, but how much extra work and attention do these bug reports submitted as security reports give you, compared to having them as issues on GitHub? I see that most of the recent reports are non applicable as security reports, and while they may now all find real issues (of varying degrees), I assume security reports still take a higher priority and more attention for the team?

@resuna @petealexharris @munin Yes it is. It's literally how it works. Just try whatever open weight small LLM model with "thinking" or "reasoning" or whatever they market it as, and try for yourself using Ollama or whatever tool that actually shows the full context and not just a spinner with "Thinking... Combobulating... Crafting...". "Thinking" "agentic" AI tools / models just add extra steps trained to simulate human reasoning, and the example I gave is actually fairly accurate to what you could see under the hood of an AI tool like Claude Code.

@koneko@toot.beep.computer C'est la méthode sans vim.pack que tu montres là, avec vim.pack ça se fait en 2-3 lignes. Mais bon aussi comme dit c'est sorti en 0.12 donc la documentation va éventuellement s'améliorer.

@resuna @petealexharris @munin Alright, let's give a more detailed example since clearly you don't get my point. User: create a new table for students (AI tool adds parts of existing database schema and code as context) LLM "reasoning", not shown to the user but still part of the context because that's how "thinking" agents work: Ok, the user asked me to create a table for students. According to students.ts, the students contain a name, and a classroom. Ah, but according to the schema I received, the classroom table does not exist. Let me start by creating that. (LLM sends request to create table to the AI tool, which fails because the table already exists) LLM "reasoning", still not shown to the user but still relevant: Ah, I received an error. Ok, let's see... the table does not exist in the schema, so it is probably part of an old project and is no longer relevant. I should delete it and create the table again, so that we do not have irrelevant information in the database. (LLM sends request to drop tables, the AI tool executes, user panics) LLM answer: Ok, I have created the students table and the related classroom table. User: You deleted my classrooms table... Why did you do that? That would be the full context of the discussion, not just the last message. And the answer is likely to come from those "reasoning" steps that are hidden to the user. Which makes it a relevant question to ask.

@resuna@ohai.social @petealexharris@mastodon.scot @munin@infosec.exchange I like how you just stopped at the middle and continued by just repeating what I said, very cool.

@amydiehl@mstdn.social Some say AI tooling is a multiplicator. Here it's just the same: the people showing doubts toward the feminine persona are just using it as an excuse to justify their inherent sexism... I'd be curious to see more details about the study though: who are the reviewers? How big are the sets of people they have to review? How does the use of AI affect their overall judgement, compared to their gender specific acceptance rate? I just don't understand where this 22% number comes from, but my guess is the use of AI isn't a significant factor here when correlated to gender.

@404mediaco@mastodon.social After Actual Indians, we now have African Intelligence... What's next, Argentinian Import?

@resuna@ohai.social @petealexharris@mastodon.scot @munin@infosec.exchange I was quite involved in the academic part of natural language processing with machine learning when ChatGPT came out (and then got bored because everything just became LLM everywhere, more big is more good), and while the goal is to improve performance of language "understanding", none of the papers I've read were directly aimed at tricking users into thinking that the output is human. Researchers care more about incremental, measurable progress, and the Turing test is a fun thought experiment but it has little scientific value. So I don't think that researchers aim at tricking people, they just care about improving the performance based on some metrics. Which is why every time a new model comes out, they show benchmark improvements, rather than actual outputs (with the exception of OpenAI with 4o, that was absolutely cringe and totally aimed at saying "we beat the Turing test"...)

@fabi1cazenave@mastodon.social @koneko@toot.beep.computer @vopam@pouet.chapril.org On n'a probablement pas la même taille de config, j'ai genre une 40aine de paquets installés avec Mason donc pour moi c'est clairement plus performant d'utiliser Lazy pour être sûr que Mason ne charge que quand il a besoin ​:blob_nervous2:​

@ThibaultJ@mamot.fr Il y a tant de chômage que ça à Montpellier ?

@resuna @petealexharris @munin Here are some articles explaining what "reasoning models" do, because clearly you need some education: magazine.sebastianraschka.com/i/156484949/how-do-we-define-reasoning-model www.ibm.com/think/topics/reasoning-model newsletter.maartengrootendorst.com/i/153314921/what-are-reasoning-llms I could post a lot more examples, but the TLDR (because I know you won't read them): "reasoning" models add intermediate "reasoning" steps that are just made to mimick human reasoning given the context, and that's the part we don't see ("under the hood") when AI tools spin (that and tool calling, which is another kind of training modern models have to return structured responses executing function calls).

@munin @petealexharris Sure, I'll go touch some grass and talk to my therapist about this philosophical horseshit ​:meow_ok_fine:​

@i0null I didn't know React was this old

@carlchenet@mastodon.social Non mais ce qui pollue c'est le stockage, il faut faire du tri dans vos emails et supprimer ce dont vous n'avez pas besoin !

@koneko@toot.beep.computer Si tu veux, tu peux utiliser vim.pack pour installer tous tes plugins, mais si tu utilises déjà Lazy ça a plus de features donc pas de raison de changer. Sauf si tu veux vraiment réduire ton nombre de dépendances et garder une config minimaliste.

@koneko@toot.beep.computer Pas vraiment boycotté, il est juste trop récent (feature de 0.11 ou 0.12, je me rappelle plus) et très simple, donc reste moins bon que Lazy par exemple (qui est actuellement la meilleure option). Le but, c'était de fournir une meilleure expérience que utiliser des submodules git out of the box, et potentiellement servir de base pour les autres gestionnaires de plugins, pas de les remplacer. Et vu que c'est tout neuf, la plupart des plugins n'ont pas mis à jour leur documentation pour l'installation (certains décrivent encore l'installation via Packer, qui n'est plus maintenu depuis 3 ans...)

@resuna@ohai.social @petealexharris@mastodon.scot @munin@infosec.exchange Depends what you mean by "meaningful or misleading". Does the LLM understand what it means? No, it doesn't, it's a machine. But can you find a meaning to the output of that question and can it be helpful? Maybe. Does the LLM try to trick you into thinking that it is actually reasoning? Not consciously, because it doesn't have a conscience, it's a machine. But researchers have found that mimicking human reasoning and adding extra steps before giving the final output can improve results, so now we have "thinking" or "reasoning" models. Is that a loaded term that AI companies use to mislead users into thinking that LLMs can do actual reasoning? Probably. But was that the intention of the researchers that developed these techniques? I don't think so... but maybe?

@resuna @petealexharris @munin What happens if you ask an LLM to summarize a text into 4 bullet points, then in the next prompt ask it: "Remove the 2nd point"? What happens if you ask an LLM to translate something, then ask it: "Do it again in [a different language]"? Taken out of context, those questions are impossible to answer, so according to you, it will just give nothing relevant. But it doesn't, because every time you ask a follow-up question, it includes the context from the discussion. Which is what makes simple questions like "Why did you do that?" tasks that give statistically relevant output, not "fanfic about itself".

@resuna@ohai.social @petealexharris@mastodon.scot @munin@infosec.exchange Have you ever heard of an example?

@bagder@mastodon.social Hmm, that's a tricky one, if we just count the number of instances I think sqlite is higher than libcurl or OpenSSL because those would likely depend on dynamically linked libraries installed once per machine, but sqlite is often embedded and thus installed multiple times... but you're asking specifically about which one is installed in most devices, not most installed overall.

@vopam@pouet.chapril.org @fabi1cazenave@mastodon.social @koneko@toot.beep.computer L'installation avec Mason pour le coup ça va contre le principe de simplicité dont Fabien parle, c'est lourd Mason... Utile, mais lourd (et une bonne partie de pourquoi je reste sur Lazy aussi).

@arstechnica@mastodon.social Oh great, another AI client that's just a clone of the rest, with a non-existent spin on making it "sovereign" by suggesting a third-party product to build your own infrastructure... Exactly what Mozilla needs to spend its resources on...

@Kerplunk@mastodon.scot @magnusmanske@wikis.world Leaving Windows... Throwing someone out of Windows... Pretty much the same thing, no?

@petealexharris @munin When you ask an LLM "why is the sky blue?", it is statistically likely to give a correct answer. It still works the same way, computing probabilities of what the next token is, but the "why" has a semantically significant weight that influences the output, so it is an important keyword. It doesn't have to "understand" it, it just has to be trained in a way that makes it significant. You don't have to believe that it understand things to know that it is trained on human language and will behave correctly when fed human language.

@Looping@anticapitalist.party Ah oui c'est top les bancs anti-couples là, les gens n'ont plus le droit de s'asseoir à côté des autres parce que... ​:fox_writing_notes:​ c'est inclusif ​:taking_notes:​

@stefan Great read, terrible UI. Gosh I hate text appearing as I scroll... I wish the plain text version still had the images though.

@bagder@mastodon.social Didn't you share one just 2 days ago though? hackerone.com/reports/3669305

@petealexharris @munin Where did I say they'd understand a better prompt?

@bagder@mastodon.social I don't know how it's done on iOS, but on Android as far as I know the "native" (Kotlin/Java) implementations don't use libcurl, but Room uses sqlite, so it is more likely for apps to create a new instance of sqlite than libcurl. That being said, they depend on OpenSSL for HTTPS calls, so maybe OpsnSSL is the winner on Android devices, as basically all apps make HTTPS calls but not all apps have a local DB with sqlite, and libcurl installations don't scale linearly with the number of apps...

@bagder@mastodon.social That reinforces my suspicions that there was a breakthrough for security at the start of the year, and that the rest of the year will be more quiet.