Replying to
@fj they should do a merry go round at Dreiländereck
Remote
Ntropic
@Ntropic@chaos.social
mastodon 4.6.9Vegan, Quantum Physics, Gym & Calisthenics, TTRPGs, Lasercutting, 3D Printing, FOSS, Self Hosting, Anti Fascist
31 Followers
102 Following
44 Posts
Joined December 06, 2024
@peter@thepit.social @jonny@neuromatch.social Spawning subagents is a skill, hence this can be said to simply be a skill issue
Replying to
@baldur On my first larger project in high school I once wrote 10.000 lines of code in one all-night session. It was one large function, I was very proud of it. It worked. It was the only time in my life that I wrote worse code than what Anthropic did here.
This company is evaluated at 2.5x what would be needed to end extreme poverty for a year.
Anthropic called Mythos a "quantum leap", like other quantum phenomena quantum leaps tend to be very small.
Replying to
@fabio ok, so the headline is wrong, the model was prompted to extrapolate the image 5 seconds into the future. It was also not trained on its output.
Replying to
@cmccullough@polymaths.social I don'T think people hate AI - it's just an easier target. They hate techno-fascism, they hate capitalist power concentration. They hate that they are made dependent on income from a craft, and that then the powerful monopolize that craft - while using their crafts output without compensation. The problem was never machine-learning, it was our economic and social system.
Replying to
@WastelandWandrr@freeradical.zone @TheBreadmonkey@beige.party The Dunning-Kruger effect is not what people think
Mythos and GPT5.6 wil be unblocked by the US government after their companies fine-tuned them to glaze Trump and right wing christian nationalism.
Replying to
@Possiblydrew Do it the old way, use a search engine to open an AI generated SEO optimized sloppage instead!
Replying to
@pojntfx@mastodon.social Can my government please start following the law or at least try to...
@thestrangelet@beige.party @cmccullough@polymaths.social Yes, I don't truly think ythat people don't hate the models. I just think that they are conflating things into a category error. In research we have been successfully using AI models for decades, to classify galaxies, supernovae, count cells in petridishes, and analyze data - if you don't hate those things my comment should be interpreted as an invitation to define a smaller subset within "AI" that one hates.
Replying to
@thelias @diritschka @SchreibeEinfach@23.social @annyhartmann@mastodon.social Ein Faktor der in einem statistischen Modell als Konstante faktorisiert kann keinen kausalen Effekt besitzen. Diese Konzepte sind mit einander verwoben.
Replying to
@thelias @diritschka @SchreibeEinfach@23.social @annyhartmann@mastodon.social 3. Da letztere einen Parameter weniger benötigt, ist sie besser laut Occam Faktoren. p(H) selbst besitzt weniger Aussagekraft, sie ist daher ein schlechteres Modell.
Replying to
@thelias @diritschka @SchreibeEinfach@23.social @annyhartmann@mastodon.social Frage hätte sich erübrigt hättest du es verstanden.
Replying to
@thelias @diritschka @SchreibeEinfach@23.social @annyhartmann@mastodon.social du hast ganz klar den Kommentar nicht verstanden. Ich habe nicht die höhere Frequenz in Frage gestellt, sondern durch Variablen ergänzt, welche den Effekt erklären und nichts mit Herkunft zu tun haben. In der Wissenschaft schauen wir nicht nur Raten an, sondern suchen nach den Ursachen und Herkunft ist eben nicht die ursächliche Variable, sondern Alter x Geschlecht und sozialer Status.
Replying to
@thelias @diritschka @SchreibeEinfach@23.social @annyhartmann@mastodon.social Experten sprechen da von konfundierenden Variablen - zB. Alter und Geschlecht (da fast alle Strataten von jungen Männern begangen werden und diese unter Syrern in D überrepresentiert sind) und Einkommen; deren Berücksichtigung führt sogar zu geringerer Kriminalität bei Ausländern, und das obwohl wichtige Aspekte, wie höhere Anzahl von Kontrollen und höhere Anzeigewahrscheinlichkeit da nich nicht einmal berücksichtigt ist.
Replying to
@petealexharris@mastodon.scot @capnthommo@c.im i use a local 30B model with large context window to code. It is so slow that using it is an active choice due to the multi second delay. If I am unfamiliar with what command i need for an operation, I add a comment line to ask it, it outputs it, and i continue coding. Removes the need to search online.
Replying to
@elphez@mas.to @russss@chaos.social @mewsleah@meow.social @gsuberland@chaos.social no jre is a major plus
Replying to
@elphez@mas.to @russss@chaos.social @mewsleah@meow.social @gsuberland@chaos.social better audio quality than spotify and has tidal_dl_ng,
@mojoaxel Klarnamenpflicht überall und für alle! Da Namen nicht immer ausreichen um richtig zu hetzen wird jeder Minderheit ein Symbol zugewiesen welches unter dem Namen getragen werden muss. Für Muslime der Mond, für Juden ein Davidsstern...
Replying to
@gabrielesvelto The comanies are currently operating at a loss, but that's not necessary. Due to the exponential scaling laws, distillation and quantization, we can run 100x smaller models at 100x lower price with almost the same performance. As investor money runs out, companies will simply have to run models at a size that covers the costs. Google will likely win that race, since they can continue operating at a loss for longer due to their remaining revenue streams and in house tensor chips
Replying to
@cmccullough how will me hosting my own models and running them on my computer make alibaba rich?
https://c.org/8bJ8Qn8yCY
Since the latest update #vw blocks #grapheneos and #api access to peoples car info. This forces people to use #google for tge very basic functions of pre-cooling ones car or sending routes/destination to it. The lack of API makes #smartcharging impossible, which is an important way to facilitate our personal contribution to #energytransition.
@proscience@toot.community Die versuchen in dem Game auch vorne mitzuspielen. Reichenrekord https://www.meinauto.de/news/2025/09/mercedes-eqs-faehrt-1-205-kilometer-ohne-ladestopp-reichweitenrekord-mit-neuer-batterie
Replying to
@randy_ @gaufff @signalapp @delta side loading. Or not using ios and android. See grapheneos for example (from which this message was sent)
Replying to
@TheBreadmonkey@beige.party In classical mechanics the fact that we don't understand it hides behind our intuition of what will happen. In quantum mechanics we can gain an intuition for what will happen, but we still have to face our lack of knowledge about an ontology that we simply refused to question in classical physics.
Replying to on infosec.exchange
@0xabad1dea@infosec.exchange federation is extremely useful though here.
Replying to
@CCC@social.bau-ha.us Just one more great opportunity to further stop the brain drain from the EU to the US.
Replying to
@briansullivan @SmudgeTheInsultCat I know how I code. I know how others code. All code is buggy amd I don't trusz the security of buggy code.
Replying to
@gabrielesvelto in the mid to long term RL post training will likely keep increasing, as this has shown the bigget benefits and its scalablility has been much improved. Model decay would come from pre-training data, and that is already available and only slightly increased between model versions - probably because they want to avoid model decay.
@thestrangelet@beige.party @cmccullough@polymaths.social [1/2] One common use case is as a search engine, and I would argue that what counts is not whether those have errors, but how one can reduce errors. In the case of LLM's these seem to mostly be hallucinations. I recommend the following paper: https://arxiv.org/abs/2512.01797
They found that certain neurons they call H-neurons light up whenever models hallucinate, indicating that they have an internal representation of when they hallucinate.
Replying to
Replying to
@Strandjunker I am afraid that that takes courage, and after having repeatedly exterminated the opposition over the last few Millenia we selected against that in our gene pool.
Replying to
@kaidu@mastodon.social @peter@thepit.social I prefer asking an expert in a closely related field, close eniugh for a decent judgement but far enough to not need funding for the thing they are commenting on.