ggdupont

@crankylinuxuser @urbanfoxe @mattsheffield No one do fine tune on context window. It's refinement on the probabilistic distribution with more tokens. There is no training that is meaningful on such small scale. There no training that shows stability on self training. LLMs are just adjusting to the inputs.

@carnage4life@mas.to :facepalm: that's called entertainment... if someone want historically accurate, they watch a documentary

@nixCraft@mastodon.social lenovo laptop ranges is much wider than what was available from IBM (and includes some shitty machines) But they still have a very decent range of ThinkPad that are some of the most repairable of the market. Not as extensible as before (but it might be also a consequence of the deeper integration of components).

@david_chisnall@infosec.exchange they should even more piss-off against US, shouldn't they?

@carnage4life@mas.to The idea that US could have built any frontier models without international research is also silly. But it fits their narrative.

@ploum @cstross @26aafa19 @ravenbait @nathanael (nitpicking but we had our share of slaves in Europe - likely for longer than the US - and we did export that approach for society)

@flomaraninchi@pouet.chapril.org Avec le réchauffement, elle va peut-être survivre...

@brunus @davduf Il connait certainement. Il fait passer un message "ce n'est pas important, regardez plutôt de ce côté", car sa ligne éditoriale a d'autres objectifs.

@gabboman @peter I prefer my bug to be artisanaly sourced and hand crafted. These industrials AI bugs do not get the same level of attention and the quality is sloppy at best.

@tshirtman @carnage4life As a European currently in China where my guess-timate is that 60/70% or cars on the road are now electric, I find it both interesting how the fossil crisis does not really exist here and depressing since it could have been the same for us if we had taken energy independence seriously some 10 years ago.

@endrift @0xabad1dea It's some kind of special crystals that are infused with electricity. They emits lights and sounds. You can communicate with them with special languages that takes years to learn. But if you're good enough you use these to summon things. Well magic.

@matt_garber @codinghorror It does apply to OpenAI (if anything, the signs are even more obvious). It might extends to any major competitors hoarding resources for this. One fundamental is often trying to scale for economic objectives without ensuring to scale benefits for most.

@gee@framapiaf.org Je connaissais pas cette serie histoire greque. J'aime beaucoup toujours le style. Y'a un petit rappel de "il était une fois l'histoire" mais les blagounettes sont mieux. Je crois que je vais retenir l'Histoire grecque bien mieux que mes cours barbant de lycée ;-) (ok ça date un peu aussi)

@Troll@maly.io Utilisation simple sur mon tel Oui des plantages de temps en temps mais rien d'insurmontable en rapport au service rendu pour moi.

@paelnever I can see one... :guillotine:

@paelnever @tchambers You're probably right on mastodon being (still) at a small scale and that it will face challenges when it will grow. I tend to feel the scale by itself is not enough and that the network need to promote "virality" by design. Mastodon does not and the decentralisation will make it very difficult to get the same virality. I would even advocate to make a fundamental safeguard in fediverse. I'm also a bit more optimist than you on the nature of "People" in general ;-)

@peter "No but it's ok, just get GPT7.2 pro max plus golden to fix it. It now costs 20x the previous model but it's definitively 2% better on that weird benchmark they just released. Also the previous model is not available... but don't worry, GPT7.2 is sooo goood, it tastes like AGI" (If you feel sarcasm, you're on the right track)

@Crayon_Laser@framapiaf.org Data is the new oil

@lmorchard@masto.hackers.town Anybody saying anything like "it's just physical work, it doesn't deserve more salary" should be asked to "just" dig holes in dirt for "just" a week.

@queen_fennec@mas.to :facepalm:

@bearstech Vu ce matin. Ça sent quand même le scam non?

@quinn @adulau We know. The agency that are using it know. The decisions maker know. They made the choice anyway. To be clear: it was already a bad choice in the initial "temporary" contract years ago.

@ploum@mamot.fr What would be needed to fork mozilla?

@ethanschoonover breaking news: in a full-on bet on AI and LLM, Zuck decides to rebrand his company "LLM-verse" => that one will be a success... for SHURE

@davidgerard@circumstances.run :facepalm: "AI is amazing, look! the coding assistant replied to 5K PR just in the last hour!"

@SamCrawley@sciences.social @carnage4life@mas.to Well two options here: 1- he's lying so people get more committed to his product (that wouldn't be the first time) 2- it's true thanks to massive optimisation in inference and models size and then it means the models can run on your laptop now... What could it be ??

@bortzmeyer @tek I understood it the same teaching approach in Nigeria where a lot of AI ghost writers was/is done.

@miko@piaille.fr 2 fois la même image?

@codinghorror@infosec.exchange @glyph@mastodon.social Deeply shared feeling. I had this serious discussion with a colleague about code review: when I ask him to review my code, I want him to have a serious look... not his claude even if he "validated" claude's review. The goal is him to get to know (and eventually crush) my code.

@Mourioche @ache @bonpoteofficiel @FdC @fourmiune Oui la température ressentie est importante pour le suivi (la prévention) des effets sur les personnes. De même la température à l'ombre normalisée est très importante pour le suivi métérologique et notamment l'accéleration des canicules. Il faut surtout retenir que "LA" température absolue n'existe pas. Que les détails des variations de quelques degrés ne sont pas si importants. La tendance de fond l'est : canicule en Avril !

@carnage4life@mas.to (I haven't read the article yet) All these numbers related to "creating artifacts productivity". What about the "creating value" productivity?

@pbloem let's call it what it is: full on nazi imperialist manifesto

@davidchavalarias je découvre gargantext avec ce fil... Très intéressant et beaucoup de potentiel. Par contre, je ne sais pas par où commencer sur le GIT... le main parait vide et peu d'information pour lancer ça. J'ai raté un truc?

@Roland@framapiaf.org Non non c'est pas raccord. "France vérité" serait plus dans leur logique.

@bearstech AIRBUS en 2) je rigole un peu quand même... J'en viens soit à douter de la qualité du classement soit à désespérer sur ceux qui viennent derrière...

@david_chisnall@infosec.exchange @sue@glasgow.social Let's not use big words like "quality". It is very subjective. Let's say it's a different shade of shite to promote diversity in shite.

@gee@framapiaf.org c'est à cause de la chaleur ? Les streams se dilatent et la bande passante tient plus ?

@paelnever @tchambers Not perfect, but the decentralisation removes incentives to scale these too much. It's not economically profittable for platform owners and thuq not for tje perpetrators. They become low background noise and not the central feed.

@lesmillemondes@re.lire.im @ploum@mamot.fr @saraschneider@tooting.ch Il manque un lien dans votre profil (j'ai trouvé dans les messages précédents)

@ben@werd.social Business idea: compile your vibe coding prompts to highlight gaps and errors in the specs. Then ask you to fix it. Then submit the right specs to proper developpers.

@simon I had this discussion/reflexion with my colleagues. We miss understood some important details of software practices. One important part was that PR review is much more about sharing knowledge and building collective know-how than "make the code better". With LLM assistants as a new tool in knowledge workers belt, this pattern of reflecting on what is the work and where are the core values is shifting (for the better I think).

@Mourioche @ache @bonpoteofficiel @FdC @fourmiune Il y a un problème similaire avec la radioactivité qui a même carrément plusieurs unités de mesure selon que l'on s'interesse au phénomène physique pur (à la source), à la ropagation dans l'environement ou aux effets possibles sur l'humain. Mesurer, c'est compliqué ;-) Mais c'est aussi hyper intéressant car c'est le fondement de la science : l'observation précise.

@carnage4life@mas.to Why does it more and more look like this whole "age restriction" stupidity is just a way to uniquely associate internet user to their ID again just for better ads modelling... :facepalm:

@tut_tuuut @bearstech Non mais là, ce sera vraiment différent, ils vont monter un institut "France attaque" soutenue par un consortium Atos et Cap Gemini

@bortzmeyer

Investment in Freedom

🫠🤣 En vrai, j'adore. C'est un troll d'un niveau vraiment sérieux : faire de l'argent pour vendre de la 💩 a des 🤡.

Le troll ultime ce serait que les bénéfices soient reversés à associations open source...

@Troll@maly.io Je suis sur pipepipe et je trouve plutôt bien. C'est quoi le manque vs libretube et flow ?

@aerique@genart.social @davidgerard@circumstances.run @ploum@mamot.fr 1000%

@david_chisnall@infosec.exchange @sue@glasgow.social there are infinite shades of shite...

@sebsauvage j'aila chance de ne jamais avoir eu à boser sur Azure (quasiment le seul gros cloud sur lequel je n'ai jamais même testé). Mais en de moment Github devient de plus en plus instable c'est incroyable... on arrive à moins de 90% de uptime (alors que tout le monde est habitué à du 99.9% ou plus)