Anders Cornelius Madsen

@anderslund@expressional.social @benjaminlj@virup.social @kkdk@mastodon.social @bettina@mastodon.nu @jeppe@uddannelse.social @simonjust@mstdn.dk @sindum@mstdn.dk AI er et værktøj, ikke andet. Folk kan bruge det forkert og ende med at miste viden og læring, men det er deres fejl - ikke teknologiens. AI kan drives på datacentre hvor det er en miljømæssig katastrofe, men man kan også drive sin egen lokale AI-agent på hardware, der tager mindre strøm end et køleskab (det gør jeg selv). Igen, det er ikke teknologien, det er menneskers beslutninger, der er problemet. Jeg køber på ingen måde din præmis om at vi er pøbelen, der er sat udenfor for indflydelse og ejerskab. Hvis vi føler, at der er lovgivning, der ikke er fulgt med udviklingen, må vi vælge politikere, der vil opdatere lovgivningen - men det er igen ikke et spørgsmål om teknologi. Jeg køber heller ikke argumentet om at it-udstyr er utilgængeligt. Ja, meget er blevet dyrere, men den er set før. Det skete midten og slutningen af 1990'erne efter et par brande, der ødelagde store dele af forsyningskæden til RAM-produktion, så vi har prøvet en ubalance mellem udbud og efterspørgsel før. Du kan som forbruger eller virksomhed stadig købe en kapabel computer for under 5.000 kr., og får 10.000 kr. kan du drive din egen LLM. I en verden hvor folk betaler 60-70 kr. for en 20-styks pakke cigaretter om dagen, kan det altså ikke ryste mig.

@georgetakei No surprise there. He also warned the EU to stay out of Hungarian politics while standing on stage IN Hungary, supporting Orban. The guy is almost as dumb as his boss.

@pmakholm@social.data.coop @anderslund@expressional.social @benjaminlj@virup.social @kkdk@mastodon.social @bettina@mastodon.nu @jeppe@uddannelse.social @simonjust@mstdn.dk @sindum@mstdn.dk Jojo - og dynamit er jo heller ikke et værktøj, medmindre man arbejder med minedrift. Mit argument er stadig, at det er os selv, der er ansvarlige for, hvordan vi bruger tingene. Vi er voksne mennesker med en egen fri vilje, så vi er ilde stedt, hvis vi altid skal have andre til at tage beslutninger for os. Du kan ikke - og det er med en fed streg under IKKE - lovgive dig ud af den slags. Det er prøvet, og det er fejlet SÅ mange gange. Hvis noget er brugbart (og det er AI altså i nogen sammenhænge, uanset hvor meget modstanderne så ellers hyler og skriger) vil folk finde veje at omgå lovgivningen på. Den eneste vej frem er efter min bedste overbevisning at uddanne folk, så de tager fornuftige beslutninger, der ikke er til skade for dem selv eller andre i den størst mulige grad.

@benjaminlj@virup.social @anderslund@expressional.social @kkdk@mastodon.social @bettina@mastodon.nu @jeppe@uddannelse.social @simonjust@mstdn.dk @sindum@mstdn.dk Jeg tror at man på sigt må være forberedt på, at LLM vil være en del af stort set alle softwareprojekter, for nu at være helt ærlig. Personligt har jeg ikke noget problem med LLM i forbindelse med softwareudvikling - det at genbruge kode har været helt naturligt helt siden Bjarne Stroustrup udgav "The C++ Programming Language" med kodeeksempler tilbage i 1985, så om man får en LLM til at lave pukkelarbejdet eller man selv sidder og finder kodestumper på GitHub. er for mig ét fedt. Der er så mange andre ting, jeg ikke bryder mig om at man bruger en LLM til, f.eks. at lave musik eller billeder, som dybest set bare er en kopi af andres kreative proces - det stinker. Men at en LLM laver kode baseret på kode, der er lagt ud som open source fra starten, kan ikke hidse mig op.

@benjaminlj@virup.social @anderslund@expressional.social @kkdk@mastodon.social @bettina@mastodon.nu @jeppe@uddannelse.social @simonjust@mstdn.dk I forhold til at finde et alternativ til Zulip er der lige kommet en (brugbar) beta af Bonfire til communities - hvis Fedibook føles for usikker, kunne Bonfire måske være et godt bud på en løsning baseret på AP-protokollen? https://bonfirenetworks.org/posts/bonfire-for-communities/

@anderslund@expressional.social @benjaminlj@virup.social @kkdk@mastodon.social @bettina@mastodon.nu @jeppe@uddannelse.social @simonjust@mstdn.dk @sindum@mstdn.dk Hvis alle, der er store virksomheder eller organisationer per definition er en fjende som man underkaster sig, hvis man er bruger af noget, de har haft fingrene i, er der ikke rigtig nogen grund til at fortsætte en diskussion - så er vores syn på verden så fundamentalt forskellig, at vi aldrig kommer i nærheden af at møde hinanden. I mit verdensbillede er dæmonisering en direkte vej til et Orwellsk samfund, hvor alle, der tænker anderledes, er en fjende, der skal bekæmpes - og "anderledes" vil altid bestemmes af den til enhver tid herskende gruppe. Jeg er en stålsat modstander af absolut kapitalisme, men en lige så stålsat modstander af absolut kommunisme. Ingen af dem virker i praksis, og i en verden, der skal være til at leve i, skal vi være i stand til at snakke sammen og finde et kompromis et sted omkring midten af de to yderpunkter. Det kan vi ikke, hvis "de andre" dæmoniseres.

@benjaminlj@virup.social @anderslund@expressional.social @kkdk@mastodon.social @bettina@mastodon.nu @jeppe@uddannelse.social @simonjust@mstdn.dk Jeg har fulgt den tråd her lidt fra sidelinjen, og i forhold til at få grupper ind i protokollen eller I applikationerne, kunne man måske kigge lidt på, hvordan det er lavet på fedibook.dk? Projektet er nok desværre enten dødt eller I dyb dvale lige nu, men måske man kunne foreslå at få den grundlæggende gruppe-arkitektur trukket over i f.eks. Mastodon? Alternativt skal nogen med viden om programmering i Go tilbyde en hjælpende hånd med Fedibook og se, om det kunne komme op i omdrejninger igen.

@EUauditors@mastodon.social Holy crap! I expected perhaps 15 pages of high-level buzzword talk, but this is obviously a completely different deck of cards. I just read the list of contents, and I like the fact that this seems to be very aware of(and hopefully addressing) some of the things that opponents of AI are (in my view, often with good reason) criticizing about the ethical and political aspects of AI and it's impact on our society. Bookmarked and shared - this should be an interesting read.

@cr1901@mastodon.social @mcc@mastodon.social Sounds like it's time to dig out the old Amstrad CPC 6128 from the attic. CP/M for the win! 😎

@LarsHenriksen@mastodon.nu Uha - det er et ret omfattende spørgsmål med en hel masse variabler, så det er ikke så let at svare på. Der er to ressourcer, der er i spil: RAM og GPU-kraft. RAM-mængden har indflydelse på kvaliteten af de svar, man kan få fra sin egen LLM. Jo større en LLM er, jo mere nuancerede og pålidelige svar kan man få, alt andet lige. GPU-kraften har indflydelse på, hvor hurtigt, man kan få sine svar. Jo større din LLM er, jo flere kræfter skal der til for at lave et svar - igen, alt andet lige. Udfordringen er altså at finde en LLM-model, der leverer svar, der er "gode nok" til at opfylde ens behov i forhold til kvalitet og hastighed, og så finde hardware, der passer til netop det behov. Jeg ved på ingen måde nok til at kunne give råd om store implementeringer af LLM-løsninger, men jeg kan illustrere nogen af udfordringerne med min egen løsning. Jeg bruger en lokal computer med Ollama til inferens og AnythingLLM som interface. Hardwaren er en AMD Ryzen 7 8700G med en indbygget Radeon 780M GPU, som har fået tildelt 16GB dedikeret RAM i BIOS. Det er altså meget begrænset hardware, ikke ulig den, man kan finde i en kraftig bærbar computer. 1/2

@LarsHenriksen@mastodon.nu Ingen årsag. :) Nu ved jeg ikke, hvor du er i forhold til selv at køre en LLM, men hvis du vil prøve at dyppe tæerne lidt i bassinet, kan jeg anbefale at installere AnythingLLM (https://anythingllm.com/download) på din lokale computer. Det kører på Windows, Mac og Linux, og det har nået et niveau, hvor systemet selv kan finde ud af at gå på nettet og finde oplysninger, der er så opdaterede som muligt. Selv på en relativ begrænset maskine (16 GB RAM er nok at foretrække) uden et dedikeret grafikkort, kan du komme afsted med at eksperimentere med mindre modeller som den ovennævnte Mistral-3:3b eller Gemma4-e2b (Gemma er Googles LLM, som ligger bag Gemini i en meget større udgave). Personligt foretrækker jeg Mistral, da Gemma har det være lidt af et sludrechatol. Der kommer tit nogle meget lange svar på et simpelt spørgsmål. :)

@benroyce @mihamarkic @rubbel @joacim @randahl Nio tried this in Denmark, and it failed pretty quickly. There are a number of issues, and the first has already been detailed above - no one wants to get a heavily worn battery in their new car when they swap. That can be fixed if you do not own the battery, but rent it instead. In that case a lower performance may be a nuisance, but not a financial burden. However, what happens if you want to sell the car, and the potential new owner lives far away from a swapping station? Still rent the battery for a signifikant monthly fee, even if you never or rarely get to use the swapping feature? We hade ONE swapping station in Denmark, so... Also, the Nio models that were on sale here in Denmark had pretty slow fast charging, so driving your car on vacation in countries without swapping stations was less than ideal, compared to a car with a fixed battery. Considering that the latest batteries from both CATL and BYD have sub-10 minutes charging times on sufficiently capable fast chargers, I don't see battery swapping going anywhere - it was a stop-gap measure while battery charging was comparably slow. Now? - a solution looking for a problem that has already been solved.

@hanscees@ieji.de @ronfromnl@mastodon.social @emmalbriant@mastodon.online I'm genuinely torn on this one. Yes, I can see how the GDID can be used to identify political enemies and nosy journalists, should an evil regime with sufficient additional surveillance want to do so. However, it is clear that in a normal society with laws and search warrants and subpoenas, identifying someone by correlating online activity with the activity on a computer with a specific GDID is no small task. From what I understand, the user made the mistake of using the same computer with and without a VPN active, and that was how they eventually identified the users identity. They did not get it from Microsoft, or as a minimum had to verify it by other means (logins to social medie) as well. I honestly don't know. I can see all the potential issues with the GDID and how it can be abused, but removing it could is not without problems of its own, both for Microsoft and for law enforcement. Of course one could make the argument that any hacker worth their money would use a Linux computer anyway and simply sidestep the GDID issue entirely, leaving it to be a tool for surveillance of non-technical users only. I can see that clearly as well. Opinions, everyone? What am I missing?

@jeppe Jeg kan på mange måder godt følge dig, og jeg er enig i, at dækningen af USA og deres vanvid er kvalmende - men jeg har det også sådan, at det er pinedød nødvendigt, at det bliver tværet lige op i ansigtet på alle dem, der af en eller anden grund stadig mener, at der er noget som helst at lære af USA, rent politisk. Ja, Morten Messerschmidt og vedhæng - det er jer, jeg kigger på. Jeg har nok også selv været der, hvor jeg har taget demokratiets beståen alt for meget for givet, og har i bund og grund ikke været klar over, hvor relativt lidt apati, der skal til, før det hele kan ramle om ørerne på os alle sammen. Der skal simpelthen være så massiv en dækning af det her, at der ALDRIG er nogen, der skal stille sig an med hængende underkæbe og sige "Nåååårh....okayyyy....ja, øhhh, altså, det var jeg da godt nok SLET ikke klar over...." Så må vi leve med kvalmen eller finde over på Mastodon, indtil tingene falder lidt til ro igen. 🙂

@ronfromnl@mastodon.social @hanscees@ieji.de @emmalbriant@mastodon.online I've been through the answer provided by Claude three times now, and there is something I can't wrap my head around - perhaps one of you can explain it so I get it. 🙂 It says that MS only needed to have the GDID, the IP and the timestamp - then the rest could be filled in from the corresponding logs at ngrok. How did ngrok get the GDID in the first place? Is it sent as a part of the network traffic, embedded in the TCP/IP packets? Handshake? Magic incantations?

@sb@metroholografix.ca I've noticed similar accounts commenting on YouTube. They have been created years ago with minimal information, and suddenly they come out and go full throttle with pro-Russian arguments in debates on videos related to the US, EU and Ukraine. It's not ideal that Hetzner is hosting the email for them, but then again - if the Hetzner account is created by a EU citizen with Russian ties, it's almost impossible to detect. Personally I would contact the account owner and ask them a couple of follow-up questions and tell them to go elsewhere if I felt something was fishy.

@benroyce@mastodon.social @EUCommission@ec.social-network.europa.eu Oh, not to worry - the reality on the ground is nowhere near the mouthpieces you see online. European politics are turning right, yes - but extremism is generally kept at bay in the real world. You'll always have the outliers shouting the loudest and claiming to speak for the oppressed and the minorities, demanding everyone else respects their ekstremist views while respecting no one.

@georgetakei Oh, he is cursed and a genuine monster, not worthy of any kind of love for sure. No two ways about it. The father, that is. 👹

@C0rn3lius@aus.social Oh, not to worry - that is a feature that is in no way limited to EVs. Whatever fuels your car, it will still be able to spy on you anyway. May I suggest a model year 2021 Hyundai Kona Essential EV? It does not have any kind of internet connectivity built-in, and can't even communicate with the Hyundai app on you phone, even when connected to Android Auto on the same phone. The car is basically a Nokia 3310 phone.

@sturmsucht@mastodon.social Someon has to explain this to me: "No one can make money there, which makes it unattractive to media outlets, brands and creators. That's why it can't scale." Exactly what does those two things have to do with each other? Why is the earnings of media creators tied to whether a platform can scale or not? That said, extremely few make real, serious money on YouTube FROM YouTube - they make money from all other kinds of partnerships and sponsorships and promotions and a host of other arrangements that pays the bill and keeps the light on. So yeah - "ethical data abusers" indeed. Can't wait for Fedibook to go full throttle and lay these asshats to rest...

@benjaoming Ah - I think I see where we fundamentally disagree about all of this. We definitely agree on the desired outcome, but not on the means to get there. :) Basically I think it boils down to the fact that you want a technical solution where different actors have a part in the responsibility and possible culpability, where I want a legal solution that isn't relying on a technical foundation, but rather on a legal framework that outlaws scraping content where it isn't explicitly stated that it is available for AI training. That brings me back to the application versus protocol again. You can have an application where the content is the payment, and where you as a user allow AI scraping as a way to pay for the service. It will have to be clearly stated that this is the agreement between you and the service provider, like it is today with e.g. Facebook and others. Think about it: You can either start playing whack-a-mole with flags added to protocols, or you can simply outlaw scraping unless explicitly allowed - protocols be damned. Either way you will need a legal framework to handle this, and a protocol flag without the legal backing is at best worthless ("Do not track"), at worst a sense of false security. I think the fight to get a general ban and a ban per protocol will be just about the same.

@juliette so if people - like me - live in a small village where most jobs are 20-30 kilometers away and they don't own an EV, they can basically go duck themselves because they chose to be poor and unable to afford to live in a large city with public transport that runs more than every two hours? And I'm sure that is really not what you meant, but that is very much the real life result of what you are suggesting. People can't just somehow magically afford an EV, just because energy prices go through the roof, and buses doesn't magically start running every 20 minutes in all directions from everywhere. And yes, I am one of the lucky ones who actually owns an EV, but given that housing is cheap in my village, there are a lot of people who are way less fortunate, and if they want to keep their job, they have no choice but to fire up the old bucket every morning and get going. Telling them to suck it up and stop whining because they have to cut down on food to be able to afford fuel seems somewhat harsh in my opinion.

@benroyce@mastodon.social @EUCommission@ec.social-network.europa.eu To be honest, I'd like to think that any high profile account entering social media would be ready for the mindless drivel from people on the fringe, and, additionally, I would hope that the would be enough sane voices countering said drivel. Let's be clear. I'm a nobody. Yet I hope that people like you and I - although divided by the Atlantic ocean - can help encourage those who has the reach and the power that we lack, to keep talking truth to power, however trite that expression may be. The world relies on that right now.

@benjaoming I'll be the devils advocate here. If the protocol forbids scraping, who is to blame in a legal fight if scraping occur? The application provider? Because the scraper never touched anything related to the protocol and never saw if the flag was set on the post. Is the flag to be set at all times? Can you prove that the content was transported by the ActivityPub protocol, and that the scaper definitely was aware of that? Because if you can't, legal action is not possible anyway. This is not about what is right and what is desireable, it's about what is possible to prove in a courtroom and who you can blame for any misdeed. I am pretty sure that a good lawyer will be able to shift the blame to the application displaying the content to the scraper, or at least make the case that the scraper had no way of knowing that this particular content was off-limit. Yes, you can implement an extension of the protocol with the field you talk about, but it will be of no use at all, unless you get all possible Fediverse clients to agree to that field and implement precautions against scraping if the user has set the field. If the scraper is successful anyway, the Fediverse client will be the one to blame for not implementing the precauitions well enough.

@wjmaggos@liberal.city @georgetakei@universeodon.com I have old friends and colleagues in Ukraine, and according to them, that election was already being influenced heavily by the Russians (probably even more than Russia is trying to influence the election in Hungary), so there is that. Other than that, I tend to agree with you. Foreign powers should keep their grubby little paws of local elections in other countries.

@tim That said, you should be aware, that whenever you boost a post, it is the same as saying "this is so interesting that I think that anyone following me should read it". So, a boost is basically a share, which is the same as posting something yourself, and that can clutter up the feed for people that follows you pretty quickly, since there is no algorithm selecting what to show to your followers - it all goes out to us. I initially made the same mistake, and I think it is one of the things that isn't entirely clear to newcomers here on Mastodon. So, use boosting in moderation and use favorites to acknowledge a post - that way nobody gets overwhelmed. In short: Boost = Share with ALL your followers Favorite = Like

@rashunda My apologies - I was just notified that I should not reply to this post since I am not following the group. I just followed the hashtag and the post came into my feed. Apologies if I offended anyone.

@rashunda Ah - sorry. Had no idea. Is this a US thing? It's not really something I've ever heard about here in Europe, but there may definitely be areas that I don't know about where this is a thing.

@sindum@mstdn.dk how about a common feed with 3 checkboxes, Groups, Friends an Follows? That would allow me to quickly limit what I want to see at the moment. Other than that, I think I would prefer three different feeds on separate tabs.

@juliette granted, my guess was wrong then. However, then you have to know first hand that public transport isn't a usable option for the vast majority of people in a case like yours. One child goes to the daycare, one to the kindergarten and the oldest to the school - and mum and dad works 20 kilometers away. There is no way that will ever be possible with public transport in small villages.

@benjaoming I don't follow the reasoning behind your example here? If the field in the ActivityPub protocol cannot be enforced (like the entry in robots.txt), why bother then? It will just be like the "Do not track"-field in your browser settings that will give those that don't know any better a false sense of security. Or do I misunderstand your example? I still firmly believe that this problem can only be solved reliably at the application level, not the protocol.

@rashunda @blackfedi Ehhh - BIPOC?

@benjaoming I'm not an expert on this in any way, but I think building a legal consent into a data transfer protocol might be difficult, especially if said protocol is encrypted. You basically could not verify what is being transferred without decrypting the data stream, and in essence perform i man-in-the-middle attack. I'm pretty sure that the legal framework has to be built on top of the application, not the protocol.

@asbjornn@expressional.social Fejl F51 på en Miele *kan* være for lavt tryk på vandet ind i maskinen - altså en knækket tilløbsslange. Hvis du trækker maskinen lidt ud af skabet, så slangerne rettes ud, er det nemt at kontrollere. Når det er sagt, så inkluderer "right to repair" også "right to fuck up by yourself". 🙂 Jeg har selv prøvet at bruge alt for lang tid på en vaskemaskine, bare for at se en hvidevaredoktor fikse tingene på et kvarter.

@asbjornn@expressional.social Start med at checke vandtilførslen og afløbsslangen. Begge kan knække sammen over tid, og pludselig har du købt en ny opvaskemaskine og ender med at sælge den gamle for 500 kr. på DBA, selvom den ikke fejler noget overhovedet. Billigste opvaskemaskine vi nogensinde har købt. 😁

@sindum@mstdn.dk TBH I don't remember how Facebook does it - I haven't had the Facebook app installed for about a year or so.

@antopatriarca @marymessall @tchambers @anildash Well, some lawmakers definitely DO make it expensive for tech companies to break the law: https://www.euronews.com/next/2025/12/17/eu-takes-on-big-tech-here-are-the-top-actions-regulators-have-taken-in-2025 No, not the US lawmakers, I get that - but there are other lawmakers in the world... 😉

@antopatriarca @marymessall @tchambers @anildash They do need a legal framework for their actions, so things definitely takes time - and perhaps sometimes to much time to be really effective. Also, there are misguided efforts like the age verification system, that most likely first came from a good place, and then was warped by political interests from bad actors. I think this sums up the two sides pretty well, actually. You can either have things happening at a slow pace (but then be pretty well debated and thought through), or happening at a faster pace but be done haphazardly in some cases. Balance is key here, and while I'm not exactly one to make excuses for politicians, I also do understand, that what might seem obvious to someone who live and breathe the open web, it's not necessarily the case for someone who does not.

@sleepy62 Holy cow - that is quite the production, you have there. 12 kW inverter? My first thought was that this was begging for at home battery, but then I saw the actual numbers - 31 kWh may be a bit over the top in order to cover your usage when there is no sun. :) Is your home fully electric? - and isn't 63 kWh on a single day still quite a lot? Our home is fully electric and we also have an EV (doing around 45000 km a year), and we're using about 21000 kWh a year, or 57 kWh a day on average, and that includes winter months with a lot more usage for heating than now.

@juliette hold on - it's not like all the oil in the world goes through that strait. We will be missing 20%, not 100%.

@rimu @juliette That is not how economy works, to be frank. When supply is lower than demand, prices will go up and make some kinds of transport (and other use of fossil fuel) expemsive enough that it is either not worth it or alternative ways of solving the same problem will be used. This basically means that things that are not necessary will be reduced or cut away. That could be anything from going on vacation to buying products that are energy intensive to produce and reasonably easy to do without. All that said, we are moving towards summer in the nothern hemisphere, and that is both reducing the need for fossil fuel and increasing the production of green energy. It can't make up for the 20%, but it will help to keep the sticher shock lower than without the help from green energy. All in all I stand by my two original statements: Not considering some kind of tax relief to those that have no alternative to using fossil fuel and are without means to go the green route is elitist, and we are not running out of fuel since supply and demand will be governed by pricing - some things will be deemed unnecessary and be put on hold until prices reach a level where their percieved value will be higher than the asking price.

@sindum@mstdn.dk that said, after thinking a bit more, I realy think I would like the three boxes (could be a dropdown that activate or deactivate a category as well, but tha may be bad UI) and be able to choose what I am in the mood for today. Sometimes I just want to know what is going on with my friends, and not necessarily have it all cluttered up by a crapload of posts from the groups I participate in, and on those days I most definitely don't want the world to interfere as well. 🙂 If three checkboxes are impractical, three tabs - then I can at least switch between the ones I am in the mood for today.

@sleepy62 I want a battery like that! 😁 With an additional car and a climate a bit colder than ours, usage is really not that extraordinary - and yes, 1 kW sustained usage with a couple of large fish tanks is probably to be expected. We're currently looking to add a 10 kWh battery to our small (7 kWp) solar system as well as a 6 kW hybrid inverter, and we're debating whether to increase both solar and inverter capacity, or add more battery capacity instead (the budget is fixed, so something has to give). Danish power distribution adds tariffs to peak usage times, so we have 4 hours in the morning and another 4 hours around dinner time where electricity is significantly more expensive, while being very cheap during off-peak hours at night. I don't think we will get to a point where we run entirely on our own electricity, not even in the summer, but it would be nice to be able to operate on battery power during peak hours, and then charge during off-peak hours from either cheap grid electricity or our own solar panels. We use 3-4 kWh during the peak usage hours in winter, which means that a 10 kWh battery is slightly smaller than needed, so right now I thing we are going with a bigger battery and live with our small solar system - it can easily charge the battery in the summer anyway.

@juliette BTW, no - we're not running out of oil. We have 47 years left in the tank at the 2024 rate of consumption, based on the reserves we currently know of. https://www.worldometers.info/oil/ Given the current advances in green technology, the transition away from fossil fuels will be done well before that, leaving plenty of reserves for the few use cases where we can't find a realistic alternative.

@benjaoming Again, it's a protocol, not an application. Think about it: It is not the protocol that exposes the content to the AI scraper - it is the application. Spam is not prevented by legal restrictions of the SMTP-protocol either - it's a feature in the application instead, and it is usually activated by the user (or at least possible to deactivate). I understand what you want, and I get the reasons for it, but restricting what protocols can carry and for whom, is usually a bad idea in the long run. There is a always a use case that you did not think of, and suddenly people starts forking or modifying new versions of the protocol, with backwards compatibility going down the drain.

@LarsHenriksen@mastodon.nu En hurtig opdatering: Jeg prøvede lige at installere AnythingLLM på en ældre HP ProBook med 8GB RAM, og det kunne godt installeres - det kunne bare ikke køre nogen form for modeller. ;) 12 GB RAM er et absolut minimum, mens 16 GB RAM nok i praksis er den nedre grænse for noget, der kan bruges til noget.

@LarsHenriksen@mastodon.nu På den hardware kører jeg primært 3 modeller: Mistral-3:3B Mistral-3:8B Mistral-3:14B Den mindste model (3b) bruger jeg primært til hurtige opslag som jeg ikke skal bruge til noget vigtigt. Her får jeg omkring 18 tokens/sekund, og svaret starter med at komme stort set øjeblikkeligt og er færdigt på under et minut. Den midterste model (8b) er den, jeg bruger når der skal laves en lidt mere omfattende indsamling af data fra flere kilder og holdes styr på dem, indtil de puttes i en sammenfattende tekst. Her får jeg omkring 8-10 tokens/sekund, hvilket giver mig et svar på 1-3 minutter. Den største model (14b) bruger jeg, hvis der skal samles data fra flere kilder og laves en grundigere analyse af dem. Her får jeg som regel 5 tokens/sekund, så et svar tager tit 5-10 minutter. Kvaliteten er ofte på et niveau, der ligger omkring eller lidt under kvaliteten på https://chat.mistral.ai/chat Alle svar kommer serielt, så ét svar skal være færdigt før det næste begynder at komme. Det vil være håbløst i et scenarie med 1000 elever, og det er derfor, jeg slet ikke tør sige noget om, hvordan det løses i praksis. :) 2/2