Marco Pfeiffer

@iam_zachi @thdxr But open source models are cheaper. Not much anymore but realistically maybe half the cost of Claude for the same task.

@thdxr People are getting ~$3000-4000 worth of usage with codex right now, so it’d be hard to fight those subsidies

i know this isn't the most appealing thing ever but i'm curious on the state of things would you pay for a $200/month sub that gives you access only to opensource models? say you got ~$800/month in usage

@melvynx It’s 1M. But it’s not like we are dealing with bug free software 😉

did you see that context window of Fable is down to 200k tokens ? 🫣

Yeah, but that still means that someone in the network has to convert that IPv8 package into an IPv4 packet since they are binary incompatible. That same concept exists for IPv6 too. It's called 464XLAT, and your phone is likely using it right now since mobile carriers are usually not providing IPv4 addresses. It’s just looking a bit nicer in v8.

@TheTrueNemo @Data_Knox Every ipv4 address exists in the 1.1.1.1.* ipv8 sub net. Transparent to ipv4.

IPv8 has been proposed... and the crazy part is how simple it is for being so powerful. - 8 octets instead of 4 - fully compatible with IPv4, no dual stacking. - Can be rolled out as a software patch, no hardware replacement - 1 device - the Zone Server - acts as DHCP, DNS, NAT (they call it XLATE8), ACL (east and west, north and south), NTP, and OAUTH - BGP routers reduced to ~175,000 prefixes vs today's 900,000+ - A new metric for routing protocols called Cost Factor derived from RTT, loss, congestion state, stability, capacity, policy, and geography Here is a blog that breaks down how it works (diagrams included): transitai.app/blog/ipv8-draf…

@Art0fLife_ What if I’m the negative person in my live?

He literally shares the proof that negative people ruin your life.

@TheMatheusFroes @Airapost Everyone was weirded out by this second wife scenario in that episode. I’m pretty sure it’s seen the same way as it is in our world.

@Airapost You can tell Noobs with this kind of reaction. This is meant to exemplify the fact that this is another world, an Isekai, different from Earth with different culture. To them, having more than one wife is normal. In fact, those following Milis and monogamy are seen as weird.

The first wife, holding her child, asking the second wife if she had fun last night. 😭

Every word comes from another. The German word Monster appears to be adopted from the French word monstre which comes from the Latin word mōnstrum. Interestingly, it appears that the English word has exactly the same source and it is coincidence that it resulted in the same spelling. But true. Ungeheuer is older.

@TheTrueNemo @GermanSimply_ Monster is a case of adapting foreign words. The actual german word is Ungeheuer

Good luck learning German. 🇩🇪

Der Witz ist, Ungeheuer ist eigentlich das Gegenstück zu geheuer. Ein Wort das fast ausgestorben ist. Aber „das ist mir nicht geheuer“ hat man vielleicht nochmal gehört. Das steht für mild, freundlich, vertraut, anheimelnd. Also genau das Gegenteil. Das Ganze stammt wohl von dem Wort „Heim“ oder „heimlich“ ab, aber da bräuchte man jetzt einen Germanisten. Es ist aber natürlich eine Sache woher ein Wort kommt und eine andere wofür man es heute benutzt.

@TheTrueNemo @GermanSimply_ Nein ein Ungeheuer ist ein Ungeheuer und man sagt selten das steht dir ungeheuerlich gut.. Also du hast an dem Ungeheuer das.lich vergessen

This is misleading. Resellers are able to get large quantities of components, pay the regular import duty (while also going through the regular checks), and then sell it to you at a reasonable cost. This is much more economical than the loophole that allowed individual parts to be shipped for basically free. The location of manufacturing isn’t relevant here. The only real downside is for very rare parts or parts that are technically illegal.

The EU just killed DIY PCB design With the new tax on imports (3€ per import of each category) I now would have to pay 3€ for every PCB component that I have to solder at home I can't buy PCB components in Germany because THEY ARE NOT EVEN MANUFACTURED HERE ANYWHERE ITS JUST CHINA AND AMERICA So instead of 20 bucks for one prototype PCB with 20 different components It's now 80€ Fuck you @vonderleyen

@DogukanUrker True, but I’m pretty sure that Kimi is a pretty good model regardless. It’s not a Gemini model after all.

Half my feed in last 2 days: "Kimi K3 beats Claude Fable 5 on this benchmark." So I trained a 0.8B model to beat GPT-4 on HumanEval. 98.8% vs 67%, on a single RTX 3060. Took 32 minutes. It's completely fake. This is why you need to be more careful on benchmark numbers 👇

@HarmAlhusen @claudeai The pro sub has a session limit of ~$20. it was never really usable with fable anyways.

@claudeai $100, that’s like 1 prompt with Fable :(

Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits. Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit. Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.

@luciascarlet @sophiiess_ And If you hold option while clicking, then the Window is captured without a shadow.

@sophiiess_ on macOS, you can hit space after hitting cmd shift 4 and just click on the window you want, which is a little quicker and easier than selecting the area if you want to screenshot a full window

did you really have to put a screenshot of your browser through a tool that gives it a background and shadow instead of just posting the screenshot

@NaMarrado_ @daniel_mac8 I think you are talking about Gemini models 😉 GLM is pretty good. Not SOTA good but very usable. I don’t have any experience with Kimi but I’m going to get some likely.

@daniel_mac8 What do you do? Own them These models are made for benchmarks, they still cant even compete But for their price point yeah they are a fantastic deal

Kimi K3 is about to have America rethinking its life choices tomorrow. What do you do with a Chinese model that’s approaching SoTA, but 50x cheaper? Oh, yea. And it’s open-source.

@bengunnnnn Die Idee ist nicht unbedingt schlecht. Wenn jemand deine credentials geklaut hat, hast du Zeit zu reagieren. Ist aber ungewöhnlich. Früher konnte man eine Überweisung auch noch innerhalb eines Tages noch zurück holen. Wie das jetzt mit Echtzeit ist weiß ich nicht.

Wir hassen Banken und die EU noch lange lange nicht genug alter 😭 ich will ne arschgefickte fucking Drecks Überweisung machen 😭 wie kann so etwas legal mehr als 10 Sekunden Aufwand dauern 😭

- Es ist nicht bei Openrouter, es ist nicht bei Cortecs. - Es gibt keine unabhängigen Benchmarks. - Qwen 3.6 ist längst raus. - Die Ministral Modelle sind nicht als "Open Source European" gekennzeichnet? Und die alten 8B und 14B Modelle von Mistral mit einem 30B Modell zu vergleichen, ist auch nur bedingt fair. - Richtig spannend wäre der Vergleich mit GPT-OSS-20B. Das ist etwas kleiner und inzwischen sehr alt aber auch sehr bekannt und hat in der Vergangenheit richtig gut performt. - Ich bin mir ehrlich gesagt nicht sicher was der chart dort überhaupt benchmarked.

Germany has launched one of the world's best open-source AI models. Soofi S, made by the Soofi consortium, is a 30B parameter model fully trained in Europe and tops the ranking for open-source AI. Huge moment for Europe, and finally some competition for Chinese open-source AI.

Ich finde es interessant dass Altersvorsorge Thema einmal ohne Geld zu betrachten. Die Idee der Umlage ist: dass jeder arbeitsfähige 20% seiner Zeit daran arbeitet die Rentner zu versorgen. Wenn ein Rentner mehr bekommen soll heißt das, dass Rentner mehr als 20% der Arbeitsleistung Deutschland bekommen sollen. Und wenn man sich die Demographie anguckt und weiterhin persönliche Betreuung haben will ist das vermutlich auch zutreffend.

Ein Eigenanteil von 3.364 Euro fürs Pflegeheim?! Wer sein Leben lang durchschnittlich verdient hat, bekommt am Ende ca. 1.500 € Rente netto. Ich bin kein Mathegenie, aber die Rechnung geht nicht auf. Ich bin ganz ehrlich: Sowas zu lesen ist alles andere als ein Anreiz für die private Altersvorsorge. Wie soll dieses System langfristig funktionieren?

Interestingly, this would still be subsidized. You could still take out more inference with those plans than you pay for. It would be really nice if the API pricing and the subscription pricing wouldn't have such a large difference. But increasing the subscription price, like shown above, or decreasing the API pricing by 10x is both out of the question, likely.

Nightmare of every vibe coder

@melvynx Use whatever works for you.

GPT 5.6 Sol > Fable 5 Change my mind

@thomaspaulmann GPT 5.6 might be a reason they can offer this. The fewer use it the less constrained they are. Although the current strategy feels scammy.

Why not just say if Fable is included or not? Are they so capacity constrained that they literally need to find GPUs to keep it running every week? Feels a bit desperate to keep folks off the new Codex/ChatGPT trying out GPT-5.6 Sol. I’m sure there are reasons. This is just how it looks from the outside.