Johannes Schmidt

creator of MidTerm - The best agent orchestrator / neo ide / dev tool

business software dev lead by day space nerd by night

[Johannes Schmidt @spaceMonster]

MidTerm ... now with doom
run Codex/Claude/OpenCode/Gemini all Terminal apps in the world
in the browser but -> agents run on your hardware
you control it remotely (iPad, tv, pc, mac, Linux)

check it out it's your missing link in the your workflow
https://t.co/I0NqnQBTcW

[Johannes Schmidt @spaceMonster]

@jarredsumner

[Johannes Schmidt @spaceMonster]

@levelsio Guys it's save to buy. I don't have any funds free for an allocation today wich guarantees it'll rip to orbit.

Aegre sciurus se alit 🤌🐿️🐿️

[Johannes Schmidt @spaceMonster]

@levelsio Guys it's save to buy. I don't have any funds free for an allocation today wich guarantees it'll rip to orbit.

Aegre sciurus se alit 🤌🐿️🐿️

[Johannes Schmidt @spaceMonster]

@levelsio @cosmo_kappa I had a breakthrough

[Johannes Schmidt @spaceMonster]

@thdxr @jarredsumner after we read that blog post
https://t.co/i3iXL6QLz7

[Johannes Schmidt @spaceMonster]

Völlig falsche Betrachtung/Annahmen

Unterscheide erstmal Training und Inferenze.

Beim Training ist nicht der Up/Down link der Limitierende Faktor sondern die Vernetzung der Satteliten untereinander der über Laser Links (Glasfaser Niveau).
Wie lange der Upload der Trainingsdaten dauert ist (in Grenzen) egal. Die Entsprechenden Satteliten können dann auch in einem Hohen Orbit schweben weil auch Latenz zur Erde keine Rolle spielt.

Für Inferenz ist Latenz der Limitierende faktor. Die eigenntlichen Datenströme sind dann nur Prompts und output Tokens. State of the art bei OpenAi/Gemini sind 70-150 Token/s (was wenige kb pro sekunde entspricht) für jeden einzelnen Job.
Es muss nicht für jeden Inferenz Job das eigentliche Model hochgeladen werden sondern nur die Prompts (mit Document attachments). Das wäre für das Beam-Confinement was Starlink schafft (Kommunikation läuft über Starlink nicht direkt vom AI Sattelit zur Erde) kein Problem hundertausende Infernze Jobs ein einen Häuserblock zu beamen.

[Johannes Schmidt @spaceMonster]

@Gfilche Since I decided not to invest right away It's basically guaranteed it will rip.

[Johannes Schmidt @spaceMonster]

@levelsio @X Had the same issue (many month ago at a client of mine) I cycled the credentials (or even created the "app" again from scratch) then it worked again.
But I guess you tried that already.

[Johannes Schmidt @spaceMonster]

@skdh Power costs rise exponentially on earth if data centers on earth buy the remaining capacity that cant grow at the same pace. Wich will result in slow permitting.

Space AI costs scale inversely (the more you deploy the cheaper it gets)

There is break even point.

[Johannes Schmidt @spaceMonster]

@levelsio The sun is getting rid of lots of heat just fine

People like to be contrarian. But the goal is to be a correct contrarian.

[Johannes Schmidt @spaceMonster]

@LimitingThe He'll come around when Rivian announces their space AI endeavours. Those guys can't help to late-copy what Elon does.

[Johannes Schmidt @spaceMonster]

@StonksReddit @PTrubey @ZacksJerryRig wrong question

[Johannes Schmidt @spaceMonster]

at 100°C you need 1.45m2 of radiator to move a kW

just plug in the number of kW you need moved and you are done.

You are falling for the misconception of launch mass priorization. In the past when launch costs were much higher everything tended to be optimized for low heat output since radiators are the highest weight part and least functional item of space equipment.

For an AI sattelite its a central part of the design wich is easily tackled.

[Johannes Schmidt @spaceMonster]

@beffjezos @theo the slowness is killing me
even if it is better than 5.5 - it's no fun to wait that long

[Johannes Schmidt @spaceMonster]

@theo the slowness is killing me
at this pace it'll take ages to compare it to 5.5.
no fuckups yet

[Johannes Schmidt @spaceMonster]

@snicklink @LiberalMut Mit dem aktuellem X Algo bleibt der höchste Traum den man erreichen kann die #3 in einer Sub-Sekte einer Mikrobubble zu werden.

[Johannes Schmidt @spaceMonster]

The whole "Cooling is hard in space"-idea stems from the misconception of launch cost and mass to orbit priority.
ISS for example had high launch cost so they sized their radiators very low so they were constantly budgeting heat ingress.
Now with lower launch cost, just strap on a bigger radiator and go.

[Johannes Schmidt @spaceMonster]

@petesandor @levelsio Damn that's a significant shift.
You need to learn the habbit of "Lüften" now.

i try to never go over 700 in my office

[Johannes Schmidt @spaceMonster]

@BitcoinArchive if he wanted the price to go up he would buy on exchanges instead of OTC

[Johannes Schmidt @spaceMonster]

@biancaztan @TansuYegen @grok Tires and lockers