Auston Horras

[Auston Horras @Austontatious]

@MystiqueMide @andy1eth (I mean ive run lots of local models, just never tried to seriously replace a frontier model with one)

[Auston Horras @Austontatious]

@MystiqueMide @andy1eth Specifically CAN use Claude and GPT-OSS with Antigravity. But I have the horsepower to run a big local model so I'm gonna try setting that up when my Codex resets at 6:30.

[Auston Horras @Austontatious]

@Real_SilverLine @MemoryReboot_ Data privacy, redundancy, cloud dependency, specialty models, refusals/limitations, customization. Your argument is like "whats the point of having a car? You can always Uber or take the bus". Looks like youre a dev - do you own a gaming/dev rig or just work in the cloud? Same

[Auston Horras @Austontatious]

@jaseeey Yep. Just hit me this last weekend. I just went to the $100 plan like 6 weeks ago, I wasnt even doing anything intensive. Smoked through my weekly limit of 5.5, my 5 hour of 5.3, AND $40 of tokens in maybe 6 hours. And no one could explain it. I canceled.

[Auston Horras @Austontatious]

@thsottiaux Are you trying to distract from the fact that you nerfed EVERYONES usage? This is ridiculous.

[Auston Horras @Austontatious]

@tinkerersanky @OpenAI @thsottiaux I got surprised by this as well! I even got $40 of top up and burned through those in 5 hours today. Im going to be taking a hard look at Gemini/Antigravity for sure, because this isnt sustainable.

[Auston Horras @Austontatious]

@TheEconomist Why would it be ok for companies benefitting from being in a rich country to offshore production to poor countries, but poor people can't access the market in reverse? It seems like the same thing, just its ok when owners do it and not when labor does.

[Auston Horras @Austontatious]

@BOJANGLES6253 @krassenstein If Iran was 3 weeks away from a weapon 2 months ago... wouldn't they already have one? Cuz its been more than 3 weeks,. Before that iN June 2025 they were about to have a weapon and we "obliterated" it. They've been 2 weeks away for almost a year. Is it 2 weeks yet?

[Auston Horras @Austontatious]

@nicksortor Pwrfect hes either launching nukes or going to Walter Reed.

[Auston Horras @Austontatious]

@hashjenni Oh mostly because you got to die at 35 so you could stay hammered basically all day from ago 10 on

[Auston Horras @Austontatious]

@Real_SilverLine @MemoryReboot_ Also depends how serious you are about AI stuff. $8k to run openclaw to check your email is dumb, $8k to rapid prototype 3 new ideas a week might not be.

[Auston Horras @Austontatious]

@Real_SilverLine @MemoryReboot_ So you have a chance when they try to turn the switch off. Youre right though, I have a big homelab and I'm running some experimental stuff on it but I do basically everything with Chat. Im tempted to go get a netbook again, Im always remote anyway.

[Auston Horras @Austontatious]

@alexocheema @TheAhmadOsman Memory and architecture support are the biggest barriers Ive hit in local AI. Not speed. I mean speed matters, but not if you OOM before you ever get that far. I agree - unconventional but these are unconventional times.

[Auston Horras @Austontatious]

@Kannanigga @alexocheema @itsry16 Its not "misinformation" anymore than 614GB/s claimed benchmark in the lab compared to real world performance is "misinformation"

[Auston Horras @Austontatious]

@Kannanigga @alexocheema @itsry16 If you have a use for that compute and you cant get a top tier mac studio until who-knows-when, you CAN do basically all the same work on the parallel MacBooks. Duh, there are tradeoffs. Obvs.

[Auston Horras @Austontatious]

@Kannanigga @alexocheema @itsry16 Yes of course it would be nice to run a B300 Blackwell Ultra, but since you cant GET one of those but you CAN get 4 A100s and match the memory or you CAN get 14 A100s and match the compute, its irrelevant that its not EXACTLY the same because of this or that bottleneck.

[Auston Horras @Austontatious]

@elonmusk Well, its not going to suck itself I guess.

[Auston Horras @Austontatious]

@sama Yes. To a point, obvs. If youre targeting a voice interface then speed (to me) is at least as important, but for most real AI use cases intelligence/accuracy are worth waiting a few seconds. Honestly its just smarter routing.

[Auston Horras @Austontatious]

@SullyOmarr Dude theres a 99% chance your info is for sale on the dark web for some dogecoin. How many notifications have you gotten from your credit card company, your bank, your hospital about breaches?