Brandon Pendleton

Principal Architect, AI & PE @ AHEAD | Professor(AI) @ UTC | Co-inventor @ SIGHTER | Fmr Head of Agentic Instructors AiGlobal @ MIT, AWS

[Brandon Pendleton @gitshipdone]

https://t.co/xKlQf6pip9

The future isn’t about believing. It’s about building systems sophisticated enough to understand what’s already here.

[Brandon Pendleton @gitshipdone]

@SouthernOrbs lots of people are, and lots of us are working towards that.

[Brandon Pendleton @gitshipdone]

@NVIDIARTXSpark but what if im already doing this with magic? also get me a 300 workstation these sparks need more juice and my xx90's and outlets cant handle anymore power!

[Brandon Pendleton @gitshipdone]

@elonmusk yall got a lot of people as customers! pretty much mapped it out. amazing work team! just know aws does this every year with us-east-1 and youve really only had one major outage @elonmusk

[Brandon Pendleton @gitshipdone]

@kenwheeler self hosting it is much better. but beyond that its cancer. and really self hosting serves a different purpose.

[Brandon Pendleton @gitshipdone]

@BULKHEAD are we getting Linux support on launch or no? If not id like to refund. Great game but cant sacrifice my dignity to play a single game.

[Brandon Pendleton @gitshipdone]

@matt503ea5sf9z5 👂

[Brandon Pendleton @gitshipdone]

@SkyXFunNext maine is friendly to everyone, we dont give a shit as long as you dont trash our state.

[Brandon Pendleton @gitshipdone]

@reactiverobot @TownAI break down compared to hermes or grok bot?

[Brandon Pendleton @gitshipdone]

@kenwheeler i’m not noticed a difference moving to the @SpaceXAI stack. local inference mixed with 4.6 and sol to review. once same and elon make up if not i’ll do my reviews with opus 5 for pay as you go.

[Brandon Pendleton @gitshipdone]

@billythalheimer @elonmusk @regentcraft and just to be fair, i can barely figure out to get a kite to fly, im amazed at this and wasnt throwing shade. appreciate individuals like yourself pushing us ALL forward.

[Brandon Pendleton @gitshipdone]

@billythalheimer @elonmusk @regentcraft even cooler!

[Brandon Pendleton @gitshipdone]

@matthew_pines you going to send it out and it’s going to arrive back in ten months, that’s when we realize we are the others

[Brandon Pendleton @gitshipdone]

@kenwheeler you all are invited down for the bbq just a heads up.

[Brandon Pendleton @gitshipdone]

agree. this is why in my theory we see massive degradation before a new model release because they are changing precision in flight to keep “uptime” and push what they do have to training cycles. but the normies don’t see it because it isn’t affecting general chat response it’s affecting deeper coding and other “reasoning” tasks.

[Brandon Pendleton @gitshipdone]

you don’t need a frontier model (cloud) to run good coding agents, I understand the “intelligence” in using say Fable 5.1, but i don’t need that to code or help design or manage a system. i don’t need to offload everything i know to a model.

date sovereignty is a thing also. in the near future the models will become more efficient, run smaller and become more intelligent. thus having the hardware ready while it’s still available is a positive.

on top of that it’s fun and a great tool to learn. i build giant neoclouds for private enterprise and it’s the same reason they are moving away from cloud providers.

[Brandon Pendleton @gitshipdone]

@elonmusk give it to me baby.

[Brandon Pendleton @gitshipdone]

@sattyyouneed not better, but we are taking single digit differences in not use cases. it’s much faster and much cheaper, in case it with pay as you go frontier for orchestration and review and you’re still better off imo.

[Brandon Pendleton @gitshipdone]

@UK_Daniel_Card put mgmt on its own vlan, vlan is only network devices, lock down all network devices (and remove shell), block internal and external ips from connecting to api.