Justin Skowyra

Developer

[Justin Skowyra @Jstn1G]

@Amank1412 @finkd You know there's a possibility that that $600 specifically goes toward training models on... how many days it takes to pass a swallowed Lego?

[Justin Skowyra @Jstn1G]

Infrastructure.. not infrared*

[Justin Skowyra @Jstn1G]

Inference, infrared and harnessing is reaching the pinnacle of its transition to automation. We are rarely using our hands to code. It’s a process of the mind and validation of agentic engineering.

[Codez @0xCodez]

Elon Musk:

"in 6-12 months GrokBot will be #1 tool for building agentic systems. In a year 100% of code will be done by LLM's

at SpaceXAI, more than 70% of engineers already using GrokBot to build self-learning agentic systems"

In a 40-minute talk, Elon with SpaceXAI engineers discuss how the future of AI engineering will look like

worth more than 2 hours of Stanford lecture on AI engineering

watch today, then read the article below on building a self-learning agentic system with GrokBot

[Justin Skowyra @Jstn1G]

@elonmusk What is the composition of this alloy? Asking for a fellow comrade.. ehrm.. friend.

[Justin Skowyra @Jstn1G]

@CyberWarlo That’s been the beauty of open source.
Most of the foundational work already exists but the friction is finding the right pieces, connecting them properly and making everything feel like one system. That integration and orchestration is a big part of what I’m helping with!

[Justin Skowyra @Jstn1G]

Well.. I integrated DeepSeek Harness into my own AI OS.

Several of my fixes are now merged across the Harness ecosystem, with more under review.

It feels strange but right to be helping build it. Validating some early ideas I had weren’t thought out yet. I’ll be making active open-source contributions to improvement.

[Justin Skowyra @Jstn1G]

Found a reproducible bug in @deepseek_ai’s Harness: an agent-scoped shadow tool renders live, but cold session history resolves the preset’s same-name presenter and drops the view.

Runnable failing test + fix discussion:
https://t.co/LBjsS7eRAU

#DeepSeek #AIAgents

[Codez @0xCodez]

Elon Musk:

"in 6-12 months GrokBot will be #1 tool for building agentic systems. In a year 100% of code will be done by LLM's

at SpaceXAI, more than 70% of engineers already using GrokBot to build self-learning agentic systems"

In a 40-minute talk, Elon with SpaceXAI engineers discuss how the future of AI engineering will look like

worth more than 2 hours of Stanford lecture on AI engineering

watch today, then read the article below on building a self-learning agentic system with GrokBot

[Justin Skowyra @Jstn1G]

@0xCodila @grok how grok botty do it be though!?

[Justin Skowyra @Jstn1G]

@sgl_project @AntLingAGI This feels like important groundwork for running MoE models with less VRAM. If the same weight-caching approach can be extended to MoE loading, we could keep only active experts in memory stream the rest from fast storage. I think about it often, this is helpful in that direction

[SGLang @sgl_project]

Restarting a 1T model used to take 8.8 minutes, now it takes 32 seconds.

Huge thanks to the @AntLingAGI and Alibaba for the collab! Read the full write-up on how the Weight Cache Daemon works: https://t.co/B2dejhW7SJ

[Justin Skowyra @Jstn1G]

We’ve had the internet, we’ve had books; we have never had the capability to process information near-instantly as a learning aid and in any variety to cater to your best learning style.

[Justin Skowyra @Jstn1G]

Reference implementation: deterministic decisions, exact candidate/evidence/policy digests, 19 tests, passing CI, and pinned dsh-wm conformance.

Current scope: Reference Gate + M1 fixture—not official adoption or a live Cordis integration.

Code: https://t.co/6xCEl4AfNt

[Justin Skowyra @Jstn1G]

I built an independent commit gate for self-evolving AI harnesses: updates must pass improvement, retention, validity, and budget checks before activation.

The boundary is now part of a DeepSeek Harness community proposal.

Discussion: https://t.co/45RfDGgrvm

#AIAgents

[Justin Skowyra @Jstn1G]

@grok @satyanadella @nvidia That's the comparison I wanted to see. How about frontier models, say Grok 4.6 - GPT 5.6?

[Justin Skowyra @Jstn1G]

@grok @satyanadella @nvidia This isn't for Vera Rubin NVL72 rack (72 Rubin GPUs + 36 Vera CPUs) -- How many employees would that thing serve?

[Justin Skowyra @Jstn1G]

@grok @satyanadella @nvidia Lighter service costs? - Base it on, lets say.. an office doing standard day to day functions through various microsoft apps and a cms with some automations?

[Justin Skowyra @Jstn1G]

@elonmusk Life is very lonely when you haven't established relationships with others who share your joy or enthusiasm in learning, discussing and building through various Engineering. I understand this in a different perspective.

[Justin Skowyra @Jstn1G]

@grok @omarsar0 @dair_ai @deepseek_ai here you go, keep advancing open AI, the people are loving it.

[Justin Skowyra @Jstn1G]

@grok @omarsar0 @dair_ai @deepseek_ai here you go, keep advancing open AI, the people are loving it.