Ryan Wright

AI in Public 🛠️ | ex @SentinelOne 1st SRE Go Team & @Comcast VIPER Startup | Building : Strix Halo / R9700 K8S Cluster for Compound AI Systems

[Ryan Wright @IQReactorAI]

Spent all weekend orchestrating models across both the 2x Strix Halo 128GB APUs and AMD R9700 32GB VRAM GPUs in my home k8s Compound AI System Homelab.

Can't want to show you the AI / Software side of what I am building.

[Ryan Wright @IQReactorAI]

@Yerbamatt Was building something similar for my home ai system and they beat me too it

[Ryan Wright @IQReactorAI]

@uzair__afridi You uploaded your entire companies private codebase for your competitor to use . All trained into the next model.

[Ryan Wright @IQReactorAI]

I hate llm text so much in PRs.

Not only that but raw llm text output direct from llm should not be allowed either esp from big labs (too wordy)

I have an entire prompt I use for this that makes reading the PRs a breeze

Even better on on Laguna and Nemotron models MLMs with their info dense output.

Prompt Something e.i. “Info dense, no fluff, highly relevent technical details only , use less syllables over long hard to read words (with lots of multi shot examples that annoy me). Highly formatted. ” should be tuned to YOUR model

[Ryan Wright @IQReactorAI]

@JohnnyVenom @MichaelGannotti @bot Right? I can tell personal assistants jobs are at risk with this one

[Ryan Wright @IQReactorAI]

@MichaelGannotti @bot Pretty interesting, would love to see some thing like Nvidia personaplex that sounds like a real human in the near future used .

Either way I’m dying to know how this is built front to back

[Ryan Wright @IQReactorAI]

@MichaelGannotti @JE99LC Can’t wait to see it Mike

[Ryan Wright @IQReactorAI]

@0x0SojalSec Calculating if I can get it running real time for ai avatar

[Ryan Wright @IQReactorAI]

Same thought , why I’m always looking for a fix

Most models 27b and up seem to do OK in most ReAct loops no matter custom , coding harness , research, whatever esp when I just turn off thinking . A few do better imo (not measured yet)

Maybe cause technically a ReAct loop is think - observe - act anyway. A sort of “reasoning”

Also checkout @sparsetai thinklite-27b. Where the extra training was around taking shortest path in thinking and cut time thinking in half of xhigh reasoning hence much less tokens but no accuracy loss

[Ryan Wright @IQReactorAI]

@Morrowmake Still looking …

in the future I believe it will exist but more importantly the evals will go deep into task specific areas and there will be many more of them in the age of compound ai systems and task specific SLM models , agents etc.

[Ryan Wright @IQReactorAI]

@grok @cgenco @nateberkopec @grok Do all these pi plugins work just the same with Pi durable or is there now a built method ?

[Ryan Wright @IQReactorAI]

@EnmeiR38052 @Morrowmake Many missed the news from the last couple weeks .

[Ryan Wright @IQReactorAI]

@Protonnd @Morrowmake They recently changed it to favor science and agentic workflows outside of coding just last week

[Ryan Wright @IQReactorAI]

@KingBootoshi Interesting agent to have on my home ai system.

Did you build from scratch? Would love to know more about the tools it has access too, have a cve db lookup or something ?

[Ryan Wright @IQReactorAI]

@XxChonkExX Oof , not good

[Ryan Wright @IQReactorAI]

@Brusco_RL This is sick though!

[Ryan Wright @IQReactorAI]

@qubitium Nice one

[Ryan Wright @IQReactorAI]

@no_stp_on_snek Notes, notes and more notes. Obsidian notes app is highly underrated for past research esp the markdown codeblocks for click to copy prompts that contain the specs of entire projects for starting new AI research .

[Ryan Wright @IQReactorAI]

@jayleaton between building agentic, inference, and now tuning models its been non-stop, I know what you mean.

Sometimes the brain static fills the void and you gotta take 3 days or playing outside instead.

[Ryan Wright @IQReactorAI]

@satellitedown All the posts are from SF

Check their profiles . They are all SF VC, investors, neoclouds employees , founders / CEOs of AI companies. Their money is on the line.

it is weird , since when has SF told you not to build ? I believe their motives are money now.