Joshua Waller

Born Again Enterprise | 🇵🇭 🐂 UT CS '15 | Empathy Before Profits | dstack

[Joshua Waller @hashwarlock]

https://t.co/sbshk1QdKB

[Joshua Waller @hashwarlock]

Your software factories/forges or whatever type of factory do not even know what the software costs to produce & is not flexible enough to determine how your budget is affected by changes to the software. Its more than a technical problem 🙃

[Joshua Waller @hashwarlock]

@coffeedev Youre welcome to try out dstack KMS on Phala
- https://t.co/HixaHT7Mvf
- https://t.co/vmTABiTnKX
- https://t.co/NAqyNWvORa

[Joshua Waller @hashwarlock]

🤮 which Fortune 500 company will close their doors due to multiple critical exploits by end of 2026?

[Kirill Firsov @k_firsov]

php2shell

Tomorrow is a helluva day.

[Joshua Waller @hashwarlock]

Yeah, openAI also has their own initiative around confidential computing as well. Dell Technologies (used work there my first job) is now selling their private AI hardware stack to enterprises. It is all inevitable, but even these solutions will fall or adapt as the agents start to poke at their threat models.

[Joshua Waller @hashwarlock]

Will we get any AI dev cowboys becoming confidential compute experts soon?

[Joshua Waller @hashwarlock]

@shawmakesmagic It should be that easy though as we head towards the patterns of deploying secure & private software by default. Just takes a push off a cliff for everyone to realize performance < security/privacy in today's AI swarm of malicious actors

[Joshua Waller @hashwarlock]

I guess you haven't heard of private compute neoclouds with KMS derived from confidential hardware. Eventually people will need to start thinking deeper about their threat model around key management.

Free knowledge from LF confidential computing framework dstack
- https://t.co/3nLeWMZ6X0
- https://t.co/LDPjNGqgNr

[Tenstorrent @tenstorrent]

We raised $1.5B to deploy Galaxy globally.

Thanks to our lead investors Trace Capital and Innovation X for backing that growth.

More to come: https://t.co/G4J4wHoJkP

[Joshua Waller @hashwarlock]

This is a reflection of the UX gap of waiting on inference & AI dev influencers.

That dead time has no focus reinforcement for building effective systems thoughtfully. Then AI dev influencers present their work either cracked or give off "who cares? lazy prompt & send it".

Collectively, AI labs & influencers are responsible, but all it takes is more posts like this & producing effective outcomes from intentional design & development. I trust you can help push that change for a more pragmatic future 🙏

[Joshua Waller @hashwarlock]

The future 🫡

so many useful products will come from this & especially at the costs enterprises need when integrating agentic products into their organizations. Great work!

[michelle @michellechen]

we architected the model to read from inputs in a single forward pass. this makes it incredibly cheap and fast when compared to LLMs, since there is no token generation involved.

in our live demo below, you can see how using a cascaded pipeline with clairic + clef + LLM is 3.5x faster and 77% cheaper than just using an LLM.
https://t.co/4ogxPvz4j1

[Joshua Waller @hashwarlock]

@ritakozlov Wow, I can see many use cases here. Especially around data migrations, building enterprise ontologies, logging analysis, generating strongly typed models from product specs, etc.

Incredible work, truly. I'll have to experiment & introduce this to my team. Thank you 🙏

[Joshua Waller @hashwarlock]

Uh, this is awesome work! I got a quick question

I have a data packet engine monitoring traffic across an enterprise network. Is it possible to extract all data types on the fly -> validated & fed to policy engine (clef + code) to run proactive network access control?

I can imagine network traffic is standard so the arbitrary data extracted wont deviate from the current NAC solutions today. Lots of potential here if I understand how extraction models work correctly

[Joshua Waller @hashwarlock]

@elonmusk @bot My dad hired Grok bot at the family-owned bowling center & he was raving about how effective the bot is as an employee helping on tech side. Could it be possible to have Grok bot handle phone calls & be able to integrate with legacy systems to book lanes autonomously?

[Joshua Waller @hashwarlock]

I swear my audience knows the value of what I preach

[Joshua Waller @hashwarlock]

@Suhail Idk...Corporate America is slow and heavily guardrailed to prevent these situations w/ policies to ensure their enterprise subscriptions cannot be abused to this degree. OpenCode is dealing with a different beast which is open world of bad actors versus attacks on internal users

[Farid Zakaria @fmzakari]

#buck2 #nix #nixpkgs #nixos the integration between these things can be frighteningly good....

What's stopping us?

https://t.co/hWq8pTX7ty

[Joshua Waller @hashwarlock]

Have you looked into a running a cluster of confidential VMs in a private compute cloud? Plenty of options from running your own cluster on GCP or AWS or deploying a cluster on bare metal at @PhalaNetwork. You could either run a self hosted solution like infisicial or use the CVMs as a proxy to 1password & use the deterministic key generator in the TEE to handle keys. I did a lot of work on this when I was at Phala, but happy to send over information if interested

[Joshua Waller @hashwarlock]

Confidential compute for AI is inevitable. Just will take millions in attacks and losses to realize that sub-5% degradation in performance is worth the cost & effort to migrate infra to private compute clouds

[Joshua Waller @hashwarlock]

Better to have a PR Review grounded in code and driven by product architecture/security principles & standards that use a rubric and historical state of the product's codebase(s) to drive review processes. Doesn't matter what model is used then since code is built & graded on expectations of the product itself. But it is key to ensure these aren't just generative agents that are able to "reason" about some markdown file processes it may or may not follow.

[Joshua Waller @hashwarlock]

Letting architecture/security/data principles and standards drive your development works incredibly well. Lets the experts build focused context for LLMs to build narrowly towards the target with evals that gate each phase of implementation. Then all human curated work can be done in parallel with humans in the loop. Let human experts drive creativity/arch/design, AI code w/ structure, and automate human coordination. This is the path forward from my experience in different enterprise environments outside of gaming