Vitali

Building out cool tech at stealth startup and angel investing. Opinions my own. Ex GAF (Google/Apple/FB). OSes, indoor GPS, Pixel Buds, VR, cloud, AI, do it all

[Vitali @vlovich]

Working at a Bay Area AI stealth startup. This week modified llama.cpp to be able to run LLama 3.2 3B fp16 and Llama 2 13B Q8 (6 GiB and 13 GiB of weights) using just 2 GiB of VRAM on 2080. First time working on AI. Looking for talented developers with C++, AI, or GPU experience.

[Vitali @vlovich]

@richardartoul I am.

[Vitali @vlovich]

Trump is talking about 2028 as a distraction from him being a lame-duck president & to quell internal dissent within the party. Ignore the distraction

[OpenInfer @openInfer]

Sharing a sneak peek into how we are bringing large AI agents and context to any device by breaking memory barriers.
https://t.co/pbV4FvJCsZ

[Vitali @vlovich]

Lots of hard work went into this rearchitecture. Most customers are enabled and we're finishing off the migration for the remainder. The biggest change is that a long cacheTTL will not make interim writes invisible. https://t.co/aoT8HQr3sp

[Vitali @vlovich]

Steps to building maintainable software at scale while retaining high velocity.

1. Good test coverage
2. No mocks/fakes except for things like network I/O (use the real database if you can)

Step 1 is a massive and nuanced topic - lots of bad cargo-culted advice.

[Vitali @vlovich]

In everyone's hands because critical for commercial Internet to function (lots of applications of AI). Lots of unforseen unintended consequences (e.g. secure E2E chat) but net positive (even E2E chat I think despite hand-wringing). AI regulation will be net positive?

[Vitali @vlovich]

Long thread. AI hype + AI regs entered the main stream. My view as a very experienced generalist software engineer, angel investor, and an observer of computer history. Not an AI expert/researcher. Technical grounding valuable to avoid hypothesizing unrealistic scenarios.

[Vitali @vlovich]

Gov (inc agencies) has middling track record regulating. Not useful for exponential tech. Gov classified cryptography as munitions (seriously). Roadblock to commercial Internet / succesful obvious 1A challenges. Cryptography best example / closest parallel I think.