Vedhas Deshpande

Engineer. I write what I find interesting. Keep some sense of humor. Not an investment advice, do your own due diligence. blog:

[Vedhas Deshpande @VedhasD]

I have been reading about how people are buying Mac Minis specifically for setting up LLMs and OpenClaw locally. That’s what prompted me to check if my own modest rig was sufficient to run LLMs. To my surprise, it is—specifically for Small Language Models (SLMs).

On my modest rig:
AMD Ryzen 5 (AM4) + MSI B550 - PCIE 4.0 16 GT/s, RAM - 32GB

NVIDIA Titan Xp – 12GB VRAM (~2017 card 😐)

I am able to run gemma4:E4B and with a context of ~32k can successfully develop simple scripts using opencode using ollama.

<link below>

[Vedhas Deshpande @VedhasD]

Perfect size for mid level VRAM GPUs

[Google Gemma @googlegemma]

Meet Gemma 4 12B!

A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.

Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇

[Vedhas Deshpande @VedhasD]

New Money:
Anthropic, OpenAI and SpaceX IPOs

New Value Stocks:
$GOOGL $AMZN $MSFT $AAPL

[Vedhas Deshpande @VedhasD]

@xadhish @Google Congratulations 🎉!

[Vedhas Deshpande @VedhasD]

I have been playing around with llama.cpp on windows and wsl2 ubuntu.

I was using the same model file downloaded in windows on wsl2 via /mnt/c.

On windows llama would quickly load model files into the VRAM but not on wsl2.

TIL wsl2 uses a network protocol 9P to bridge the file system between NTFS (win) and ext4(linux). Hence the system calls are slow when made in wsl2 on windows drive.

[Vedhas Deshpande @VedhasD]

@pitdesi @jaredhecht Same vibe:

[Vedhas Deshpande @VedhasD]

It’s actually bought an RTX 3090 instead of Porsche.

[Itamar Golan 🤓 @ItakGol]

Men in their 40s used to have cool midlife crises… now they just have agentic workflows.

Bought a Claude subscription instead of a Porsche lmao.

[Vedhas Deshpande @VedhasD]

This is good advice. Even if you work at a higher level of abstraction, it is beneficial to know the space and time layout of the system.

[Abhishek🌱 @Abhishekcur]

The best thing you can do as an engineer is learn C and C++ deeply.

once you see memory, CPUs, caches, threads, and I/O without layers of abstraction hiding them, you start appreciating how much of modern computing is actually systems programming.

even a bit of assembly, hehe :)

[Vedhas Deshpande @VedhasD]

I can't answer your original question. Regarding market share, that might be true until Q1 2026, I believe. Post-Q1, token spend is being rationed, and frontier model availability has also suffered. In comparison, I have been defaulting to Cursor 2 and 2.5, which are based on Kimi 2 and 2.6. There's real value in open-source models, but I don't know how sustainable it is. I am seeing a lot of on-prem capacity building with Gemma 4 and Qwen 3.6. I think it's going to be hybrid in the future, with corporates building on-prem OSS models and reserving frontier-level models for high-end, difficult, and non-sensitive tasks.

[Vedhas Deshpande @VedhasD]

What’s a lightweight PDF alternative for Adobe on Mac?

I don’t like the Adobe overlays in the Chrome extension. It’s too bloated, and the app is sticky—I always have to force quit it when installing a macOS update.

[Vedhas Deshpande @VedhasD]

@typesfast My version:

When my selected options translate correctly across codeshare international flights.

[Vedhas Deshpande @VedhasD]

@NoahKingJr It depends on which harness you say hello 😂

[Vedhas Deshpande @VedhasD]

Trimble $TRMB linked SketchUp with Anthropic's Claude AI to bring conversational, AI-powered capabilities directly into 3D modeling workflows.

https://t.co/Tm8wlWJw7m

[Vedhas Deshpande @VedhasD]

I remember playing with Google SketchUp software.

Today I checked if it was in the Google Graveyard list, and it turns out it is not.

It was sold to Trimble Inc. in 2012.

It used to be my favorite indoor pastime during monsoon rains.

[Y Combinator @ycombinator]

It’s never been easier to design your dream house.

Draw a shape. Define your rooms. Set your constraints.

@DraftedAI generates complete floor plans, elevations, and 3D home designs in seconds.

Over the last month, 120,000 people generated 325,000+ home designs with https://t.co/XqC0LP5n3y.

[Vedhas Deshpande @VedhasD]

Is there a way I can merge 2-3 separate chats into one in gemini app?

[Vedhas Deshpande @VedhasD]

@david_nix There’s technically 2x PCIe x1 expansion slot but not for GPUs. It’s something like wifi/bt nic upgrades.

[Vedhas Deshpande @VedhasD]

Based. I have been postponing building llama.cpp and been using ollama. Now I can try this quickly!

[Georgi Gerganov @ggerganov]

llama.cpp now has an official website: https://t.co/vztdUpdBWL

Our goal is to make local AI accessible to everyone, and improving the user experience is a big part of that. On the new landing page you’ll find a single-line cross-platform installer. The installation provides a single unified `llama` entrypoint which you can use to run/serve models and interface with 3rd-party agentic applications.

While oriented towards simplified user experience, the new `llama` application also provides all the advanced functionality of the existing llama.cpp tooling with which experienced users are already familiar. Also note that all GGUF models that you might have already downloaded with llama.cpp in the past will be automatically available to use without downloading again (they are stored in the common HF cache on your machine).

We have many improvements in the pipeline both at the UX and at the engine level and we plan to iteratively ship new things over the coming months. One of the main focuses will be seamless integration with local-friendly 3rd-party agents (such as Pi). In the meantime, we’ll continue to listen for feedback from the community and adjust accordingly, so keep letting us know what you think and need.

[Vedhas Deshpande @VedhasD]

@david_nix Me too, every man's pre-midlife crisis :D.

[Vedhas Deshpande @VedhasD]

@paulg Few people have this luxury.

[Vedhas Deshpande @VedhasD]

everything is cyclic. it's good that hardware is getting the spot light but when the supply, demand and capacity stabilizes the spotlight moves to next money making layer (SaaS in the last decade). Hopefully the spotlight remains on hardware, infra, energy, quantum, and materials for long. :)