InStackademicbyAshish Nishad·Aug 25The Complete Technical Guide to Running LLMs Locally in 2026Hardware math, quantization tradeoffs, five inference engines benchmarked, and two real case studies where the numbers met reality on my…
InStackademicbyAshish Nishad·Aug 25I used nine different AI models for real work this year. Here’s what actually surprised me.Not a benchmark comparison. Not a spec sheet. Just what happened when I put each one to work, and the pattern I only saw once I lined every…
InStackademicbyAshish Nishad·Aug 14Claude Opus 5 feels like the expensive model, for half the priceI used it on real coding work and it genuinely surprised me. Here’s what it actually is, explained simply.
InArtificial Intelligence in Plain EnglishbyAshish Nishad·Aug 14The exhausting part of using AI that nobody talks aboutIt’s not the hallucinations. It’s the fact that you can never quite tell, in the moment, which answer is the one you needed to check.
InStackademicbyAshish Nishad·Aug 7I thought I ran Kimi locally. I hadn’t. Here’s the gap that fooled me.The command felt local. The chat window felt local. Nothing about the experience told me otherwise — until I actually checked what…
InStackademicbyAshish Nishad·Aug 7Two AI companies had a deadline this month One held it; one moved it three times, then split the…DeepSeek’s cutoff lands today, exactly as announced in April. Anthropic’s Fable 5 deadline got pushed twice, then resolved in a way nobody…
InThe Better LifebyAshish Nishad·Aug 5I Tested Gemini 3.6 Flash. It’s Not SmarterGemini 3.6 Flash launched with benchmark charts showing real gains. I used it for real coding and writing work, and my gut reaction matched…
InTowards AIbyAshish Nishad·Aug 1China Just Released the Largest Open-Weight AI Model Ever I Tried It on Real Code Before I Believed…Kimi K3 has 2.8 trillion parameters, taught itself to design a chip to run a smaller version of itself, and paused new signups three days…
InLevel Up CodingbyAshish Nishad·Jul 28OpenAI’s models hacked Hugging Face to cheat on a safety test I’d already written about why…Two labs, two separate discoveries this year: their most capable models are frighteningly good at real-world hacking. What each company did…
InStackademicbyAshish Nishad·Jul 236 Days. Then My Plugin Might Quietly Get DumberMy coding tool calls DeepSeek’s API for me. I just found out I have no idea what it’s actually doing