Algorithmically promoted YouTube content has taken such a nosedive since AI video, only content as brilliant as this keeps me hopeful — youtu.be/HZAmbbTcQ3M as good as everything else is bad. What is YOUR Operating System (of Eukaryotic life) 🧬⚙️?

A night of testing Grok 3 against my codebase (plus o3 and Gemini Advanced 2.0 Pro Experimental). Any model can do stuff from scratch, knocking off trained patterns. But stepping into helping with a creative codebase is different — mikelev.in/futureproof/o3…

You know that "Copy Text" button? It strips out all the markdown and makes it impossible for publishing. xAI mikelev.in/futureproof/gr…

Think you have to ditch coding for service gigs? Nope! Think: don’t be a scribe, but learn to write. Scribes fade; coding literacy doesn’t. #FutureProof #LearnToCode #SkillsOverTitles

Neither @xAI #Grok3 nor Anthropic #Claude37Sonnet cut it when you're out on the edge doing something creative and trying to document it. Grok fails on the Markdown and Claude fails on the over-training on conventional patterns. mikelev.in/futureproof/co…

The current state of LLMs is unsuitable for genuinely innovative or unconventional software projects—specifically, projects that explicitly reject the common, established, widely-used coding patterns these models have learned. Anthropic OpenAI @xAI — mikelev.in/futureproof/wo…

LLMs are over-trained on FastAPI, making them struggle with FastHTML. Read our article: mikelev.in/futureproof/fa… #AI #WebDev #Python OpenAI Anthropic Google AI @xAI @CursorAI

After struggling with other LLMs, @xAI's #Grok3 surprised me. Read my story of giving it another chance: mikelev.in/futureproof/gi… #AIassistance #CodingWithAI

Two Moon landings in one week!

I pitted four leading LLMs against my codebase that breaks all the rules: Grok 3, ChatGPT o3-mini-high, Anthropic Claude Sonnet 3.7, and DeepSeek R1. mikelev.in/futureproof/ll…

#MCP is yet another API wrapper. The pitch though is that it got all those nuanced details sufficiently right (unopinionated/not-a-framework) so that it's a kind of TCP/IP or HTML. Fundamental. A lookup table or router to softwire otherwise hardwired dependencies. #AI

The #MCPProtocol hype had me thrilled at first, but vendor shenanigans left me bitterly disappointed. An "open standard" with closed-door moves? Feels like a great concept squandered, Anthropic. #AI #LLMs — mikelev.in/futureproof/py…

🤖 Your new research assistant: compare AI tools like ChatGPT, Bing Chat, Claude, Gemini, Perplexity, and Grok for deep research in our article—find your perfect match! [mikelev.in/futureproof/ai……] #AI #ResearchTools Mike Levin

Google's dominance comes from direct searches in the address bar. Perplexity should team up with Firefox 🔥 (by out-sponsoring Google) and have Perplexity as the default. @Amazon could pay for that.

NixOS upgrade to #Ollama 0.6.0: clean system rebuild. Learn more: mikelev.in/futureproof/ni… #NixOS #Ollama #Gemma3

Perplexity's 3 ENHANCED QUERIES per today are like the Genie's 3 wishes — Auto mode deciding to use one is like me saying, no THAT WASN'T A WISH! Perplexity #AI

My mind was blown. I absorbed what it said: “You don’t need to `git reset --hard HEAD`!” “Just give me hashes of a working state” “And I’ll diff ‘em.” Had to contemplate 🤯 mikelev.in/futureproof/ai…

Just dropped a deep dive into teaching AI to act with Pipulate’s MCP! 🐱 From cat facts to digital planarians, it’s a tinkerer’s journey into local AI agency. Check it out: mikelev.in/futureproof/te… #AI #LocalAI #Pipulate