I have been working with Anthropic Opus 4.5 for a few days and it does edge out Gemini 3 Pro and ChatGPT 5.1 as a sort of 'technical work companion'. I don't use it to write code in the Claude Code sense but I do bounce ideas off it and ask for suggestions. ( I am basically

hey Nikita Bier there really should be a way to eliminate engagement bait

Congratulations to the Dallas Cowboys fans. Really an amazing turnaround!

The X algorithm has decided I need more Japanese X in my life and it’s the most amazing thing.

Performative nonsense. By 19 hours of no sleep you are as effective as a drunk person. By 24 hours executive function degrades by 20-40%. You’d be way more effective if you slept for 8 hours and worked the other 28 then working 36.

This is surprisingly capitalist of the new mayor. Pleasantly surprised. Hope it continues.

I am feeling the AGI today. First time ever. Claude Opus 4.5.

China is stomping us in open models and the US AI czar is fishing for platitudes 🙄

I think it’s incredibly sad that China has top frontier class open models and now Europe has top frontier class open models while the best the US has is Llama 3 :(

Wow. Bun was going to strong ☠️

I want to especially thank Mistral AI for releasing the base models for Mistral 3. Fewer companies are sharing base models and this opens many use cases from custom instruct to non-instruct cases

An interesting observation. If you use ChatGPT 5.1 thinking ( through the app ) and offer it a novella that fits into the context and ask it to summarize it will fire up python and engage in all sorts of mechanics to dice the file, scan through each slice, and so on and so

Opus 4.5 is something else. 200k characters of context and I suddenly get a ‘bad request’ from the api. No other data. I pop the whole exchange as json into another Opus instance ( as a file so it uses a tool to read it ), it does some reading, doesn’t even think and then ‘here

Another AGI moment courtesy of Anthropic I am working on the stack that I use to communicate with Opus ( so Opus can read the code for that stack ). While working through a request Claude unprompted, decided an additional tool would make a task easier, and decided to write

Today, we’re excited to introduce Rnj-1, Essential AI's first open model; a world-class 8B base + instruct pair, built with scientific rigor, intentional design, and a belief that the advancement and equitable distribution of AI depend on building in the open. We bring

Men used to go to war …

I've put together an overview of the local LLM stack I use that leverages Parallax Gradient and a bunch of tooling I've built from scratch - primarily to learn more about models but also to be able to configure them for agentic use. Please let me know you think. ( synthetic

A quick tour of my home parallax LLM inference cluster including a 5 minute 'how to' so you can setup your own (cc Gradient).