| About | https://vitobotta.com/ |
| SprintPulse | https://sprintpulse.io/ |
| Github | https://github.com/vitobotta |
| Location | Finland 🇫🇮 |
Codex/GPT 5.5 feels much more likely than Claude to check its own work without being asked.
Run the command, read the file back, test the change.
With Claude I still often have to say: please verify this.
Have you noticed that too?
Gemma 4 12B is the first one in a while that makes me want to download it today.
Image and audio go straight into the LLM backbone, no separate encoders, and Google says it gets near the 26B model on 16GB VRAM/unified memory.
Apache 2.0. Ollama, LM Studio, LiteRT-LM, MLX, vLLM.
I don't know much of IPOs but I sense SpaceX going out at $135 a share and a $1.75T target is absurd to look at.
Not in a bad way. It is the rare startup IPO with revenue, myth, launch cadence, and founder pull all stacked together.
Public markets have to price the Elon premium properly now. Cap table nerds will have a week.
22-year-old founder, $300M Series C, $1.8B valuation. Mach Industries is three years old and building defence tech. The bit that stuck with me is the production side.
Four new facilities by the end of 2026. Fundraising headlines are loud, factories are the hard part. That bit is easier to miss.
In a wild ride for 22-year-old founder and CEO Ethan Thornton, Mach Industries has raised another $300 million. It already has five autonomous vehicles in development and completed a major acquisition.
OK, I got sold on ChatGPT Pro. Kept seeing people say GPT-5.5 beats Claude on some jobs, tested the $20 plan, OMG, upgraded to $200/mo.
Main reasons: high limits, Hermes Agent/OpenCode, custom harnesses, and security work without the Anthropic subscription drama.
Been testing MiniMax M3 for a few hours now. Early days but it's doing well so far. Reviews I've seen from others are mixed, which makes me more curious to keep pushing it.
Not sure yet if I'll switch my default to it, need more time. But what MiniMax has done with ~230B params is impressive.
A model this size competing with things several times its weight class. The 1M context window doesn't hurt either. More testing to come.
