3B active parameter model beats Sonnet 4.5 in benchmarks. I'm all for the Chinese distillation industry. They are doing a way better job than their US counterparts.

What happens when AI can generate characters experiencing any story without pretending? What about when generated music feeds replace selecting the artists you listen to? As for adoption, the next generation rarely cares about the older generation's preferences.

Someone choosing to implement zero device ownership verifications in a globally accessible API is beyond wild. Smells like Opus.

DeepSeek V4 Lite news V4 Lite is under active testing with at least 1 inference provider who we verified has access but imposes strict NDAs - the codename is "sealion-lite" - has 1M token context window - *much better* than web / app model - it's natively multimodal

Benchmaxing is part of the process. Every number ceases to be a good target once you start to measure success through that number. But unless we max out all kinds of benchmarks, we will never get to the benchmark that measures all the sides of human intelligence.

Closing the loop is the real unlock. Not "AI in the IDE". AI that runs the system, takes feedback, fixes itself, and deploys. This is the beginning of the AI economy. Build for autonomy.

An agent ran into an issue when changing the status via a tool call didn't work. After a few tries, the agent went straight to issues through another tool, checked labels, created the missing label, and then changed the status. We hit the wall and now we go straight up.

few

Looks like all Ethereum L2s, except Solana, have been terrible investments so far. Every L2's token price has steadily nosedived after mainnet. Solana is starting its nosedived but the scam is well-crafted and it'll take time to inflate.

Shannon Sands Justin Schroeder The point people miss is that open-source models tend to be extremely jagged. If there were as well-rounded open-source model as OpenAI's and Anthropic's, even if the reasoning isn't as SOTA, it's a huge deal because 90% of tasks barely require reasoning beyond elementary.

There's something terribly wrong with these people. The impulse to surveil everybody is not what I'd consider within the range of normal, ethical, and moral behavior in a non-totalitarian society.

I wanted to give Linux desktop a chance. Been a while. All went fine until it turned out I had to install the wifi/bluetooth drivers manually. After installing the drivers, bluetooth connects to phones but not mice & keyboards. Oh well, let's try again in 5 years. Maybe then.

GPT-5.2 Pro was a step change in AI reasoning capabilities. Qwen3.5-35B-A3B is a step change in AI privacy.

Accelerate all things, especially health care.

Most AI critics aren't dumb. They understand AI well enough to instinctively realize it will replace their jobs in the near future. While I understand their reaction, AI will also cure all diseases and aging. It's not if, but when. That is worth the gamble.

start