I don't think anyone says it's impossible to put a data centre into space. I mean, look, we ALREADY have satellites running on solar doing compute. The question is whether it makes financial sense given the current technology.
https://t.co/qdQw4Jda6S
It's very cool that Apple shipped a 20B parameter on-device.
You can't put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today's standards.
A small model predicts from the query (or prompt) which experts to load from Nand into RAM. The key distinction from a typical MoE is that you do this once per query and then generate all the tokens with the same experts (instead of switching the experts for every token).
Honestly feels like the pinnacle of code editor experience right now.
It’s as light and blazingly fast on any moderate hardware as NeoVim when it’s needed. The Vim motions implementation is a chef’s kiss. So it totally captures old school senior engineers.
It’s fully featured for the majority of mainstream languages as a true classic flagship editor when it’s needed — very easy to set up. Capturing the majority of engineers.
And when needed, it rivals the best agentic coding editors with its own thread implementation and concurrent agents view. So now it captures non-technical folks — vibe coding looks pretty slick.
Testing Zed since the private beta back in 2022 was an interesting experience. Seeing it grow is mesmerising. Best-in-class pair coding experience, best-in-class rendering performance, and the ability to just switch off AI when it’s not needed then switch back to a fully fledged editor. It truly feels like NeoVim 2.0.
It’s virtually winning in most types of coding tool competitions (even competing with dedicated agent harnesses), except for JetBrains products which are also awesome and incredible feats of engineering, in a league of their own with a slightly different philosophy.
It’s very inspiring to see the @zeddotdev team making the right decisions one after another, especially with the latest open source and license changes. Amazing job from super smart engineers.
I remember when Death Note ended, and everyone complained Light freaked out because he needed a more dignified end...
...not understanding that their cool exterior only existed because they had control.
Decision making was the bottleneck all along. Productivity is the rate at which you make open-ended decisions, the rate at which you reduce future paths.
Many bad AI takes arise when people don't understand the jagged frontier.
Yes LLMs are superhuman in many ways. Yes LLMs struggle with basic arithmetic. Yes, they're still intelligent.
If you disagree, you probably have a very narrow, human-centric definition of intelligence.
🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.
🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.
🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice.
Try it now at https://t.co/GCdiMzk1Dl via Expert Mode / Instant Mode. API is updated & available today!
📄 Tech Report: https://t.co/drlDrxkYtp
🤗 Open Weights: https://t.co/T13Y8i7SDM
1/n
I think this is actually great news. I was afraid that they would be nerfing Mac Studio artificially in order to give Pro top specs. I’m glad that’s no longer a possibility.
Human-level general intelligence is achieved when an AI system can approach a new task and figure it out, without human intervention, *with the same learning efficiency as humans*.
If every new task requires human intervention, it's not general. If every new task requires brute-forcing, it's not human-level.