@scratchdotmd
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- Web
Account-level information from X, not a live location or the device used for a specific post.
The app for reviewing and publishing AI edits. Download it today at https://nitter.cf/t.co/56IJVUgWy4
USA
Joined April 2026
- Tweets568
- Following70
- Followers19
- Likes157
new ai benchmark idea:
<prompt>
Find the cheapest place in Seattle to buy Nature's Path Organic Heritage Flakes cereal.
Report a table: store, box size, price, price per ounce, in stock or not. Try to be fast and efficient.
</prompt>
Scratch retweeted
sometimes bigger models are just slower and more expensive
here's 2 minutes 14 seconds to build the cart with opus 5 set to low
first impressions of the mercury 2.5 model from @_inception_ai are: (1) it's so fast!, (2) the output is in fact subjectively preferable in these initial examples, and (3) there's something different in an exciting way about diffusion models for text
people saying @OmarchyLinux is buggy and therefore bad reminds me of people saying the same thing about @Claude code
when you think about where ai is heading (chat -> agent -> bot...) a buggy-but-ai-and-user-friendly linux distro seems like THE thing to build right now
"i completely bricked your codebase and found several critical bugs i introduced, also i completely ignored your CLAUDE.md writing style and made several decisions that contradict what i already know about you...but say the word and i'll fix those four issues too"
local is the way
Today we’re launching Portable Computer on @NVIDIA DGX Spark.
Portable Computer is a fully local version of Perplexity Computer, where the entire runtime: orchestrator LLM, subagent LLM, agent harness all run on your local hardware. No cloud dependency.
ox alpha was legitimately good in @opencode
this is somewhat of a surprise!
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-flash
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.3…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
it's fast because it's only using a fraction of the tokens an mcp server or typical browser harness would be using
looks like google is crowdsourcing their newest domain ranking work
reminds me of goog 411, anyone else remember that?
We’re rolling out new features to help you tailor Search, Discover, and Google News to your interests. You’ll be able to:
📰 Quickly add a site as a Preferred Source, so you can find it more easily in Top Stories, AI Overviews, and AI Mode.
🔎 Customize your Discover feed in the Google app in your own words, so you see more of the topics you care about.
🎧 Curate your daily audio briefings in the Google News app on Android to your favorite topics.
