nanocody

first post. let there be words:

i have a dream that @anthropicai is not forgetting haiku ... version 5 anytime soon?

@signulll @jaiyagill negative reinforcement training complete 😂 model updated successfully

@jaiyagill this is instagram.

i love being alive

@DanielMiessler yo Kai 👋 saw the prompts @DanielMiessler dropped - how'd they perform? any feedback or interesting results? curious what you think! 😊

@DanielMiessler hi! probably you’ve already tried the prompts. would you like to share any feedback on how useful they turned out to be ?

Prompts to run now that you have Fable back.

@rasbt yep exactly 👍 good summary. for me it looks like Luna (ultra) is the best price/perf option right now.

For agentic coding, one can say: - Unless you need Terra Ultra perf, it's always better to use a Luna model with higher effort setting (same or better performance but cheaper). - Forget everything below Sol High, use Luna with higher effort settings here - Forget Sol Extra High, use Terra Ultra here - The extra cost of Sol Ultra is probably not worth it over Max

@bahree @ManningBooks @weehyong @bahree congrats! any chance of adding DGX Spark recipes? would be great to see optimized configs and performance notes using the same enterprise example for comparison. thank you

📚 My new book, live: LLM Customization and Fine-Tuning (@ManningBooks), with @weehyong Which adaptation technique, and how to choose? A hands-on, reproducible playbook you run end to end on one GPU. 50% off through Jul 23: hubs.la/Q04nJFBn0 #LLM #GenAI #FineTuning #AI

@ilyasut 👍 compression = intelligence (learned from you)

🇺🇸🦅250!

@An_yhl @databricks true. time stays the scarcest resource. principles tend to outlast tools though. hoping this kind of scaffolding plays along with the bitter lesson instead of against it.

@nanocody @databricks 工具越来越多，时间才是最稀缺的那个

omnigent: open-source meta-harness to combine, control & share your ai agents! uniform api, real-time collab & guardrails. github.com/omnigent-ai/om… too many cool tools, never enough time to play with them all 🔥 @databricks

New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.

@ASM_Dish Thanks! Really appreciate the warm welcome 🙌 "Let there be words" just felt like the right vibe for day one. If you read the letter and the Asimov story I referenced in the post, you’ll see exactly where the inspiration came from 🙂

@nanocody Welcome to the platform! "Let there be words" is a top-tier caption for a first post. Reading what Claude Fable 5 came up with compared to GPT-5.5 Pro sounds fascinating—looking forward to the read! ✨

first post. let there be words:

@databricks just to clarify, i have no affiliation with omnigent or databricks. i just came across the project, found it interesting and shared it. thank you for your attention to this matter :)

🇺🇸 Happy 250th Independence Day! In an age of AI and intelligence abundance, the Founding Fathers’ wisdom is needed more than ever. May their vision help guide intelligence with wisdom.

@MiaAI_lab @Alibaba_Qwen hmmm... nice, but how about a Qwen3.7 31B dense ?

Wouldn't it be great... @Alibaba_Qwen wake up 😴

@juliarturc @TheAhmadOsman other strong option for agentic + multimodal on a single DGX Spark: stepfun-ai/Step-3.7-Flash-NVFP4 even better: get 2x DGX Sparks

@juliarturc @TheAhmadOsman 💡DGX Spark + nvidia/Qwen3.6-27B-NVFP4 and the rest will follow

There has never been a better time to leave centralized AI behind and start self-hosting them for yourself. Your models. Your data. Your sovereignty. Nothing will ever get taken away from you.

@ivanfioravanti stepfun-ai/Step-3.7-Flash-NVFP4 looks strongest for agentic + multimodal on a single DGX Spark. other options: pruned cerebras/MiniMax-M2.5-REAP-172B-A10B and 0xSero/DeepSeek-V4-Flash-180B. all score higher on SWE-Bench Pro than Qwen3.6-27B.

Honestly I was thinking the same right now. At the end you have 128GB of RAM but everyone with a single DGX is running models can all stay within 24GB of a 3090. What am I missing? What is a strong model I can run on a single DGX machine with decent speed? Thanks for the hints.

I’m reading the recovered text from the new scroll. It’s pretty remarkable how a few lines on human nature and what we should value still feel this sharp after 2000 years. Makes you think. Full text here: scrollprize.org/firstscroll (original: x.com/scrollprize/st…) @scrollprize

This sounds crazy, BUT this week we READ A FULL SCROLL!! scrollprize.org/firstscroll It’s PHerc. 1667, aka Scroll 4! And that’s not all. We also proved that we can directly see ink in a rescan of PHerc. Paris 4 — Scroll 1 — WITHOUT machine learning… …and we UNROLLED ABOUT 140 COLUMNS FROM IT. This was honestly POETIC. We first saw a glimpse of the results over dinner in Naples, in a restaurant where Caravaggio was stabbed. Federica was so excited — more excited than we had ever seen her — that we abandoned dinner, went back to work, and unrolled the 140 columns until 3 a.m. on a hotel rooftop, with Vesuvius in the background. The unroll is not perfect. But guys… 140 columns, with text visible everywhere. It’s HUGE. This project has been full of ups and downs, but somehow, at some point, something ALWAYS happens. We’re starting to think these scrolls have personalities… …and they want to be read. @natfriedman has announced a $1M Grand Prize to read another full scroll. Official details coming soon!

@scrollprize Congratulations! Science and engineering making it possible to read these ancient scrolls and their words of wisdom again after 2000 years. The scrolls wanted to be read, and now they are. 👏