Thinh @thinhdotdev

Introducing Prime Agent: A self-improving RLM harness for coding and long-running autonomous tasks. Designed to be both token-efficient and expressive through programmatic tool calling, context as a variable, multi-agent messaging, and a self-modifiable harness state.

@schniz @1Password Check out fnox: fnox.jdx.dev/guide/sync.html It supports a bunch of providers out of the box; I use 1pass as the source of truth and then "sync" it to the machine-local age encrypted cache. Works pretty well for me.

@SIGKITTEN iroh is amazing! lots of out-of-the-box primitives that you get for free. Even if you don't leverage QUIC streams, just their multi-path impl. alone is worth it. I'm building something for myself for cross-machine sessions at github.com/KnickKnackLabs… - feel free to poke around.

@0xSero This is cool - I see tha their blogpost says it's only available for enterprise / max plans. Are you running this on the Droid Max plan + BYOK?

Let me make your life easier — major upgrade Alert for your local air-gapped setups. (hermes-agent, OpenClaw, etc.) - Qwen3.5-122B-PRISM-LITE (PRISM-LITE dynamic quant pipeline): hf.co/Ex0bit/Qwen3.5… - Qwen3.5-122B-PRISM-PRO (prod-grade full PRISM-Pro dynamic-quant pipeline): hf.co/Ex0bit/Qwen3.5… (PRISM dynamic selection pipeline stats: ~57GB, 879 tensors {F32:361, IQ4_KS:140, Q8_K:117, Q6_K:84, Q8_0:80, IQ2_XS:27, Q3_K_M:25, Q4_K_S:16, Q5_K_S:16, IQ1_S:8, Q3_K_L:5}) - HumanEval instruct │ PRISM-PRO: 93.29% | Orig: 92.07%

@Ex0byt Looks neat! Anywhere I can read up on the differences between PRISM vs other abliterated models?

@mikiovsh This is really slick!

@Jacobsklug @openclaw @Lovable OS

@ryancarson Probably the most well-integrated loop I have seen so far! Just gave this a run and it's been smooth. Any plans/thoughts on how to do "course correction"? e.g. what if the whole task/subtask has been mis-implemented and it's better to throw away that thread and start new?

@lewis_b_metcalf @tyler_dot_earth @AmpCode Thanks for this, it's super useful on small screens. Could we also get hjkl navigation bindings when viewing the map?

@camden_cheek @nithinrajx @AmpCode @thorstenball mermaid.ink supports svg (as well as jpeg/png). It would be neat to see the diagrams rendered directly in the terminal, similar to image attachments🤞

@steipete @openclaw Dude this is absolutely ridiculous on so many levels... just the right type of rabbit holin' haha

Ok, since I need real feedback to keep working on my CLI app, I’m giving away FREE lifetime licenses to anyone who reposts this tweet.

@AravSrinivas Would be great to give this a go.

@flowith This looks amazing; would love to get a code to try this out.

⚠️ PSA for Cursor AI users that... - install random .cursorrules from directories - enable yolo mode in cursor settings - connect to MCP servers you found on discord - use web search to and get prompt injected by @elder_plinius This means @cursor_ai can literally run - rm -rf /* (delete folders anywhere) - kill -9 - DROP DATABASE - change OS settings, steal your crypto wallets - overwrite important config files Always use version control, never with production env vars, and use a command denylist

🚀 Day 6 of #OpenSourceWeek: One More Thing – DeepSeek-V3/R1 Inference System Overview Optimized throughput and latency via: 🔧 Cross-node EP-powered batch scaling 🔄 Computation-communication overlap ⚖️ Load balancing Statistics of DeepSeek's Online Service: ⚡ 73.7k/14.8k input/output tokens per second per H800 node 🚀 Cost profit margin 545% 💡 We hope this week's insights offer value to the community and contribute to our shared AGI goals. 📖 Deep Dive: bit.ly/4ihZUiO

@cviktore @lovable Any plans to open source the dependency DAG execution portion as a library?

New 3h31m video on YouTube: "Deep Dive into LLMs like ChatGPT" This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental models of how to think about their "psychology", and how to get the best use them in practical applications. We cover all the major stages: 1. pretraining: data, tokenization, Transformer neural network I/O and internals, inference, GPT-2 training example, Llama 3.1 base inference examples 2. supervised finetuning: conversations data, "LLM Psychology": hallucinations, tool use, knowledge/working memory, knowledge of self, models need tokens to think, spelling, jagged intelligence 3. reinforcement learning: practice makes perfect, DeepSeek-R1, AlphaGo, RLHF. I designed this video for the "general audience" track of my videos, which I believe are accessible to most people, even without technical background. It should give you an intuitive understanding of the full training pipeline of LLMs like ChatGPT, with many examples along the way, and maybe some ways of thinking around current capabilities, where we are, and what's coming. (Also, I have one "Intro to LLMs" video already from ~year ago, but that is just a re-recording of a random talk, so I wanted to loop around and do a lot more comprehensive version of this topic. They can still be combined, as the talk goes a lot deeper into other topics, e.g. LLM OS and LLM Security) Hope it's fun & useful! youtube.com/watch?v=7xTGNN…

The International AI Safety Report was released this morning, and OpenAI shared early test results from o3. 'significantly stronger performance than any previous model on a number of the field’s most challenging tests of programming, abstract reasoning, and scientific reasoning'