Olivia Scribbles @PuzzleChall

𝗦𝗶𝗹𝗶𝗰𝗼𝗻 𝗩𝗮𝗹𝗹𝗲𝘆 𝘄𝗮𝘀 𝘄𝗮𝘆 𝗮𝗵𝗲𝗮𝗱 𝗼𝗳 𝘁𝗶𝗺𝗲. What started as a joke in Silicon Valley feels surprisingly real in 2026, because one of the most interesting things about AI is not what it helps us do, but what it helps us avoid. We spent years talking about AI as a tool for automating tasks, writing code faster, summarizing meetings, and boosting productivity. But a surprising use case is emerging. People are increasingly using AI to draft difficult messages, navigate awkward conversations, soften conflict, and sometimes avoid interaction altogether. That says something deeper about us. 𝗪𝗲 𝗯𝘂𝗶𝗹𝘁 𝗔𝗜 𝘁𝗼 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲 𝘄𝗼𝗿𝗸. 𝗦𝗼𝗺𝗲 𝗼𝗳 𝘂𝘀 𝗮𝗿𝗲 𝘂𝘀𝗶𝗻𝗴 𝗶𝘁 𝘁𝗼 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻. That may make life more efficient, but it also raises a strange question: if AI keeps getting better at handling communication, what happens to the human skills we stop practicing? What human interaction should never be outsourced to AI? #Humics #AgenticAI #DigitalBehavior #FutureOfWork

THIS IS HOW HERMES AGENT FINDS THE RIGHT FILE IN SECONDS WHEN YOUR VAULT IS PROPERLY STRUCTURED. Most people blame the agent when it cannot find anything. The agent is not the problem. The vault is. A flat pile of 2000 notes with random names forces Hermes to search blind every time. A structured vault does the opposite: - Consistent naming - Clear folders - Properties it can filter on - Maps of Content it can navigate Give it structure and retrieval stops being a search. It becomes a lookup. Same agent. Same notes. Completely different speed. Bookmark this. Follow @cyrilXBT

CyrilXBT @cyrilXBT

2 days ago

x.com/i/article/2072…

12 46 381 375K 1K

Tim O'Reilly @timoreilly used to do these posts about the developer ecosystem. He had a famous book publishing company that published books for programmers (they were the ones with the cute animals on the cover). Every year, he would publish a list of developer platforms and which ones were more popular with developers than others. The fact that Nvidia is going with Rust would, I assume, get him to re-rank the whole frickin' list.

Omar Kilani @omarkilani

22 hours ago

The LPU team at @nvidia is building the future of low latency inference. We're going all in on Rust. If you're a cracked Rust engineer and love making hardware do things, we want you. nvidia.wd5.myworkdayjobs.com/NVIDIAExternal…

11 26 249 25K 124

Anytime I see a post that claims to be 10 times faster, I take it much more seriously.

Jonathan Stephens @jonstephens85

20 hours ago

10x faster. That's how much Lightwheel compressed Geely's humanoid training cycle. That's from months to train a task down to weeks. Geely's self-developed humanoid robots are now running on the Auto production line, sorting and sequencing parts alongside human operators.

2 4 23 10K 14

My teenagers will never learn to drive unless it is on a race track.

I made 23 LLMs play speed chess against stockfish, on a real clock. api latency counts as thinking time - run out and you lose, even if you're winning on the board. basically: how smart is a model per second? the leaderboard looks very different to the usual ones.

THE BIGGEST PRODUCTIVITY UPGRADE THIS YEAR ISN'T A SMARTER AI. It's an AI that finally remembers. Think about how much time disappears every week just repeating yourself. You explain the project. The goals. The decisions you've already made. The mistakes you don't want repeated. Then the chat ends... and next time you start from zero again. That's backwards. The real breakthrough isn't another model with a higher benchmark score. It's giving your AI a memory that survives the conversation. Every project, every decision, every lesson compounds instead of disappearing into yesterday's chat history. At some point you're no longer managing context—the system is managing it for you. I think that's when AI stops feeling like software. It starts feeling like a teammate who's been working beside you for months.

kepo @kepochnik

4 days ago

x.com/i/article/2071…

10 11 63 33K 111

OCR hallucinations poison downstream workflows. We built research-driven safeguards that reduce hallucinations to near-zero in our benchmark. And our word-level bboxes and confidence scores let you check any potential hallucinations that slip through.

building and watching @ctnzr all things Nemotron and @NVIDIAAI if you didnt know he is the primary leader and driving force behind Nemotron youtube.com/watch?v=Oojrfd…

Bryan Catanzaro @ctnzr

2 days ago

Why are open technologies for AI so important? How and why is NVIDIA building Nemotron? What can we learn from China’s AI efforts? Great conversation with Matt Turck. youtu.be/Oojrfdl42LI

8 22 79 17K 20

Orca: The World is in Your Mind A general world foundation model that learns a unified latent space from multimodal world signals. It moves beyond next-token and next-frame prediction to Next-State-Prediction. This enables understanding, prediction, and action upon the world.

Everyone's arguing about which model today. Wrong question. My agent reads a coffee distributor's order emails. It's run 24/7 on a Raspberry Pi 5 since mid-June. Not once has the model been the thing that kept it alive. What keeps it alive is the part nobody posts about: — survives a restart without forgetting what it was doing — nothing ships until a second model reviews it — nothing ships until a human clicks yes — when it chokes at 3am it fails loud, not silent Swap in a cheaper brain, fine. A smart model on a weak harness is just a faster way to send the wrong order with total confidence. The edge was never the model. It's the plumbing. What's the least glamorous part of your stack that's actually holding the whole thing up?

this is actually insane. How to Build Your First AI Agent (Full Guide) save this. if i had this a year ago, i would’ve built my first AI agent in a single day instead of spending 2 weeks figuring everything out. everything is explained in a way that just makes sense. if you’re serious about AI, don’t skip this.

MIKE @mikenevermiss

3 days ago

x.com/i/article/2072…

21 26 219 347K 547

35 billion parameter model. fits on a DGX Spark / MacBook. runs at >200 tok/sec with vLLM/MTP. just as good as frontier models on a lot of tasks considered hard.

0xSero @0xSero

3 days ago

This is a real task I gave to Qwen3.6-35B running on my DGX Spark, take 3 videos stitch them, and speed the first by 4x. This is something i have been using LLMs for over the last few months, Qwen was flawless! It contributed to local studio fixing the alignment issue u see

17 10 301 55K 163

We spend so much time trying to make the right decision on-chain. Should I use this network? Is there a cheaper route? Will this transaction go through? Could I have saved more on fees? It’s something almost everyone deals with That’s why something Ken said about @WireNetwork got me thinking The interesting part is that WireNetwork has been working toward this for years. It’s not a new idea they’re chasing because AI is trending. If this is where on-chain transactions are heading, the experience of using blockchain could become much simpler than it is today. FT: @WireNetwork × @KenDicross

Cursor for iOS 👀 remote/hybrid workers are eating insanely good

Cursor @cursor_ai

5 days ago

Introducing Cursor for iOS. Build from anywhere by launching always-on cloud agents. Or remotely control agents running on your computer from the app. Composer 2.5 is 75% off in the app now through July 5.

955 1K 13K 6.3M 4K

Having ADHD in the AI era is a gift. You can do everything you imagine with agents.

Kekius Maximus @Kekius_Sage

6 days ago

People with ADHD often have a hyper curious mind. A 2026 study of 521 adults shows that ADHD traits like hyperactivity and impulsivity lead to higher curiosity and more exploration.

keyboard_arrow_left Previous keyboard_arrow_right Next

123 101 909 62K 157

That’s super cool! Maybe it’s time to talk about including StepFun’s models there

Cline @cline

5 days ago

We’ve been impressed with GLM-5.2 and so are introducing a $9.99/month subscription to give you 2-5x discounted access to it and other open weight models like DeepSeek, Kimi, MiniMax, Mimo, Qwen. Use it on Cline CLI & IDE with $1.99 special promo if sign up via: npm i -g cline

341 261 3K 1.0M 2K

The assumptions that break early-stage companies are usually the ones nobody put in writing. They live in the founder's head as givens, never shared, never tested. A bad assumption at the architecture level costs you six months before you find it.

Try typing "the founder who wrote one great paper in 2021, left to build something, and would actually reply to a cold email" into LinkedIn search. You get a dropdown for job title and company. That's it. Everything that actually matters gets ignored. Clodo (@clodo_ai) takes that whole sentence, searches 1.3 billion profiles across the live web, finds who fits, and gives you back verified emails and the signals it found.Kind of scary how well it works. Also really useful.

Sid Rajaram @sixtensor

5 days ago

Introducing CLODO, trusted by @mercor_ai, @usnews, @UseCorgi, the U.S. Intelligence Community, and more. Describe exactly who you’re looking for, and find them every single time. 1.2B profiles. We find, enrich, and reach out to them for you. It’s that easy. Backed by

136 287 1K 1.2M 801