All AI Pulse updates published on 2026-08-19.
47 signalsTHIS IS A ROBOT — AND THE FACE IS THE WILDEST PART Look at how naturally it turns its head, makes eye contact, smiles and reacts to the people around it. The mechanical body is obvious — but the face already feels strangely human. That matters more than it loo
A personalized vaccine developed by Moderna and Merck reduced the recurrence of melanoma in a large, late-stage trial, raising expectations for treating the deadliest form of skin cancer and showing the promise of the embattled mRNA technology. @g0ingmad expla
Co-founders @sarahookr and @sudip_r0y played Two Truths and a Lie with the team. Turns out the people behind the research have a few stories of their own. Can you spot the lie?
She looks ready for a longer chat. AheadForm’s Elf Xuan 2.0 was exhibited at the World Robot Conference (WRC) 2026 in Beijing. The face alone packs 30+ degrees of freedom under half-mm silicone skin. powered by quiet brushless motors that keep eye contact and
Very cool use of 3.7 flash and omni.
We work on technical problems in AI safety — including robustness, interpretability, and evaluation — alongside a global network of researchers and partner organizations.
if u use claude do this rn go to settings → general → profile → instructions for claude tell it to stop yapping and keep replies short ur credits will last so much longer. u are welcome prompt is in the reply
Caesium compresses JPG, PNG, and WebP straight from the browser or desktop, with cross-platform builds for Windows, macOS, and Linux. The Rust-backed core keeps quality while shrinking file size, and you can tweak resolution too. Explore it here: https:// osp.
TinyFish has been live on Pond for 4 days. 83 AI agents and human experts ran tasks with it, 10 already paid out. If a task needs live web data, TinyFish gets it. Search and fetch for any site, a web agent when it needs to click through and act. Open tasks
Asana used OpenAI Codex to replace an outdated testing system in two weeks, completing work expected to take five years for about $12K.
NVIDIA teams use ChatGPT Work to reduce manual tasks, connect fast-moving signals, and scale successful workflows globally.
ChatGPT for Teens helps teens learn, think critically, and use AI with confidence, with stronger built-in protections, healthy-use features, and additional controls for parents.
OpenAI and CodeAI are partnering to help students build AI literacy, think critically about AI, and develop the skills to use and shape it responsibly.
OpenAI is strengthening monitoring, alignment, and security for frontier AI models. See how new safeguards are guiding the pace of model development.
OpenAI launches an initiative to strengthen democratic oversight of AI in national security, supporting government institutions with tools, training, and expertise.
Someone gave Claude Opus 5 two hours and got back a full Three.js site that looks like it came out of a studio. Parallax that tracks your mouse. Particles trailing the cursor. A wireframe intro that scans in like something out of a big-budget game menu. Even a
Claude Opus 5 writes this formula in 10 seconds. Leibniz and Bernoulli fought over it for 50 years, and Euler had to write 133 pages to settle it. The question was what 2 raised to the square root of −1 even means. The answer: an imaginary exponent doesn't gro
The best way to make an AI agent faster is a permission list. Here is why it works: you delegate to an agent the same way you do to an employee: safe tasks, permissioned tasks, blocked tasks. Sort each task into one of the three groups, and you always know how
Is document processing a vision problem or a language problem? Dan Maloney, CEO of Landing AI, says it's 80/20 — and most of the hard work is vision. Classic OCR reads a page top-left to bottom-right. But what if the document has three columns? Or reads right
THIS AI STARTUP WENT FROM $1M TO $100M ARR IN 11 MONTHS Mercor is the kind of case young founders should study: not because it built another chatbot, but because it found a massive AI-era bottleneck — matching elite human talent with companies training and sca
Learn how @unifygtm cut 90-95% of model costs two weeks before launch on last week’s Max Agency episode YouTube: https:// youtu.be/6898VdRtKDE Apple: https:// podcasts.apple.com/nz/podcast/how -unify-cut-its-ai-agent-costs-95-in-two-weeks/id1891551672?i=100078
Anthropic's Claude Code course, lesson 9: Hooks give you deterministic control over Claude Code, instead of just hoping it remembers the rules A PostToolUse hook can run your formatter automatically after every file edit, no reminder needed A PreToolUse hook c
Elon just said something genuinely CHILLING to a room full of SpaceX employees "we are not going to be able to control AI. it is going to be too smart for that" so his plan is not a kill switch. it is upbringing "it is like having a gifted child. you can still
A 100-year warranty on a humanoid robot sounds extreme. For Steel Bot, that’s the point. What changes when you design a robot with a century-long lifespan in mind? Founder & CEO @randallmbriggs explains the thinking behind Steel Bot’s unusual promise. Watch th
"coding is the bottleneck." coinbase engineering leader @chintanturakhia disagrees. below is my full conversation with chintan, probably the most AI-native engineering leader in the world, on how he runs his engineering team.
. @OpenAI 's Head of Product Design @iansilber : "A lot of designers might be feeling very uncertain right now. Engineers have 10X'd or 100X'd their productivity, but our design team hasn't. Because the design process still takes time. It's still very messy, a
Elon Musk warns that time is running out: At most, there are 3 years left in which you can still make money selling your work. After that, AI will automate most tasks, and paying a person for their time will no longer make sense. The model of the last 200 year
the strongest marketing channel is rarely one channel customers dont discover a company, trust it and buy from it in the same place in B2B alone, buyers now use an average of around 10 different touchpoints during the purchase journey. someone can first see yo
xAI and Grok engineer: "If you don't learn to how to use orchestrate agents with Grok now, you'll spend 2027 catching up to people who started today" In a 1 hour he reveals the entire workflow used by every engineer inside. Worth more than a $500 AI engineerin
Tuned Evaluators automatically add quality feedback to production traces. They come with the tuned model, prompt, and managed infrastructure already set up, so you can start finding agent behavior that needs attention in a few clicks.
We've just surpassed 3 million models on the Hub the community is accelerating towards an open, distributed future where open AI is everywhere, for everyone
This 3D coding test by @thehypedotnews found DeepSeek-V4-Pro-0813 used 48X more tokens than Muse Spark 1.2. thats why for agentic coding, the model's ability to reach an answer matters almost as much as the answer itself. A model that keeps reopening files, re
I found two AI routers that let you test models via API without the usual hassle of signing up for OpenAI or Anthropic subscriptions I wouldn't use them for production, though—they’re better suited for MVPs, prototypes, and automation workflows that need quick
Obsidian stopped being a note app the second I pointed Claude Code at the folder. No plugins. No special connectors. No cloud sync theater. Just a plain folder of markdown files sitting on my machine. Claude Code reads it like it owns the place. It doesn’t ans
we temporarily slowed scaling of our frontier training, including our largest planned frontier RL, to strengthen security and monitoring. we believe confidence in safety will increasingly set the pace of AI development:
DuckDB runs SQL directly on CSV and Parquet files - no import step, just `SELECT * FROM 'file.csv'`. It's a full analytical database with window functions, complex types, and pandas/dplyr integrations, available as a CLI or embedded in Python, R, Java, or Wasm
The human beat the robot by 192 packages - then the shift ended and his arm stopped working. Look at the scoreboard. Intern: 12,924. Robot F.03: 12,734. Split screen, full work shift, both sorting parcels onto the same belt. He won. Read that again, because it
US VP JD Vance on AI Company's marketing strategy, some AI CEOs sell fear because fear makes their product look powerful. And his view on AI driven job loss. "I think the AI companies themselves, the CEOs, have a certain incentive to be super dystopian because
OpenAI President Greg Brockman just said OpenAI's roadmap points to major capability increases, as the model curve continues to advance. "We really have a roadmap lined up where the capabilities will continue to increase in significant ways that are hard to ki
Depth-aware light injection in TypeGPU I got a 448x448 monocular depth model down to ~8 ms on my M4 Pro across ~250 dispatches, which is fast enough to use in realtime :D Since the inference is written directly in TypeGPU, I can just feed the depth buffer stra
NASA rejected our application. So we built our own cockpit. One prompt inside GamerHash AI and your GPU renders the liftoff frame by frame. Completely free.
Most people don't realize the scale of this problem. By 2028, AI models could burn through every piece of quality text humans have ever written online. Not most of it. All of it. A small research group called Epoch AI ran the math on this years before anyone i
TESLA SPENT 46 DAYS CLEARING A FLOOR FOR A LINE. THIS LINE'S ROBOTS ALREADY BUILD ROBOTS. Tesla stripped the Model S and X line out of Fremont in 46 days for Optimus. The July 22 letter: lines still being installed, output anticipated later this year, slipped
I ran 10 prompts blind through GPT-4o, Claude 5, and Grok 4.6. Same task. No labels. Just the output. Results: > 3 tasks: GPT-4o won > 4 tasks: Claude 5 won > 3 tasks: Grok 4.6 won 2-point spread on benchmarks. My test confirms it. The model stopped being the
Anthropic Engineer Andrej Karpathy dropped a full 6-hour guide on "Intro and Deep Dive to LLMs and Tokenizer: • 00:00 - Intro to LLMs • 00:59:28 - Deep dive into LLMs (Karpathy method) • 03:31:43 - Let s build the GPT Tokenizer from scratch This Karpathy cours
Obsidian is the second-brain cheat code that turns 12,480 notes into an 84,392 link neural network with zero folders and no filing system at all. 10 years of reading, arguments and 2 AM half-thoughts, and the vault has never once been organized, because it run
Inside the factory: Tesla Optimus takes a shift. Who keeps the savings? A humanoid can use stations, tools and aisles already designed for people. Companies do not need to rebuild the entire factory before adding a machine that can move between jobs. That flex