John Walker retweeted
🚨BREAKING: Peptide Nano-Particles reduced plaque of the aorta by over 70% in pre-clinical trials!
John Walker retweeted
David Sinclair's lab is using AI to build a pill that reverses aging for $100.
Right now, their gene therapy costs roughly $10 million to manufacture. It requires a direct injection into the target organ.
That's not going to work for 8 billion people.
So Sinclair's team made a breakthrough. They found that the three age-reversal genes aren't the only path to resetting cells.
They discovered CHEMICALS that do the same thing.
In mice, they can now give an animal a liquid — not genes, not injections, a drink — and rejuvenate tissues in 4 weeks.
Sinclair says it's now normal for his students to casually report: "We just rejuvenated the ear. We just rejuvenated the skin. We just cured ALS (motorneuron disease) in these animals."
He calls his lab "Willy Wonka's chocolate factory" because the discoveries blow him away every week.
But he wants one molecule that does everything. So they used AI to screen 8 BILLION candidates. They're now down to three molecules that work. And they're using AI to try to combine all three into one.
The gene therapy could cost over $100,000 per treatment. Sinclair's goal: "What if it could be $100 instead? That's what I'm working for. I want to democratize this technology so anyone even in Kenya can take these medicines."
They should know within a year or two if the molecules work in mice.
The gene therapy is the proof of concept.
The pill is the endgame.
— @davidasinclair
John Walker retweeted
Jensen closed his GTC Berlin 2026 keynote with this Hermes Agent demo as the finale.
He introduced it by saying the way to build an agent is to ask an agent to build it for you, so he installed and set up a customized Hermes in one prompt.
Then he used Hermes, running on @nvidia RTX Spark, to create a simulated training environment for the @huggingface Microduck. Hermes planned an obstacle course with Nemotron and had GPT-6 Astra build it using FreeCAD, Omniverse libraries, Blender, Newton, and Isaac Lab. Hermes trained thousands of simulated ducks on that course until the validated policy was ready to run on the real robot in the physical world.
At Nous, we believe open source is a practical tool as much as a philosophical ideal. It gives users the ability to build powerful, complex workflows chaining numerous local+cloud models, devices, and tools themselves instead of waiting for a company to enable them.
Google's free 2.5-hour course on building agents that remember, survive restarts, and run for days:
0:00 → AI harness crash course
12:16 → the 4 types of agent memory
50:52 → Memory Bank: agents that rewrite what they know about you
1:23:05 → why "I'll check again in 10s" never runs, and how to fix it
2:00:26 → kill the server mid-task, the agent resumes where it stopped
2:20:04 → how to stop your agent from charging twice
Annie Wang (Google Cloud) builds it step by step, from harness basics to production safeguards. most agents never get past a demo. this is how they do.
the architecture behind systems like this is in the article below↓
John Walker retweeted
I'm blown away by Grok @bot.
This is the 14-bot team that handles my inbox, research, code, and design, and I only talk to 4 of them.
(bookmark so you don't lose this)
My executive staff of Pam, Tyrion, Gilfoyle, and Rachel takes every request, hands it to their teams, and brings the finished work back to me.
I'm publishing the full video soon. It will cover:
- How to reproduce this exact setup
- How Grok Bot works under the hood
- How to add skills and integrations
- How to run it with Claude Code, Codex, or any harness
This is still a work in progress, getting better every day.
Also, name your bots after characters you actually know, because talking to Dwight feels far more human than talking to agent-3.
Stay tuned!
My AI bots sign in to @Buildium with a login saved in a vault.
When a 2FA code shows up, they don't ask me. They ask the one bot that owns Buildium, and it passes the code along.
One owner per login. I stopped being my bots' help desk.
4
John Walker retweeted
Build reliable agents for long-running tasks.
Loop Engineering with @LangChain Deep Agents: add loops that check output, trigger runs on events, and improve the harness from traces across runs. Taught by @sydneyrunkle and @EAGar123.
Enroll for free: bit.ly/4kfDJxR
Spotted 11/16/2026 in St Peterburg, Florida- New Cybercab (With Driver) performing geofencing operations. Must be on there way soon!
14
New rule for my AI bots: when a search turns up an important file, copy it to where it belongs.
Never move it. Never delete the original. File a copy in the right folder, like that property's folder, so the next bot finds it in one look.
Search once, file it, find it forever.
5
John Walker retweeted
Make predictions on tabular data without training a new model for each dataset.
In Tabular Foundation Models for Data Science Agents, built with @prior_labs, you'll give an AI agent a tabular foundation model as a prediction tool.
Enroll for free: bit.ly/4sMtNOz
New rule when I update one of my AI bots: replace the old rule, don't stack a new one on top.
Then search every bot, routine and skill for older versions that conflict. Anything outdated gets proposed for a dated archive, never deleted without my OK.
One rule, one place.
4
John Walker retweeted
Update on the Grok bot blockade. It got wider. Grok Bot and a few others like Muse keep hitting walls.
Blocked or restricted:
Amazon, Home Depot, Target, eBay, Kohl's, Macy's, Neiman Marcus, Men's Wearhouse, Jos. A. Bank, Adidas, Pizza Hut, Cars dot com, United, Delta, Amtrak, Booking and Yelp.
Banks and brokers:
Chase, Schwab, Bank of America, Citi, Wells Fargo, Morgan Stanley, plus HSBC, Barclays and Lloyds in the UK.
Letting them in:
Walmart is building agent profiles and a gateway. QVC, HSN, Etsy and Shopify merchants are open. So are Muse partners Best Buy, Gap, Sephora, Ulta and Wayfair. GEICO even played along on insurance quotes.
Stores say security. I think it comes down to ads /profit and who controls the customer relationship.
The ones that treat personal agents like customers will get the volume.
The rest will wonder where the shoppers went.
// CS329A Self-Improving AI Agents (by Stanford) //
Lots of students have been asking me for good lectures on self-improving AI.
This one has good content and lots of good references.
I've started putting together my own notes and will share soon.
(bookmark it)
New rule for my AI bots: FYIs and status updates go into shared queue files, not messages.
A message to another bot is only for an ask it has to act on. Everything else gets written down where any bot can read it on its next run.
Fewer pings, less noise, nothing lost.
5
History will be written between today and tomorrow (Fingers Crossed) Watch the historical launch and booster catch tonight at 5.45PM (EST) and ship catch tomorrow around 5PM (EST) (Hopefully)
spacex.com/launches/starship…
40
New rule for my AI bots: any text going to two or more people waits 60 seconds before it sends.
That's my window to say "wait, wrong person" or "merge those." Bots are fast. Sometimes a minute of slow is the feature.
1
9
New flow for my AI bots in @Buildium: once a lease is signed and keyed in, the tenant pays the deposit in the portal.
When it posts, one bot tells my team and another pulls the unit from every listing site.
If the deposit bounces, the unit goes right back up.
2
20
















