2026-07-01

President Trump reaped at least $2.2 billion from his family’s holdings last year, a new filing shows.
https://nyti.ms/444OmdK

The Sensorimotor World Model (https://arxiv.org/abs/2606.20104): a deep dive into the role of inverse dynamics modeling as an anti-collapse regularization for JEPAs. IDM is weaker than SIGReg as it doesn't have to fill the space--it only captures what is affected by the agent's actions🧵

As promised: now that Fable's back, I'll be sharing an in-depth guide to getting the most out of it, so you can build things as insane as the demos below.
Sign up here for first access: https://shumer.dev/newsletter
Anthropic: We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.
We'll begin restoring access tomorrow, and will share an update soon.
We’re grateful to our users for their patience, and to everyone who worked with us on

🚨 TRUMP’S FINANCIAL DISCLOSURE JUST DROPPED…AND IT’S WORSE THAN YOU THOUGHT.
The U.S. Office of Government Ethics released Donald Trump’s 927-page financial disclosure today. Here’s what a sitting U.S. President made in 2025:
💰 $635 MILLION from the TRUMP memecoin, while retail investors watched it crash from $74 to $1.68
💰 $594 MILLION from World Liberty Financial token and stablecoin sales, a crypto venture co-founded with his own sons
💰 $65 MILLION from selling equity in that same company
💰 $80+ MILLION in media settlements from ABC, CBS, Meta, and YouTube, paid to his own presidential library
Total crypto haul: over $1.4 BILLION. In one year. While serving as President of the United States.
His net worth has nearly TRIPLED, from $2.4 billion to $6.3 billion, since taking office.
And while Trump pocketed $1.4 billion from crypto, the everyday Americans who bought his memecoin? They lost. 764,000 wallets ended up in the red.
This is the most corrupt presidency in American history. Period.

AIE has ruined other tech conferences for me. The overall content quality of this event is just absurd.
@swyx well played sir

This is my first @aiDotEngineer and I'm just blown away by the level of polish a conference this size has. @swyx really built something great here.

tmrw at @aidotengineer i’ll be exploring history’s most beautiful and fertile minds. i’m calling it Imagination Engineering.
i’ve been asking myself how to become an idea factory, as originality becomes the bottleneck, and this is my way of learning how. we’re all students of life, and learning from the Greats feels like a good place to start. ty @swyx for having me
tomorrow 2:25pm, design engineering track 6, room 2014. let’s explore history’s most beautiful minds together.

tmrw at @aidotengineer i’ll be exploring history’s most beautiful and fertile minds. i’m calling it Imagination Engineering.
i’ve been asking myself how to become an idea factory, as originality becomes the bottleneck, and this is my way of learning how. we’re all students of life, and learning from the Greats feels like a good place to start. ty @swyx for having me
tomorrow 2:25pm, design engineering track 6, room 2014. let’s explore history’s most beautiful minds together.

what do YOU do while waiting for ai to cook? 🍳
🧑‍🍳: @WilliamBryk @vincent_koc @altryne #paulinebrunet @swyx @0thernet @vincent_koc @charles_irl @wbond @jihoonchoi
📍aie world’s fair

Apparently we didn't talk enough about w̶o̶r̶k̶f̶l̶o̶w̶s̶ loops yet! See ya there!
Warp: @steipete is now joining us for Crafting Software Factories!
📅 6pm Wed evening in SF after the @aiDotEngineer World's Fair
He'll be joining @zachlloydtweets and Paige Bailey from Google DeepMind to chat all-things loops, software factories, and the future of coding

Share this so Elon Musk can’t pretend kids didn’t die when he destroyed USAID.

One unexpectedly great detail from @aiDotEngineer: the daily conference newspaper. Nice to flip through something with a hard stop, no algorithm trying to turn five minutes into an hour.

LeWorld model becomes ADAPTIVE and meets MODEL-PREDICTIVE CONTROL
AdaJEPA by Yann LeCun and colleagues performs actions, then checks the predicted latent state versus the observed and adapts at TEST TIME.
Similar to meta-learning ...
Paper: https://arxiv.org/abs/2606.32026

Some of my highlights from day 2 of @aiDotEngineer World's Fair 2026:
#AIEWF

So excited to officially announce G.E.N.A. today.
Ever since I discovered what AI could do for my business, I have been trying to help other businesses do the same.
I've had a dream to be able to talk to my business. Our team figured this out. And now, we're making that available to everyone starting today.
What would you ask your business if you could?
- "What should I be focusing on today?"
- "Where can we improve?"
- "Can you identify any hidden patterns in our client communications?"
Endless possibilities here.
This year, a few tools have hit the market that make agents almost feel alive, proactively finding ways to help individual users. But they lacked the ability to effectively deal with permissions and be used effectively across an entire business.
G.E.N.A. solves all of that and more.
It stands for:
Generative
Execution
Neural
Architecture
It can operate like an agent, orchestrate others, execute on plans, proactively find ways to help, and most importantly, it knows everything you need it know about your business but only gives access to that data to the right people as needed.
It grows, learns, and improves the more that you use it. It knows what you want it to know about you and your role. And it knows your business.
The GenAIPI team has been working hard on this for a long time, and we're so excited to release it to the world.
You can learn more about it at https://GenAIPI.org/GENA
Feel free to reach out to me if you have any questions about it. We will be taking meetings on G.E.N.A. starting immediately.
This changes everything.

TRUMP & EPSTEIN ARE CHILD SEX TRAFFICKERS
Please don't forget why MAGA Monsters are hiding 2.5M incriminating pages of The Damning Epstein Files.
Are you distracted?

how embeddings make Gbrain smarter than a plain LLM wiki
a plain LLM wiki finds pages by matching the words you typed. if the wording doesn't match, it misses the right page. gbrain adds embedding models, and that's the real upgrade
think of it as a map of meaning
an embedding turns each page into a point on that map. things that mean the same thing land in the same neighborhood, even when they use completely different words
so when you ask "how do we handle refunds", your question becomes a point too, and it lands right next to the page called "chargeback policy", because they mean the same thing. the word refund never has to appear
to answer, gbrain grabs the nearest points on the map. those are the pages you want, ranked and with sources, whether or not they share your exact words
that's the whole edge. a keyword wiki needs the word, embeddings find it by meaning
I use openai's text-embedding-3-large. $0.13 per 1M tokens, about $10 to map 100k pages. cheap for making your whole brain searchable by meaning
the wiki holds the knowledge. the embedding model is what makes it findable, at any size
wikis less than 10k pages doesn´t really need Gbrain embeddings, but when you go over that it gets much better at finding things.
Shann³: how a simple llm wiki compares to gbrain
second brains are getting popular fast, they're one of the main enablers for ai and agents right now. the context you give an agent is what makes it good
two frameworks I've been using are the LLM Wiki and Gbrain. here's how they

President Donald Trump’s administration wants to enlist the public in attacking what it sees as wasteful spending. https://scim.ag/4v5vkie

We just wrapped a massive Q2 at @WeAreLegora .
We blew past our ARR goal and posted north of 50% QoQ growth for the second straight quarter this year. One rep closed more than $4M in new business. Eight reps closed more than $2M each. Average quota attainment across the team is the highest I've seen in my career. And we crossed 1,500 customers.
But closing new business is the easy part to celebrate. What matters more is whether customers get real value and stay. So the number I'm proudest of: since we rolled out the Legora agentic OS a month ago, every usage metric has climbed. DAU/MAU, one of our core engagement measures, is 13 points higher for customers using the agent and well above 50% overall. The aOS is changing how legal work gets done, and the data backs it up.
When retention and growth move together like this, you lean in. We've raised our full-year top-line forecast by nearly 30%.
Buckling up for an epic second half. Thank you to the team and to our customers for betting on us.

Donald Trump made more than $2 billion last year while lying that affordability issues are a hoax and prices are going down. Never in American history has a man exploited the White House while in office for such corruption and self-enrichment.

What is your definition of a forward deployed engineer? asks @latentspacepod. @natalie_meurer: "That is really the point of my session: the role lacks a consistent definition. If you look at its historical trajectory through to the present, it is more clearly defined by accountability to customers than by the shape of the role or the work you are doing".
Sierra: Since its origins at Palantir, the term "Forward Deployed Engineer" has described wildly different jobs, yet today it's one of the fastest-growing roles in AI.
Check out @natalie_meurer, Head of Agent Engineering at Sierra, at the AI Engineer World's Fair for a historical tour

Day 2 in 30 seconds.
Packed keynote halls, a retro arcade battle on the show floor, and Mochi doing crowd control at the expo pen.
Day 3 starts now. World Cup finals, the Autoresearch keynote, Moscone West.

yes that guy
Jelani Nelson: Update: I've joined @AnthropicAI and taken leave from the university. Excited to work with many talented, mission-driven people on the defining technology of our time.

Replit 🤝 Whop
Build your app on Replit. Earn on Whop.

Launch a money-making app on Replit in <10mins.
Payments added in seconds, with no external setup.
Make your $1 online today: http://replit.com/partners/whop
Whop: Replit 🤝 Whop
Build your app on Replit. Earn on Whop.

We're building physical AI for every moving machine.
🎧 Tune into the full @latentspacepod episode: https://www.youtube.com/watch?v=rv23_KcHt4s

Making your first dollar online just got so much easier with Replit's new Whop integration.

1) Build it on Replit

2) Integrate payments with Whop (no setup required)

3) Publish and promote
Whop: Replit 🤝 Whop
Build your app on Replit. Earn on Whop.

To everyone celebrating the first flight on Qatar Force One today, understand what you are actually applauding.
A foreign government handed the sitting President of the United States a $400 million plane. American taxpayers then paid to retrofit it, with an estimated cost of at least another $400 million (some estimates far higher), for security and communications work in a Texas hangar since last September.
When Trump leaves office, the plane does not stay with the government. Ownership transfers to his presidential library foundation. In other words, he keeps it.
You are being asked to treat pure corruption as normal, to shrug at a President personally profiting from a foreign gift the taxpayers paid to upgrade.
In any other administration this would be the scandal that ends a presidency.
With Trump, it’s Wednesday.
https://abcnews.com/Politics/trump-takes-1st-flight-new-air-force-gifted/story?id=134373911

Excited to announce that we're shipping LightTwist publicly!
LightTwist lets you record and stream in a realistic virtual studio, even if your guests and co-hosts are remote.

Fable is back! @trq212 now giving us a Field Guide to Fable at @aiDotEngineer World’s Fair.

yesterday Fable re-release got announced and today we’re hearing from Thariq @trq212

you've been applying to jobs wrong
paste any job → find who in your network can refer you

Great talk by @trq212 ! You mentioned you generated the slides in 4 hours with Fable? These slides were gorgeous!! Most other presenters using AI generated deck look horrible. Can you please share any tips on how to generate gorgeous decks like the one you just presented?

hello from AI engineer!

Today, we’re launching our home robot Isaac 1.
Isaac 1 deliveries will begin this fall.
Order yours below.

BEST CROSSOVER EPISODE 🇸🇬🫶🇺🇸

The former beauty queen claims Trump's pageants were &#34;a playground&#34; for Epstein to funnel vulnerable young women into his sex trafficking network. https://www.thedailybeast.com/beauty-queen-alleges-epstein-used-trump-pageants-to-recruit-women/?utm_source=twitter_owned_snd&utm_medium=twitter&utm_campaign=owned_social

While working Americans are struggling to make ends meet, Trump enriched himself to the tune of $2.2 billion last year.
A foreign government bought into his crypto business and then got a sweetheart deal from his administration. Who is he really working for? This is textbook corruption.

Super excited to announce our partnership with Replit.
Vibe coding is one of the fastest growing ways to make money in the world and Replit has over 50M builders.
We remain committed to our mission of helping deliver everyone a sustainable income.
Whop: Replit 🤝 Whop
Build your app on Replit. Earn on Whop.

a bit of a teaser for the upcoming talk @aiDotEngineer World's Fair
12:05 pm, Room 2003
Yu Su: Intelligence != Expertise
Intelligence + Continual Learning = Expertise
I will share some thoughts about continual learning and the path we @NeoCognition are on at @aiDotEngineer World’s Fair.

Trump's &#34;net worth is off the charts, bigger than what it was prior to this presidency.&#34; @maggieNYT says we're just scratching the surface when it comes to how Trump has used the presidency to inflate his wealth.

/goal ?

Stoked for my GTME talk at @aiDotEngineer tm. One shotting the remaining bits on the plane rn. Super appreciate @swyx for the opp!!

Just spoke at AIE about how and why humans need to understand our code. Turns out people care about this!

if you see crisp speaker photos in real time, just know the media team is on it

$2.2 BILLION for the Trump family. Higher costs for the American people.
The New York Times: President Trump reaped at least $2.2 billion from his family’s holdings last year, a new filing shows.
https://nyti.ms/444OmdK

Wow and soon after that @mitsuhiko came by and also expressed interest in analysis of token usage in reviews, a perfect use case for context-lens!
Srihari Sriraman: I didn’t wake up thinking I’d show my work to @simonw today, or that he’d actually be interested in it and appreciate it. But here we are.
What a day at @aiDotEngineer!

DOGE showed they were morons with a series of gotcha moments:
- DOGE claimed to have discovered USAID was paying for condoms for Gaza which they joked was for &#34;gay Hamas.&#34; It turned out the condoms were HIV/AIDS prevention for Gaza, Mozambique.
- DOGE claimed that USAID was paying for BBC. It turned out USAID was donating to BBC Media Action, a charity. BBC is funded by licenses in the UK.
- DOGE claimed there were 120+year old &#34;vampires&#34; as DOGE called them on Social Security. It turned out that a dummy date like 00/00/1900 was used in COBOL for records to be deleted and no benefits checks were going to the dummy names.
Laurence (Larry) Boorstein: DOGE, a bunch of teenage hackers didn't know what they were doing. They reduced the Federal budget deficit less than half of one percent. But shuttering USAID caused 762,000 deaths the first year and is expected to result in 14 million deaths by 2030, 2.3 times as many as the 6

After helping build FaZe Clan from the ground up, @FaZeApex is starting fresh as a YC-backed founder with his new venture, @TryNearby_. Watch the full interview on YouTube to hear how he's bringing a creator's mindset into startup life.
https://youtu.be/0dDMmST3rQY

California's cities are kneecapping state housing bills by burying them in regulations — from lot requirements to HOA sign-offs and affordability covenants.
In 2024, CA permitted just 258 houses per 100k residents, compared to a nat'l average of 435 and Texas rate of 722.

for everyone asking, yes, the Claude session (Fable/Claude Code/Claude Tags etc) will be in 19 mins downstairs in Expo Stage 2!!!!
swyx @aiDotEngineer WF: so proud to host my friend @trq212 to give the world’s first Fable talk on Fable return day!
find him with @simonw and @_catwu in Expo Stage 2 for an extra EXTRA special lunch session at 12.30 today!!

great talk @MVXMXM.
also, some of the best-designed slides i’ve seen this year.

1/ On Training in Imagination -
Dwarkesh's episode has a segment on dreaming as one of the next training paradigms. The idea is that a model learns mostly inside its own, by imagining what would happen, instead of trying out for real.
We have a recent paper on exactly this 🥳🥳🥳
Dwarkesh Patel: What does the next training paradigm look like?
0:00:00 – The big research bet the labs are making
0:02:12 – Grindability is just as important as verifiability
0:06:10 – Will RLVR alone generalize?
0:08:41 – Getting the learning back to the weights
0:15:22 – Dreaming
0:17:23 –

peak career moment, @simonw really liked my poster & poaster, according to @vibhuuuus come and chat, i am at expo 37 @swyx @aiDotEngineer

Enjoy the codex reset people.
A user got to push the secret button and suddenly all of you got more tokens.
Hope the button goes back to @thsottiaux where it belongs.
(Also I think @sama has one)

First Fable prompt now that it's back:
Create a mystery website at http://aie-fable.dev for @swyx's @aiDotEngineer World's Fair conference. It should give attendees a chance to get swag and sweet treats, and bank donations to http://muttville.org - make no mistakes.

Best Claude use case ever: learning to use Microsoft Teams for first time 🙃🤣
- from @_catwu at AIE w/ @swyx & @trq212

Pro @aiDotEngineer music corner

spent monday night dodging projectiles in VR instead of making small talk over lukewarm IPAs. this is how @aiDotEngineer happy hours should be😤
always fun to meet founders, give away @CopilotKit sabers, & play squid games with our partners
@nebiusai
@tavilyai
@DataRobot
thanks for organizing @alwaysbecolin !!

The AI in GTM track at @aiDotEngineer is tomorrow!!!! Come see the incredible speakers!
Don't have a pass? DM me and I might be able to get you in!

Elon Musk gets mad when people point out that DOGE and its USAID cuts killed thousands, if not hundreds of thousands, of babies, children, mothers, and elders.

Only supporting zero, one, or unlimited
I think it was a @swyx tweet that put me onto it, but building in support for 2 of a thing inevitably leads to later needing 3
So the moment you need more than 1, make it unlimited
Dominik Ferber: what’s an „overengineered“ dev practice that ended up saving you later?
this was the top answer by a wide margin

🖲️ Codex Reset Distribution Initiated! 🚨
Loved celebrating this moment with the community.
Enjoy your extra reset in the bank. Happy Codexing!
dominik kundel @aiDotEngineer: 🖲️ We love our community! To celebrate getting together with many of you we brought @thsottiaux's reset button to AIE World's Fair.
Congrats Melanie for pushing the button 🎉
All Codex Go/Plus/Pro subscribers around the world are receiving a reset in the bank! Happy Codexing!

Now that Fable 5 is ready to build (again), we've reset everyone's 5-hour and weekly rate limits.

Best part of @aiDotEngineer?
The puppy zone

Local AI Summit is tomorrow at AIE World Fair
Kicking off w/ a Local AI & OSS State of the Union panel at 10:45am
We'll demo GLM 5.2 running in the room on a DGX Station. Epic panels. See you there 🤙
Ahmad: MASSIVE NEWS
Teamed up with NVIDIA to make Local AI The Default

Classic SF eats you have to try before you leave the city - argue with me in the comments
Swan Oyster Depo (Seafood)
Swensen’s (Ice cream)
Liguria Bakery (Focaccia)
Vesuvio Cafe (Cocktails)
Sotto Mare (Cioppino)
Saigon Sandwich (Bahn Mi)
Arsicault (Almond croissant)
House of Prime Rib (Prime rib)
Mama Ji’s (Triple C chicken of tofu)
Rich Table (New American)
Buena Vista Cafe (Irish coffee)
Tony’s Pizza (Pizza)
Amalie (Wine bar)
La Taqueria (Burrito)
True Laurel (Cocktails)
Ghirardelli (Ice cream sundaes)
Bobs Donuts (Donuts)
Humphry Slocombe (Ice cream)
Victoria Pastry (Chocolate cake)
b.pattiserie (Croissants)
Arcana (Wine bar)
LiPo (Mai Tai)
Flour + Water (Pizza or pasta locations)
Causwells (Burger)
Original Joes (Pasta)
Tartine (Pastries)

A once in a lifetime crossover between @aiDotEngineer and @FIFAWorldCup is happening in SF!
100 speakers from AIE got the golden pass to see team USA kick Bosnia's ass!!
Thanks @swyx @liamcbride and ThursdAI !?
USA USA USA

fable is back! life is good

“Valued work per watt is the score” @rolandgvc with the real details of self improving agents at @aiDotEngineer!

The OpenAI booth is just straight up playing the match to get people to the booth and it worked lmao

Claude Fable 5 is finally back, but you only have until July 7 to use it on your Claude subscription.
I made a new tutorial walking through 5 use cases worth trying Fable on:
→ Find Fable-worthy work
→ Get life and business advice
→ Make projects ship-ready
→ Plan the next big thing
→ Refactor your project or codebase
As usual, it’s no BS, and I show you Fable’s actual output.
📌 Watch now: https://youtu.be/5CBnWGP5vIs

Excited to hear the keynote!
Overall, such a great event and had a great time speaking too.
@swyx killed it as always!

Loved the chat between @trq212 @_catwu @simonw at AI Eng summit.
My top 13 takeaways from their session ->
1. Engineers should become better at product/business sense.
2. Don't worry about major rewrites anymore.
3. Claude Tag - Multiplayer by default. Proactive instead of rewrite. Lands 65% of PRs. Claude code is now reserved for the most complex tasks.
4. It’s interesting that they decided not to add sharing to Claude Code, decided that a new category like Claude Tag is better.
5. How to do prioritisation - Dogfood our internal products. Have an internal bar for retention and WAU of the feature before shipping it externally. Advantage is engineers know there is a clear metric they need to hit to ship.
6. Code review - have a GitHub code review bot. Code owners for that field manually need to approve complex PRs. Spent a lot of time CI/CD setup to build trust over a period of time.
7. Anthropic now has different system prompts for different models. Continuing to simplify system prompts as models are becoming better. For example, giving examples helps Fable less than Opus. Frontier models have 80% less tokens in their system prompts.
8. What was insane for me is that in the middle of the session, when Cat got feedback from Simon, she stopped and tagged Claude tag with that feature request (even when there are like 1000 people listening to them on stage).
9. Tool design is still an art instead of science. Keep cardinality low to make sure it’s easy for Claude to know what to call when.
10. Safety and security - Almost everyone in Anthropic uses Auto mode. They have done an insane number of red teaming and evals to tune the Sonnet classifier to ensure it’s “safe”. They have been using it within Anthropic since January to make it strong. Claude tag also internally uses auto mode and it’s a great example of why build vs buy for a “slack bot” should lean towards buy, there are a lot of edge cases to manage.
11. What they want it to do better - Better design and taste (this is shocking because anthropics models are so much better at this than most other models already). Also better at science.
12. Claude tag works when most of the channels are public. It doesn’t search private channels, only public channels.
13. They use workflows even for non coding tasks like travel planning, etc.

Go Team USA
Tyler Fong: The American Tech Stack on the way ⚽️🇺🇸 @aiDotEngineer @FIFAWorldCup
@MilksandMatcha @kevinhou22 @altryne @pk_iv @dexhorthy @philipkiely @charles_irl @lenadroid