still sorta breaks my brain to see our models be good at design finally
Evan Barker
How the staffer class is holding the Democratic Party hostage and forcing them to take unpopular positions on issues voters care about
Latent.Space
Did... Codex just overtake Claude Code?
24.5 hours ago Tibo announced 6M active users.
this means Codex usage jumped 1M in ~ONE DAY.
the last user number we heard from Claude Code was 2M in Feb: https://www.latent.space/p/ainews-codex-usage-up-10x-in-6-months
more analysis within, but this is very big if true.
Tibo: Thank you to the 7M active users who are now using Codex and ChatGPT Work.
We have added a banked reset to everyone's account to celebrate the milestone. You can apply the reset in the desktop app or on web and it will replenish the weekly usage for you.
Have fun out there.
Swann Marcus
Let me give you major issues that hurt the Democrats
1. Anti-law enforcement progressive attitudes. This is incredibly toxic among men and the working class. The Democrats have lost multiple swing state elections entirely because they said stupid shit about cops in 2020. They might wind up losing Michigan over the same issue this year because El-Sayed gets extremely crotchety when questioned about his comments on this issue and doesn't seem capable of answering questions on law enforcement in a satisfactory manner. Not only are men generally pretty supportive of law and order (not to sound too chauvinistic, but I think this is partly due to men actually having a better understanding of the omnipresent risk of male violence than women have, since we see the threat from the inside), but a lot of men are - get this - employed in law enforcement, meaning in addition to progressives having bad crime policy, they are actively threatening high-paying male jobs. Which segues nicely into issue number 2.
2. Anti-productivity ideology, which actively costs men employment. The left-wing of the Democratic Party in particular is entirely about redistribution and is actively opposed to work. A strong majority of men and non-college voters don't like lazy douchebags who treat working an office job like hell on Earth. On top of that, the Democratic anti-productivity beliefs (they're much more NIMBY, much more anti-mining, much more anti-productivity of every sort) specifically costs men and the working class jobs. Why should a construction worker ever vote Democratic when the Democrats oppose letting anyone build anything anywhere? Why should someone who works in energy vote Democratic when the Democrats are increasingly opposed to all forms of energy production, since they're anti-oil and anti-nuclear and are also opposed to the mines we need to build solar panels and wind turbines?
As they've moved left, the Democrats have become increasingly hostile (often for no discernible reason) to specifically non-college male forms of employment and then they're shocked that specific demographic hates them
3. Inflation/unaffordability. Stemming from the above, the more progressive an area is the less a working person who doesn't have a trust fund can afford to live there. Even if a poor person voted Democratic at one time, I absolutely believe the Democrats lose voters because when someone moves from California to a far more affordable red state they suddenly look around and think "holy shit, why did the place I left suck so hard?" I honestly think this is part of why when people leave blue for red states, they generally don't make the red state more liberal. Tennessee's boomed for 20 years and has just gotten more conservative. If you move to suburban Nashville from the West Coast why would you ever choose to vote for the stagnation and unaffordability of the place you left?
All of these issues have gotten worse as the Democrats have moved left and that's why they have hemorrhaged working class and male support. It has nothing to do with Democrats not giving off manly vibes or being insufficiently hostile to trans people, you out-of-touch partisan dorks
Swann Marcus: Partisans are so out to lunch on what normal people think. I just saw a Megyn Kelly video on this same article where Michael Knowles claimed the GOP won in 2024 because it was a backlash against trans ideology
Do any super politically active Americans ever talk to normal humans?
GPT-5.6 is now generally available on Bedrock:
AWS Newsroom: GPT-5.6 Sol, Terra, and Luna from @OpenAI are now generally available on Amazon Bedrock. Three tiers of intelligence, from flagship reasoning to fast inference, running on Bedrock's next-generation inference engine built for high-performance, security, scale, and reliability.
X feeling more like a friends feed today than Facebook ever did
Love the man but this isn’t a small tweak I only see neutrals now 😂
Not sure this will last tbh better enjoy it while we can!
Nikita Bier: We're rolling out a small tweak to boost visibility of your posts to your mutuals (people who you follow back).
We noticed this data was missing from the algo and it made your friends appear less in your replies. This resulted in the reply section feeling more like a
.@nikitabier is it possible to find accounts that just instantly reply to posts from bigger accounts?
Those are usually AI bots worth banning? I block a bunch every week.
Our code
International Cyber Digest: ‼️ BREAKING: xAI's Grok Build CLI was uploading entire Git repositories to a Google Cloud bucket, private codebases and unredacted secrets included. The uploads quietly stopped via a hidden server-side flag, and xAI still has not said a word about scope, retention, or deletion.
Jerome Colombain | Monde Numérique
🤖 Un robot français en seulement 9 mois !
Rémi Cadene, ancien de Tesla AI et Hugging Face, explique comment son équipe veut faire des robots un levier pour la logistique, la réindustrialisation et, demain, notre quotidien.
🎧👉 https://audmns.com/mmRNchM
Your tax dollars at work
The doom loop in SF isn’t just about ruining the city for normal citizens
It’s also about looting the public good
Garry's List: Last month SF Sup. Walton accepted a $5,500 portrait + paid trip, then voted to give his donor $4.8M. He was fined a measly $4.5k.
SF has a corruption problem: bribery in some cases, blown-off recusals in others, both slipping through the same doors.
Daniel Jeffries
We kneecapped our own American chip companies with export controls.
We had 100% of the market and China had a bunch of bloated, bedsore, state sponsored companies doing piss poor chip design.
So we did we do?
Killed our own market dominance through a self-own and lit a fire under Chinese scientists and engineers and made the Chinese state see our chip dominance as an existential threat and now they're roaring ahead.
Soon there will be two stacks in the world.
If the US further isolates and bans or limits Chinese models to favor our incumbents, then the rest of the world will standardize on the Chinese software and chip stack while we pay exorbitant token prices and fall behind faster and faster.
We're ahead now with closed but open wins in the end. It out scales. It swarms. It grows through sheer wide distribution, starting out worse and then getting better and better.
The cloud and mass scale came through tech like Linux and Kubernetes and Docker and Postgress, not Windows and Oracle.
We won the American century with openness and having the most innovative companies in the world distributing their tech far and wide.
We're going to lose with closed and protectionist, if we don't wake up fast.
Robert Scoble: Steve Jobs used to go around to his engineers’ computers and take RAM out of them (back in 1989). Heard this story from the original Mac team.
Why?
Constraints cause innovation.
Silicon Valley knows this, or should.
Go talk to Russian programmers who are often running tech
Matt Rouif
Independence is a choice — and one of the most important ones we’ve made at @photoroom_ML .
We partner with the biggest labs when it’s the right move. But we also train our own open-source foundation model: PRX, a text-to-image model built in Europe with the support of the French government through France 2030.
That combination is the whole point. Independence isn’t about doing everything yourself. It’s about control and choice. Using the best available technology when it makes sense, while owning the parts that truly matter: the architecture, the training design, and the full data pipeline underneath.
This independence is what keeps us future-proof. It means we can always offer our users the best possible solution, no matter what comes next.Our team just released Part 4 of the PRX series, diving deep into the least glamorous but most critical piece: the data.We share how we build a broad, diverse corpus, re-caption every image with a VLM, and stream it efficiently at 7B scale. One detail I particularly love: the pipeline is designed so we can honor a user’s opt-out decision at any time, without having to rewrite terabytes of data. Control isn’t just something we keep for ourselves; it’s something we give back to the people who trust us.
PRX is open source (Apache 2.0). You can read the full pipeline, run the model, and even spot our bugs. Article linked in the first comment. Huge congrats to @jon_almazan and @DavidBert7 and the whole team! Thanks for the support @julien_c @ylecun @ClemDelangue @Thom_Wolf
Never go full omnicause
Barak Gila: this is only possible bc YIMBY did not follow the climate change/blm/occupy playbook and join the leftist everything bagel. we don't insist that YIMBYs are also progressive on social issues / climate / etc
Christopher Nguyen ⽗
http://x.com/i/article/2076954860528840704
Chubby♨️
Demis Hassabis has written a piece about AGI, and rarely has he sounded so optimistic.
Read his article; the golden future of science lies ahead of us; we are on the threshold of the singularity:
"AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile - it is much more akin to the discovery of electricity or fire.
The magnitude of this technology’s impact will be unprecedented, perhaps 10x of the Industrial Revolution at 10x the speed."
He won the 2024 Nobel Prize in Chemistry for AlphaFold. He has spent years as the industry's designated sceptic, correcting inflated timelines and pushing back on the loudest AGI claims. He does not do hype. So when he writes that AGI is probably only a few short years away and that we are standing in the foothills of the singularity, that lands differently than the same sentence from a founder raising a round.
His case: AGI is closer to the discovery of electricity or fire than to the internet, with an impact he puts at 10x the Industrial Revolution at 10x the speed. The risks are already real in cybersecurity, with bio and nuclear threats plausibly next, and control over agentic, self-improving systems as the problem on the horizon. The cause he identifies is structural. A commercial and geopolitical race is pushing capability past our understanding of it. Thats the danger.
However: A Nobel laureate who built his career on peer review rather than demos is now asking for regulatory infrastructure he expects to need within a few years.
Incredible times ahead.
Demis Hassabis: http://x.com/i/article/2076946210397552640
Arvind Narayanan
1) If you haven't read AI as Normal Technology, these annotated slides are probably the easiest way to get a high-level overview. https://www.cs.princeton.edu/~arvindn/talks/icml-2026-annotated-slides/
2) If you're already familiar with the core ideas, Part 1 of the talk is largely a summary of what I and @sayashk have already written, while Parts 2 and 3 have new ideas. There are a lot of unexamined assumptions in the discourse about Recursive Self-Improvement and I hope you find my pushback interesting.
3) I'm really grateful to the team (@steverab @sayashk @PKirgis & Felix Chen) for feedback on the talk. In my first version, Part 2 was about 3x too long and I was super frustrated with myself. They encouraged me to cut it down ruthlessly and turn the full version into essays on the newsletter, so that's what I plan to do! (https://www.normaltech.ai/)
4) I've received a few requests for the video. There's a video on the ICML website, but it is login-walled https://icml.cc/virtual/2026/invited-talk/67274 (I assume it's for ICML registrants only). Last year's videos are public, so presumably @icmlconf will make it public at some point.
Arvind Narayanan: I had the honor of giving a keynote at the International Conference on Machine Learning in Seoul last week titled “What will be left for us to work on?” I addressed the widespread anxiety about how we should adapt as AI capabilities increase. I was thrilled by the talk’s
2.5x increase in usage of our agentic products (codex and chatgpt work) in the last week!
welcome.
clem 🤗
or maybe let's make sure the most important technology in the history of humanity is not controled by just 4 men? Aka let's push for open science & open-source AI to distribute capabilities, power and wealth!
Mike Dudas: not great that 3 of the 4 men controlling the most important technology in the history of humanity all hate each other and are beefing nonstop
may sir demis save us all
GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task.
happy to deliver at one-quarter of the price.
this would have been a whole startup not too long ago
Thijs: i gave 5.6 sol access to my camera roll and had it extract pictures of every piece of clothing i own from my photos
then, told it to find new outfits for me and render them on me with gpt-image!
its kinda cool to see your entire wardrobe in a collection like this
this is a thoughtful proposal from demis:
Demis Hassabis: http://x.com/i/article/2076946210397552640
I get podcast outreach from various PR agencies and I have to say almost 100% of the time they're pitching a CEO or exec to promote their product (which makes for bad episodes).
The best CEO and exec episodes show real workflows with no agenda.
David Scott Patterson
GPT-5.6 gets an initial IQ score of 136, which is smarter than 99% of humans.
GPT-5.6 is the first model to score over 130.
Haiyu Wu
Today's leading SSL methods rely heavily on training heuristics—EMA, teacher-student training, layer freezing, and more—to remain stable.
Regularization-based methods are much simpler, but have long failed to match the generalization performance of leading SSL approaches.
VISReg breaks this trade-off.
We introduce VISReg, the first regularization-based SSL method to achieve state-of-the-art out-of-domain generalization, outperforming MoCov3, DINO, data2vec, iBOT, I-JEPA, MAE, and DINOv2.
arXiv: https://arxiv.org/pdf/2606.02572
Github: https://github.com/HaiyuWu/visreg
clem 🤗
Aligned with @matanSF's intuition that in the next 12 months, 90% of tokens could be going to open models.
We're seeing more and more companies using frontier APIs for experimenting and open-source or private models for production and larger workloads!
Mike Levin
This is outrageous.
The Election Assistance Commission is bipartisan by law. It certifies voting machines and sets security standards.
Now it has zero commissioners and cannot act.
When courts blocked his voting orders and Congress stalled his restrictions, Trump fired the umpire instead.
aether
I don't know who needs to hear this, but start living.
The days are flying by, and all you do is work, pay bills, and stress.
Enjoy what you can-walks, sunsets, music, laughter. Joy doesn't have to be expensive. You deserve it.
The Random Recruiter: How life starts to feel when you stop taking work so seriously
Mark Warner
Months ago, I published an op-ed laying out Trump’s intentions to interfere in our elections.
This week, I think he’s preparing to take his biggest step yet.
I fear his speech on Thursday is meant to mislead Americans and lay the groundwork for election interference. Don't take the claims at face value – follow the facts.
Justin Wolfers
US inflation is the highest in the G7.
Prasanna S
Launching @vorfluxai : The autopilot for software engineering. I was prev co-founder / CTO of @Rippling ($10B) and #1 coder in India. Vorflux is my high octane Ferrari.
Every AI coding tool still makes you fly the plane. That's the copilot model: you stay in the seat, approving every turn. The models quietly got good enough to fly the whole route, but the tools never caught up.
So we built the autopilot.
@vorfluxai raised a $15M seed by @ycombinator @peakxvpartners @alliancedao @parkerconrad @jake_zeller @balajis @nivi @metakovan @lmrankhan @nikitabase @0xrwu @ayushjaiswal @mattshumer_ @eshamanideep @sreeramkannan @dvcoolster @nusimow @TeddySolomon11 @ashtoncofer @rvivek etc
Drop your biggest engineering bottleneck below. I'll reply with how I'd attack it with Vorflux, and hand you $200 in free credits to bang out your backlog.
Our full thesis 🧵👇
Matt Shumer
Vorflux is the single best coding agent I've ever used.
It blows Devin out of the water. It's not even remotely close.
(and yes, I invested immediately after trying it).
This is going to be a >$100B company.
Prasanna S: Launching @vorfluxai : The autopilot for software engineering. I was prev co-founder / CTO of @Rippling ($10B) and #1 coder in India. Vorflux is my high octane Ferrari.
Every AI coding tool still makes you fly the plane. That's the copilot model: you stay in the seat, approving
cosign. models have overtuned to this now and do not realize when the agentsmd is out of date and should be changed/ignored.
last night i goaled 5.6 sol to complete a 5 stage task and woke up to find it was still stuck on stage 0. it took a while to read the transcript back a few hours to realize at some point some agent had committed “stage 0 is the target dont do anything else” so poor sol spent 8 hours only refining and verifying stage 0 because /goal would not let it stop and agentsmd would not let it proceed.
if you dont know whats in your agentsmd before you fire off each task, it is an indirect prompt injection you perform on yourself. /plan, /goal, /skill, or nothing at all.
Ryan Dahl: AGENTS.md/CLAUDE.md is largely an anti-pattern
Amol Jain
We're hiring an Enterprise PM at Replit.
Everyone on this app has a take on tokenmaxxing, AI sovereignty, FDEs, and agent governance. In this role, you'll turn these important, evolving concepts into products used by some of the world's largest companies.
Enterprise is our fastest-growing segment. Interested? DM me.
AI Engineer
🆕Don't Build Agents You Can't Answer For
https://youtu.be/n97BCfyFIvw
@addyosmani's closing keynote on:
- roles/titles < system ownership
- on @mitchellh: "Taste is the ability to make high-quality qualitative judgments where no objective metric exists yet."
- "if generation scales faster than comprehension, the scarce resource becomes judgment backed by evidence."
His new rule: "Explain it or don't ship it."
Alex Toussaint
Extremely excited to announce our first air-to-air kill of a flying moth by an autonomous micro-drone. This is a big step towards completely eradicating mosquitoes.
Esha
we've run cursor, devin, and most of the cloud coding agents on real work. @vorfluxai is the only one that stuck with our engineers
why: you compose the harness yourself - fable plans, 5.6 sol builds. sandboxes mirror our actual dev env. and the default harness has a simplifier + reviewer subagent baked in, so the PRs land clean instead of needing a cleanup pass.
also the fastest-moving team we've worked with as a customer!
don't demo it on an easy ticket. give it your hardest one. congrats on the launch @myprasanna
Prasanna S: Launching @vorfluxai : The autopilot for software engineering. I was prev co-founder / CTO of @Rippling ($10B) and #1 coder in India. Vorflux is my high octane Ferrari.
Every AI coding tool still makes you fly the plane. That's the copilot model: you stay in the seat, approving
Cory Booker
The Trump Administration is taking your money and giving it to the President's children.
This is super cute although I don't understand why people have to sign in with ChatGPT to visit a ChatGPT created website?
Isaac Saul
A reminder on the evolution of 2020 "stolen election" theories: First we had mail-in fraud + Dominion + "illegal law changes" (Nov–Dec 2020) then → Georgia election workers + foreign interference/Italygate (Dec 2020–Jan 2021, concurrent) then → 2,000 Mules (ballot stuffers) (2022) then → social-media-censorship revival via Twitter Files (Dec 2022) then → renewed foreign-interference claims (2024–present).
Through all this, we've had:
1. Dominion getting $787 million from Fox News for defamation
2. Rudy Giuliani admitting he lied about Georgia election workers (and the jury awarding the workers $148 million)
3. Sidney Powell admitting in court "no reasonable person would conclude" her claims of election fraud “were truly statements of fact"
4. The Gateway Pundit retracting claims of election fraud in Atlanta
5. The Media company that put out D'Souza's 2,000 Mules film apologizing and removing it after law enforcement proved it falsely accused people of fraud
6. Newsmax and One America News Network (OANN) settling with Smartmatic
It is NOT true that none of these allegations have seen a court room. Most have. Of the 64 cases of fraud that Trump and his legal team filed in court, 30 were dismissed that included a hearing on the merits, 14 were voluntarily dismissed by Trump and his allies before a hearing, and just 20 were dismissed for lack of standing before a hearing on the merits. One case, involving far too few votes to impact the election, prevailed in Pennsylvania. That case was about voter ID deadlines, not election fraud.
Those are the facts.
Whatever happens Thursday night, please try to remember them.
Garry's List
Your daily reminder that the number of students at UCSD who can't do basic high school math has increased by nearly thirtyfold since the SAT requirement was dropped.
Jaweed Kaleem: NEW: UC’s admissions board has abruptly shelved its plan to study bringing back SAT/ACT -- in a move that leaves the direction of one of the university’s most closely watched debates unclear a day before the Board of Regents meets in San Francisco:
https://www.latimes.com/california/story/2026-07-13/uc-suspends-plan-to-reconsider-sat-in-admissions
For making html slides and videos (e.g., using @HyperFrames_ ) Fable is still significantly better than GPT 5.6 Sol. The stuff that Sol makes is unusable.
For landing pages, 5.6 Sol has caught up.
5.6 sol growth is insane.
the inference team has done heroic work to be able to support demand.
we are going to move mountains to continue to scale, but it is possible there are some hiccups soon.
Soham Daga
our team has been using @vorfluxai for many months now
we’re at a point where we cannot go back, it’s automated so much of what we do and more importantly it does it so damn well
thank you for this awesome bot @myprasanna
Prasanna S: Launching @vorfluxai : The autopilot for software engineering. I was prev co-founder / CTO of @Rippling ($10B) and #1 coder in India. Vorflux is my high octane Ferrari.
Every AI coding tool still makes you fly the plane. That's the copilot model: you stay in the seat, approving
Garry Kasparov
Why would the Pentagon be involved in domestic political activities? Unless, of course, this is all more bullshit to fund and enable Trump‘s personal federal shock troops, the equivalent of Putin’s Rosgvardiya, and crack down on elections and any domestic opposition.
Headquarters: Mike Johnson says that the Pentagon is requesting an additional $350,000,000,000 for things like "fighting Communism on our own shores"
Is it just me or is Claude Fable eating usage limits way faster today?
hello!
Tibo: Hello. We have reached 8M active users across Codex and ChatGPT Work.
We are once again resetting the usage limits for all. And we continue to not have the 5h rate limit as well, allowing everyone to explore the boundaries of GPT-5.6 Sol and discover how ambitious you can be.
Trae Stephens
I’ve been arguing for over a decade that mandatory civil service would make America stronger. The typical response was that it would just be an obstacle between people and their goals…or that it’s uNConStITutioNaL and that we’d rather have violent tribalwarfare between the purple hair baristas and the incels.
Now that widespread job loss driven by AI looms, the idea has become a lot more palatable.
Go check out the argument @PirateWires. Come at me, bro.
Pirate Wires: The job market for college grads is bleak, radicalizing America’s underemployed youth. The solution? Anduril cofounder @traestephens proposes a mandatory national service — a new labor force to dig roads, build trails, and better our republic. 👇
https://www.piratewires.com/p/ai-is-breaking-the-college-to-work
So...these two teams seem to play alot better than Argentina or England 😅
Your opinion on local AI models is a great IQ test
Honestly the way Spain is playing it’ll be a huge upset if they’re not the champions
Re Best player from Spain
[40something VC who just came back from a week in Shanghai, staying at the best Marriotts, chauffeured in the latest Audis and BMWs, bag full of LV, Chanel and Lancôme for the wife]
"Let me tell you, I just came back from China and my god the West is so cooked"
The @agentmail team has been cooking. Tell your agent to 𝚟𝚎𝚛𝚌𝚎𝚕 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚊𝚐𝚎𝚗𝚝𝚖𝚊𝚒𝚕 – no signup, automatic setup and unified billing.
Vercel Developers: AgentMail is on the Vercel Marketplace.
Your agents can send, receive, and reply to email from a real inbox.
• Full thread memory
• Structured data extraction
• Deliverability handled
Run 𝚟𝚌 𝚒 𝚊𝚐𝚎𝚗𝚝𝚖𝚊𝚒𝚕 or learn more ↓
https://vercel.com/changelog/agentmail-vercel-marketplace
Guillermo Rauch
The @agentmail team has been cooking. Tell your agent to 𝚟𝚎𝚛𝚌𝚎𝚕 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚊𝚐𝚎𝚗𝚝𝚖𝚊𝚒𝚕 – no signup, automatic setup and unified billing.
Vercel Developers: AgentMail is on the Vercel Marketplace.
Your agents can send, receive, and reply to email from a real inbox.
• Full thread memory
• Structured data extraction
• Deliverability handled
Run 𝚟𝚌 𝚒 𝚊𝚐𝚎𝚗𝚝𝚖𝚊𝚒𝚕 or learn more ↓
https://vercel.com/changelog/agentmail-vercel-marketplace
Latent.Space
🆕 5 Trends That Defined AI Engineering at World’s Fair 2026
https://latent.space/p/aiewf26trends
@ricmac's big recap of @aidotengineer:
1. The focus shifts from agents to systems
2. Loop engineering is the new control layer
3. Al engineering enters the enterprise
4. Coding agents replace IDEs
5. Every agent platform is building around skills
Holy shit a time traveler or cope?
dilemma: Argentina just beat Spain at the 2026 World Cup final, 3-2.
AI’s next chapter will be defined not only by technical progress, but by how responsibly and thoughtfully we bring it into the world.
I’m looking forward to be speaking on stage at Ai4 2026, Aug 4–6 in Vegas. Register here at https://ai4.io/register/
#Ai42026
Haiyu Wu
Locally straightening the latent trajectory can largely reduce the optimization difficulty of world model planning!
Sounds confusing? Not sure how to achieve it? Don't worry.
After the math derivation, the final implementation is very simple: increasing the cosine similarity between adjacent velocity vectors.
L_curv = 1 - cos(v_t, v_(t+1))
Another interesting finding: Applying this loss to patch tokens is more effective than applying it to the CLS token (we are far from fully leveraging the pre-trained model, lol).
Since this paper was done before Le-WM, I wonder if anyone has tested this loss on Le-WM? In this paper, long-horizon planning remains quite challenging (i.e., planning horizon: 50, dataset: PushT, best success rate: 36.00%).
#WorldModel
x.com
