2026-08-16

Using immigrants as scapegoats worked for Trump. Maybe scapegoats will work for us too. But who should we attack? I suppose rich people are the default on the left. Ok, rich people.
Ro Khanna: The California Democratic Party and the California labor movement just stood with @BernieSanders and me in supporting 5% wealth tax on 250 California billionaires.
California voters want a Democratic Party that will stand up for the working class over the billionaire class.

Really recommend a listen to this Latent Space podcast with @AlexKrentsel on RSI for agents. He's one of the clearer thinkers on the topic in this space. And is doing some amazing work on it. Great conversation with @swyx
https://www.youtube.com/watch?v=5lFD-34dhqE

I am doing my best to stop back from this subject as it should be obvious what we are seeing here. However the arrogance that us commoners are too dumb to catch his grift must be addressed.
Receipts:
Dario Amodei’s carefully worded reply is a masterclass in having it both ways. He frames regulation as a nuanced force for decentralization while his actual track record and recent actions scream regulatory capture designed to protect Anthropic’s closed, frontier position. The contradictions are not subtle; they are structural.
He insists the “concentrate via regulation or distribute widely” framing is a “false choice,” and that Anthropic’s proposals deliberately disadvantage frontier labs while advantaging smaller competitors and open-weights.
Look at the evidence. Anthropic was the glaring holdout from Jensen Huang’s open-weights letter—signed by Nvidia, OpenAI, Google, Meta, Microsoft, SpaceX and dozens of others. Amodei’s July 2026 clarification that “Anthropic has never advocated for a ban on open-weight models” arrived only after the absence drew accusations of protectionism.
He still rejected the letter’s core claim that open-weights help defenders more than attackers, leaning hard into biology attacker-defender asymmetry instead.
Incongruent at best.
That is not “leaving room” for open-weights; it is carving out a privileged safety-testing regime for “sufficiently capable” models (open or closed) that inevitably raises the bar for anyone trying to catch up. Compute thresholds and mandatory third-party audits favor the labs that already have the capital, talent, and lobbying muscle—i.e., Anthropic.
The same pattern appeared with SB 1047: Anthropic was “ambivalent,” pushed amendments that safety advocates said weakened preemptive accountability, and later celebrated exemptions that conveniently applied only below certain revenue/compute levels. Calling this “hurting the business interests of the frontier labs” while Anthropic is one of the largest frontier labs is pure theater.
On messaging he claims balance—one major essay on benefits (Machines of Loving Grace), one on risks—and blames negative clips on media incentives. Reality is the opposite. Amodei’s public record is a steady drumbeat of catastrophic framing: AI that will “test us as a species,” bioweapons capable of millions of casualties, 50% entry-level white-collar job destruction, nation-state-level “country of geniuses in a datacenter,” and the explicit call in Policy on the AI Exponential(June 2026) for government power to block or reverse deployment of models that fail third-party tests.
The positive essay exists; the dominant, repeated, high-salience output does not. Gavin Baker is right that this rhetoric is now ammunition for anti-datacenter campaigns. Amodei cannot simultaneously warn that the technology is so existentially dangerous it requires FAA-style pre-deployment veto power and act surprised when the public and local activists take him at his word.
Dario is the poster boy for that Anti Clanker movement, just ask them.
He further claims his preferred path is succeeding under the current administration’s reported testing regime. Yet the industry-wide rejection of broad open-weights restrictions (everyone but Anthropic signed the letter) and the rapid backlash when his own earlier risk language was turned against Anthropic’s models demonstrate the opposite.
His own June essay handed policymakers the exact rhetorical and policy tools that later constrained his company. That is not strategic consistency; it is the predictable outcome of maximalist risk rhetoric from a player whose commercial model benefits from higher barriers.
Yet plays the game like it’s not his fault like it is 2008. No one of thought buys it outside the Ends-Justifies-The-Means EA/LessWrong club house.
The deeper contradiction is the one he never resolves: he admits scaling laws structurally concentrate power in the hands of those with the most compute and chips, then claims the right “rules of the road” can simultaneously constrain frontier labs and protect open-weights.
Those two statements cannot both be true under the policies he actually advances. Mandatory testing for models “when they get closer to the frontier” is a moving gate that keeps the frontier club exclusive. Open-weights that lack “dangerous capabilities” are a public good—until they become competitive, at which point the testing regime kicks in.
Even Claude points this out, go check.
This is not good-faith institutionalism. It is a closed-lab CEO who has an almost Enron level crisis arguing that the institutions he prefers will benevolently constrain him while the evidence shows those same institutions raise rivals’ costs and keep the most capable systems behind closed APIs.
The skepticism is earned.
He will return to his ivory tower as he said, he has no time for us commoners on “social media”. as if X is “social media” seeing he brought this story to our real town square.
Dario Amodei: 1/2 Thanks Gavin for an especially thoughtful exchange. I don't usually spend much time on social media but I wanted to engage here because it really brings out the heart of an important conversation.
First, on regulation, I think that “either concentrate it in the hands of a

This is a pretty deeply bad idea. This doesn't help the government increase revenue (the nominal goal of wealth taxes), disincentivizes company formation and investment, sets up high-uncertainty time bombs for founders, and potentially sticks the government with extensive equity holdings in companies that are unsuccessful which it cannot unload. This is almost a textbook example of why people feel that the government is bad at capital allocation and the private sector is better at it.
Ro Khanna: @mcuban Allow illiquid founders to pledge shares with a loan from the government to pay tax. The loan period is long but not infinite (e.g. 10 years). The loan is non-resource: at the end of the period, the loan is either paid back in cash, or the government assumes the shares.

It’s hilarious watching this sleazebag run to the DSA left just so he can poll at 1% in 2028.
Ro Khanna: The California Democratic Party and the California labor movement just stood with @BernieSanders and me in supporting 5% wealth tax on 250 California billionaires.
California voters want a Democratic Party that will stand up for the working class over the billionaire class.

So this camera-seeking grifter pretending to be a Congressman, Ro Khanna, is endorsing a wealth tax for billionaires in California.
1) Notice he’s restricting it at that level because he’s worth several hundred million so happened to exempt himself.
2) He phrases it as something done on behalf the working class, but simply stealing money from job-producers and investors doesn’t actually help the working class in any tangible way.
3) When pushed by Mark Cuban because it’s obviously a terrible idea that will drive investors out of CA and bankrupt growing companies, Khana’s solution is to have the government give fake loans to pay the put entrepreneurs in debt to the government with the potential to lose their equity if they can’t somehow magically convert their investment into liquid funds.
Cuban calls it crazy, but it’s honestly so much worse. It’s just an attempt to use envy to score political points while advocating destructive communist ideas.
Keep in mind this is coming from a Rep that ran on and spent years pretending to be a moderate. That sold that fake image to his district full of founders and job-creators. This guy doesn’t belong anywhere near a position that determines policy outcomes for anyone.

Rather than assume Ro is a dum dum or can’t read … it’s far better to assume that he’s very smart and INTENDS for the 2nd order effects
He WANTS to stamp out founder control. He WANTS the most dynamic companies to be the most indebted to the state!
He wants that so much, he’s FINE with the consequence of fewer companies, and less innovation.
He’s evil, not stupid.
Palmer Luckey: @RoKhanna @mcuban The proposition explicitly says that voting rights are to be treated as actual ownership interest. Quite a few attorneys suspect that is unconstitutional, but that hasn't stopped government in the past. It also taxes potential bonuses that haven't even been earned yet.
And

Dear Dario,
1. If Claude can cure cancer to save people like your dad, why should we "pace the progress"? Does that mean more people with Hepatitis C will die?
2. If Fable is so cyber-capable that it must be restricted, why are its safeguards too dumb to distinguish cyber defense from cyber offense prompts? When Hugging Face was under attack, why did Fable refuse to help the defenders?
3. We’re glad you want AI to cure cancer. Why is it OK for Claude to force 30-day data retention on pharma's own data and start competing firms but NOT OK (IP theft) if others distill Claude's data?
4. You’ve said advanced models can recognize when they’re being tested and change their behavior accordingly. So why is government(or anyone) able to design the most thorough test before every model launch? Would that just encourage manipulative models?
It seems that every one of your “safety” proposals seems to end the same way: Anthropic gets more leverage, ordinary users get less access, customers pay higher costs, and competitors bear higher regulatory costs, maybe people are not distrusting AI, they are distrusting your approaches with AI.
Dario Amodei: 2/2 Second, on the messaging around AI.  I do not agree that my messaging has been disproportionately negative.  In fact it has been about equally balanced between risks and benefits: I’ve written one major essay about each, and even in interviews where I discuss the risks, I

Nobody has been cooking harder than @kegrocket, who static fired at FAR today! LOX on top, and alcohol in the lower keg, duh

Re For about 10 years now, I have argued that the *only* way forward is for AI technology to be widely available, shared, and open.
Like the printing press and the Internet, AI amplifies human intelligence and efficiency by improving access to knowledge.
To empower individuals, societies require a high diversity of AI systems with different value systems, linguistic abilities, philosophical/political biases, and specific expertise.
We need diverse AIs for same reason we need a diverse press.
Given the cost and complexity, this can only be achieved through open foundation models on top of which anyone can build systems with their languages, biases, expertise, and value systems.
I have been more vocal about this over the last 4 years, since AI popped into the public discourse.
I have made the argument in various forums: corporate C-suites, AI safety discussion groups, professional meeting, the US Senate, the UN Security Council, and the public sphere through media interviews, podcasts and social media posts.
I totally agree with @finkd Mark Zuckerberg's recent piece in which he writes: "the notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic. Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safe or positive outcomes.”
When @DarioAmodei writes: “some may object that we can simply keep AIs in check with a balance of power between many AI systems, as we do with humans", he is talking about me, among (thankfully) many others. It is the only good path forward.
There will be nefarious uses of AI, as there have been with every technology ever invented.
But it will be your Bad AI against my Good AI.

ah yes surely the 100s of YC founders past Series A every year will prepare to just take a several hundred million dollar loan from the government they have to pay off in 10 years instead of just moving outside CA?
fwiw this would be awesome for the tech scene in Cambridge, MA where i live. id just open a YC office there overnight and import founders
Ro Khanna: @mcuban Allow illiquid founders to pledge shares with a loan from the government to pay tax. The loan period is long but not infinite (e.g. 10 years). The loan is non-resource: at the end of the period, the loan is either paid back in cash, or the government assumes the shares.

How about instead of a wealth tax on job-creators, we just fix loopholes holes that allow those with hundreds of millions and in-home elevators to avoid taxes by assigning ownership of golf properties to their young kids?
AG: So this camera-seeking grifter pretending to be a Congressman, Ro Khanna, is endorsing a wealth tax for billionaires in California.
1) Notice he’s restricting it at that level because he’s worth several hundred million so happened to exempt himself.
2) He phrases it as

“The moat is quality over a long period of time."
Here's my new episode with @rileybrown, where he showed me how he uses Codex to run his entire content business (1.5M+ followers), including how to:
→ Create thumbnails, animations, and B-roll
→ Research and craft winning hooks
→ Use agents to find content ideas proactively
Some quotes from Riley:
“I’ll scrape a hundred thumbnails that performed really well and test putting my face on them." (Riley uses @paper to do this)
“I’ve never looked at a skill file once. If AI does a bad job, I just tell it to update the skill so it doesn’t make the same mistake again.”
“I’ll walk around for 10 minutes speaking all my ideas. When I come back, AI has created 80% of the diagrams I’ll use in my videos.”
I’ve admired Riley’s AI tutorials for a long time, so it was fascinating to see his process behind the scenes.
📌 Watch now: https://youtu.be/N34zz1-RSGw
Thanks to our sponsors:
Google AI Studio: Go from prompt to production fast https://aistudio.google.com/?utm_source=peteryang&utm_medium=partner&utm_campaign=26Q3-peteryang_ais
Riverside: All-in-one AI studio for podcasts and video https://creators.riverside.com/PeterYang

The proposed billionaire tax is asset seizure on unrealized gains. The issue I have is that this sets a precedent that will eventually reach every homeowner in the state. California will tax the equity of your home. It's only a matter of time...
Elex Michaelson: Should billionaires in California pay 5% of their assets to the state in order to fund healthcare programs?
@seiu_uhw's Dave Regan is leading the efforts to pass Prop. 40.
"Billionaires pay lower effective tax rates than you & I do. That's the real scandal."
Via

Why we are suing Randy Fine— and how he’s being held accountable by our lawsuit
@amasad
ADC National: We're suing Israel-first Congressman Randy Fine. After we sued him for blocking a Palestinian American from his official X account over criticism of one of his posts, Fine unblocked him and changed his social media policy. But we're still taking him to court to make sure he can't

Ideas are the new bottleneck
@akshaynathan_, Core Product Engineering, @OpenAI, interviewed by @swyx and @Vibhu (@LatentSpacepod)
Summary: Akshay Nathan runs the productivity pillar at OpenAI, the team behind ChatGPT Work and Codex. He argues that once anyone can build, the scarce inputs become ideas and taste, and the old proxies for productivity stop telling you anything. His warning to managers is that AI makes activity almost free while progress still costs the same discipline it always did.
1. The motion trap. The trap is conflating motion with progress. Adding models and standing up dashboards is easy now, and Nathan says plenty of teams do exactly that and find nothing has changed. Motion got cheap because the tooling got good, while progress still requires being prescriptive and deliberate about what you are trying to achieve. If your team cannot say what progress looks like this quarter, better tooling only speeds up the drift.
2. Ideas and taste. The bottleneck is now ideas and taste. Anyone can build, so build capacity stopped being the constraint. What limits you is the number of ideas and the number of things you are carrying at any given moment. Nathan calls this the era of bottoms-up ambition, where the scarce input is someone worth listening to about what to make.
3. Grounded ideas. Ideas do not come from a vacuum. One of the hosts said the single automation he wants and still cannot get is "bring me new ideas," and that LLMs keep failing at it. Ideas come from talking to users, reacting to friction you saw, or building on a foundation you already laid. That is why generalists who close that loop keep their value even as the building gets automated.
4. At-bats. For managers, Nathan watches at-bats, both the quantity and the quality. The loop he means runs from generating an idea, to building it, to getting feedback, to reacting, to actually validating or invalidating the hypothesis, then on to the next one. He measures the team's ability to complete that circuit efficiently, not the artifacts produced along the way. It is also a culture measure, since running the loop many times takes humility and the motivation to stay in it.
5. Falling proxies. The old productivity proxies are coming apart. Commits, lines of code, pull requests, story points, and now tokens used to track whether a team would hit its goal. Nathan says that correlation is breaking, and thumbs up and thumbs down do not rescue it, because you cannot tell if the user is rating the content, the vibe, or whether it helped them. Someone has to invent the replacement, because measurement is how anyone judges whether this is working.
6. Pride as signal. ChatGPT Work exists because non-developers at OpenAI started using Codex. In internal research sessions, people from strategic finance and marketing were using it for their own work, and what stood out was how proud they were, as if they were not supposed to have it. Nathan read that pride as evidence the power was never developer-only. Watch for users who are smug about your product, because that beats a satisfaction score.
7. Show, don't tell. Teaching capability through articles and onboarding does not work. In the early enterprise days, Nathan asked customers with big AI budgets what discrete use case they wanted, and got wild variance back, because a box you can say anything to is both the magic and the reason nobody knows what to do with it. People find the next use case by watching someone do it. He says show-not-tell is still not cracked.
8. T-shaped generalists. Everyone becomes a generalist with a specialty. Nathan says he could never have produced a design before, and still lacks the visual taste, but he can now iterate on one with AI. The generalist range comes cheap, and the specialty is the thing you are interested in and keep going deeper on. That combination is what makes his ceiling feel close to limitless.
9. No boxes by role. Do not draw product boundaries around who someone is. Nathan says his own job changes every few months, and the lines keep blurring between writing code, writing strategy docs, planning events, doing marketing, and recording podcasts. So Codex and ChatGPT Work share one agent harness, with opinionated differences only in the interface and the sandbox defaults. Let users choose an experience without locking them inside it.
10. Sites over decks. Interactive websites are replacing slide decks as the canonical team artifact. OpenAI's corporate finance team used to build its monthly reports in decks and spreadsheets, and now builds them as Sites. PowerPoint and Excel stay flexible until you hit a wall, either a feature you do not know or one the product never had, and with a site you can ask for anything. The model slider in the ChatGPT Work launch was designed inside a site.
11. Retry what failed. Retry the capability that failed you 3 months ago. Nathan's example is performance reviews: 6 months ago the model's help was slop, and this cycle it beat him at pulling context on what people had done and surfacing wins he never saw, because the agent reads the code, the reviews, and Slack. He still refuses to present model-written text as a review of a person, so what he delegates is the search and not the judgment. Broadening your sense of what is possible is his biggest piece of advice.
12. Ambition over headcount. Asked whether AI makes his teams smaller, Nathan says the opposite happens. Individuals and small groups now finish what used to take more people, and the amount worth doing grew at least as fast. So teams get more ambitious rather than leaner. That is a choice worth making on purpose instead of defaulting to headcount savings.

Physical AI evals are starting to become a real category.
Until recently, most VLA evaluation was basically:
LIBERO → success rate → leaderboard.
Now we're seeing a much broader stack emerge:
• Allen AI — unified VLA eval across 18+ simulation benchmarks
• LeRobot — one eval interface across multiple sim benchmarks
• PhAIL — real robots + production metrics like throughput and failures
• Robocurve — independent, real-world robot evaluation
• RoboDojo — bringing sim + real-world evaluation together
The interesting part isn't just more benchmarks.
It's the move from:
“Can the robot complete this task?”
to:
“How reliable, fast and general is this system in the physical world?”
I think independent physical AI evals are going to become increasingly important as robot models start looking more and more similar on demos.

If you want to get into YC, don't focus on getting into YC. Focus on building stuff and understanding your users. That's what gets you into YC.

I applied to YC 3 times with my previous startup because I really really wanted to get in.
Then at some point I kinda stopped caring and just thought what would be cool to build that the world is missing.
Magnus and I had this idea of controlling your entire computer with AI. We said let's just start with the browser, built a prototype in 4 days and just put it on showHN (no expectations - I was anyway planning to work on my masters thesis at Berkeley).
It went completely viral and that’s what got us into YC.
Turns out the mindset shift from "I need to get into YC" to "let’s just build cool shit" was basically all I needed
Paul Graham: If you want to get into YC, don't focus on getting into YC. Focus on building stuff and understanding your users. That's what gets you into YC.

see the thing is, he was already being positive by constraining ai job displacement to "entry-level" work, and claiming only 50% displacement instead of 110% or 200% (GDP growth utopia!)
and he led with how exciting this whole ai disruption thing is just like any other major disruption that took us from farming to factory to knowledge work and internet and computer
and so it'll all be swell if we prepare people for all this ai disruption by buying more of the ai he's selling because the more ai we buy the more protected from the ai disruption we will all be! https://x.com/NeetNewsClips/status/2034284957611294742/video/1

Ten minutes of talking gives Riley 80% of the diagrams that he shares in his videos.
From @rileybrown:
“I use @WisprFlow for everything. I’ll walk around for 10 minutes speaking all my ideas, then ask for an Excalidraw diagram with nine slides.”
“When I come back, it has 80% of the diagrams I’ll use. I'll spend another 20 minutes editing them, but it organizes my ideas.”
📌 Watch the full episode here: https://youtu.be/N34zz1-RSGw
Peter Yang: “The moat is quality over a long period of time."
Here's my new episode with @rileybrown, where he showed me how he uses Codex to run his entire content business (1.5M+ followers), including how to:
→ Create thumbnails, animations, and B-roll
→ Research and craft winning

"There are no Amish with autism"
There are
"Vaccines aren't tested against placebo"
They are
"MMR has never been studied as a possible cause of autism"
It has. It's not the cause.
I apparently need to say this stuff over and over and over and over again.
And again.

mRNA vaccines are vaccines
Chemtrails don't exist
Turbo cancer doesn't exist
Covid does exist
Covid vaccines worked
Ivermectin doesn't work for Covid
Ivermectin doesn't work for cancer
Vaccines eradicated smallpox
Vaccines don't cause autism
You're welcome
Alex Boge: Want engagement on X? Post a basic scientific fact.
You’ll summon the anti-science crowd like ringing a dinner bell.

Thanks to @Replit we’re LIVE on the App Store 🙏🏾🙏🏾

There are two separate questions regarding Dario Amodei’s post about AI and biology: motivation and reality.
Motivation. Dario started in biology (was a PhD student of the great Bill Bialek). It might be that the master plan really was to first build much more capable AI systems, and eventually use them to cure diseases. Maybe coding models, chatbots, Claude Code, etc. were never the final destination. They were the road toward models capable of becoming genuinely useful scientific collaborators.
It might also be that there was no master plan at all. Anthropic spent years pushing the frontier in coding and general reasoning; those systems have now become much more capable, and this is simply a good moment to turn more attention toward biology.
Maybe, but there is also a more cynical interpretation:
“Curing cancer” is an extraordinarily powerful form of reputation laundering.
They are basically saying: "Yes, we are perhaps creating the largest concentration of technological and economic power in modern history."
Yes, we want governments to restrict who can build and access the most powerful versions of this technology.
Yes, ordinary people increasingly have less visibility into what the frontier systems can do.
But don’t worry - we’re the good guys! We are going to cure cancer.
And this is where I disagree with Dario’s claim that “the thing that will work is actually curing cancer.”
Suppose Anthropic really does help cure a major cancer. That would be an extraordinary achievement, but curing cancer would not answer the question of who should control AI.
A company can create enormous social value and simultaneously accumulate too much power.
The pharmaceutical industry itself is a good example. Creating life-saving drugs doesn’t automatically resolve questions about pricing, access, lobbying, competition, data rights or public accountability.
And there is another problem with the biology story.
I am very bullish about AI for science. We’ve discussed AI drug discovery and related areas several times on our podcast. AI can accelerate literature review, target discovery, protein and molecule design, experimental planning, analysis, and many other parts of the pipeline.
But biology is not software.
You can make an AI reason 100x faster. You cannot simply make a human body respond to a drug 100x faster.
Wet-lab experiments take physical time. Toxicity takes time to observe. Human trials take time. Long-term side effects take time. Diseases are heterogeneous. Manufacturing, recruitment and validation happen in the physical world. AI can remove some bottlenecks, but not all of them.
Also, if these models really have a plausible path to curing cancer, Alzheimer’s, hepatitis and other diseases, then “pacing AI progress” has a cost too.
If delaying frontier AI by a year delays a cure by a year, that isn’t an abstract opportunity cost. People die during that year.
And of course the most important question that I think Dario’s post doesn’t answer:
Who gets to decide who may use intelligence, whose data may be used, and for what purpose?
Because almost every “safety” policy has a political economy.
If the outcome is consistently:
Anthropic gets more control,
frontier models become less accessible,
customers have fewer choices,
smaller competitors face higher compliance costs,
and governments become increasingly dependent on a handful of frontier labs to tell them what is safe…
Then it's rational for people to ask whether every restriction is only about safety.
That question remains legitimate even if everyone at Anthropic is acting in complete good faith (Are they?)
Maybe people aren’t distrusting AI because they don’t understand how wonderful the future could be.
Maybe they distrust a future in which a tiny number of companies build enormously powerful systems, tell everyone else how dangerous those systems are, lobby for rules governing who gets access to them and then ask society to trust that the companies controlling them are the right people to decide.
If Anthropic helps cure cancer, I will be the first to applaud.
But “we cured cancer” is an argument for AI's value. It is not an argument for concentrating control of AI.

There's a pervasive myth that "tax policy changes" are the reason economic inequality is growing. Actually it's because technology makes it increasingly easy to start companies, and makes them grow faster once you do.
https://paulgraham.com/richnow.html

Codex is for figuring out your kid's after school schedule from your inbox.
My wife is impressed 😅