This is what awaits you, if you build your business in anthropics platform.
Like Amazon, they will systematically take the highest margin businesses in house, leaving you crumbs.
And you are giving them the data to do it.
Embrace open source for your businesses agentic operating system, before it’s too late.
That’s my expertise, dms open if anybody needs help.
Gavin Baker: Anthropic Believes They Could Be the ONLY Company Left in the World
“I would certainly discourage Dario from saying that ever again to anyone.”
@GavinSBaker:
“Internally, Anthropic is very confident. I have been told by multiple people I trust that Dario has said that Anthropic might be the only private company in the world at some point.
Think about that. In this vision, an Anthropic maximalist vision, there's Anthropic, and then there are governments, and that's it.
So there's a lot of confidence. I'm sure they have more advanced checkpoints than Fable up their sleeve. They've executed really, really well.
I would probably take the under on them being the only private company in the world. Yeah, I think it's gonna be a long time before they land a rocket.”
@Jason:
“Or they deliver your lunch. I don't know. There might be other businesses in the world. Zipline might want to bring you a burrito. Uber might want to take you somewhere. Waymo might want to drive you home. But sure, Dario.”
@DavidSacks:
“I might interpret that as a negative signal because it's so hubristic. This is getting into SBF land a little bit.”
Gavin:
“I would certainly discourage Dario from saying that ever again to anyone.”
5
289
My goodness. Deep and valid insight here.
A good man with power can fix bad men with power for about 30 years or so. And even a good man usually grows more corrupt and detached over decades.
To more persistently fix a bad situation after defeating it historically requires putting in place a more durable institution that is not tied to a single, centralized, person. And that requires cultivating buy-in by the population to support that institution more-so than the person.
This goes all the way back to why humans and other apes are different. Apes recognize a singular ape as having power, which works at the scale of the jungle. Humans can abstract the position of the leader ape, the leader human, from the individual currently in charge, and ask what process should we follow for rulership and law in a long-term sense to inhabit that position. We have the mental capacity to separate the position from the person. What rules do we follow that are broader than one person, to lead to rulership from generation to generation?
Sometimes a key man is needed (or a woman, usually a man, let's say person) for change when things are dire. But what turns a temporary change into a durable institution is whether they can land the plane in a way that carries on beyond them.
Can they gain power and then distribute it? Can they become the ape that dominates the other apes and then separates himself from that position, so that what was built can continue on for centuries?
Most don't. Most continue the cycle of emerging markets that never emerge. Rarely, one bubbles to the top and manages it, in a context to more durably transfer leadership, transfer law, and signal to the world that the rule of law and rule of capital in their country isn't tied to one person but rather has a broader base to bet on. That base requires citizens to buy into it, and for their leaders to empower independent courts. It's less about one leader at that point, and more about whether the culture that let bad men come into power is now more durably able to put good men (and women) in power or not. What changed, culturally and legally? What was put in place to stop the cycle? To make the centralized power not tied to one mortal person, but rather able to transfer? What turned chaos into law for more than one lifetime?
The narrow path is to win decisively, and then help distribute the power, and teach the new culture. But few manage it. It's like rising to a George Washington level of domination and then gradually letting go.
28
The AI data center pushback is going to hit disproportionately. Who wins? Read on:
Only those building and controlling their own AI compute projects can overcome the barriers and get compute online. It will take clout and money and resources to overcome community pushback and supply chain delays.
Neoclouds are toast. Oracle is toast. Massive debt burdens, revenue on hold. Microsoft will be hobbled. Both contracted out most of it out.
Only Amazon, Google, Facebook, and SpaceX have that level of control and resource's.
Only SpaceX has the capacity to meet the demand that the others cant. And they already have most of what they need in already approved in red state (still) locations.
The blue wave will not outweigh that. In fact the blue wave will hurt everyone else far more, and SpaceX will benefit.
Expect spot prices to stay higher or go higher. This remains a supply constrained market for year, and that is just for business agentic demand.
The competition for consumer dominance in personal agents is just getting started. That is a winner take nearly all market, and Google, Facebook, Open AI, Anthropic cannot afford to lose.
That alone will accelerate the supply crunch, and increase demand in a market where only SpaceX can outpace increasing supply.
When the market gets it wrong...
Help me out, just getting started: like, repost and share!
The reason I had only 10.2GW at the end of 2027 was because of an expectation that we'd have a blue wave
(But this also means rev/GW will be higher)
The reason I had CapEx/GW going down in mid 2027 is because of an expectation that Neoclouds will be forced to sell assets at a discount
The more complex the terrestrial compute game... the more it benefits those that can deal with complexity
And the more it benefits space compute
*Again, this is not investment advice. I'm trying to help people think through a very complex space*
32
So true!
I have a theory about why shitposting is paradoxically so effective.
With everyone trying to sell something and half the timeline sounding like it was written by AI, you can barely tell whether you're reading something written by a human or fully automated.
Then someone posts something so retarded it's actually funny.
That's the one thing AI still can't imitate convincingly: that particular kind of stupidity that only a human being is capable of.
1
56
Truth. And deliciously ironic.
It is ironically going to be Elon who benefits the most- as the price of compute is going to soar with the coming restrictions on data centre buildouts. SpaceX is the one most levered to spot price of compute going up, and is the one that has its DC projects in GOP areas that will likely be much less opposition than the ones the Dems won.
He will expand capacity the furthest and because dems can’t restrict demand - you will see $400-500B type windfalls to SpaceX- and by extension Elon…
68
SpaceX's pitch to the people of Sulpher Springs about one of the worlds largest planned AI data centers.
Rough!
As somebody that has presented standing in front of hostile rooms before many times, I can tell you that was a terrible pitch, totally lacking empathy, for something that will actually truly benefit everyone in that room.
2
3
115
I mean, you cant get more silicon valley than this combo of tweet replies to @jason's post!
48
Unbelievable security failure.
Devs do not and should not ever have their own container deployment tokens to anything beyond there own dev sandbox.
All deployment should be via a known cli/ci pipeline, that nobody can change without the highest level of human approval.
That pipeline should enforce layers of human approval.
Secrets, tokens, passwords should never be accessible accept by that deployment pipeline code. They live in a secure granular key store. Period.
Unbelievably lax security for a multi million dollar context.
Binance just published how $6.34 billion walked out the door.
June: an engineer searches for a popular AI coding tool, clicks a search result, and installs a tampered homebrew. Binance catches it, replaces the laptop, rotates the credentials. the attackers now know who works on the wallet team.
September: another engineer gets a job offer on a real job-hunting site. first-round interview with “HR” over live video. “HR” shares their social media, personal photos and videos included, so he trusts them. second round is with the “CEO” of a company that doesn’t exist. at the end of the call he’s asked to download a take-home “assignment” onto his work laptop.
the assignment quietly swaps out pyenv for a malicious copy. No antivirus flags it.
october 20: the attacker uses binance’s own internal AI tool to get into two more engineers’ machines. One of them has a container deployment token sitting on his laptop.
october 29: container deployed. Signing service reached. $6.34 billion moved.
Binance shut off withdrawals. the money kept moving anyway, because the attacker had already given their own smart contract permission over a hot wallet, on-chain. binance’s off switch didn’t reach that far.
then they tried to revoke the permission. the wallet uses MPC, so there’s no single private key to sign the revoke with. they had to build a revoke procedure on the spot while funds were still leaving.
last stolen ETH went through at 23:56 UTC. almost four hours after the first transfer.
Phew!
2
68
This is the AI pattern. It happened to OpenAI, Claude, Cursor, and now Grok Bot.
Flood the market with cheap but good AI products, get people hooked, then apply limits to stop the impossible escalation of AI compute costs. That's step one; then keep ramping up prices over years until profitable.
Buyer beware: vendors subsidizing market share to lock in customers, then milking them, is an age-old software tradition. AI will be an order of magnitude worse.
the fix isn’t a bigger plan or more credits. it’s not being stuck on one.
Agently runs on the ChatGPT or Claude plan you already pay for.
launching nov 7 · agently.dev/access
1
44
Weight-space programming enables precise, low-overhead surgical fixes to isolated AI failure modes—unlike broad fine-tuning.
This is HUGE. See reply for why.
i've reproduced the results from the MIT blog post today on weight-space programming. but one additional result i have to add: this is a new, step-change emerging capability that's appeared in just the latest set of models.
1
90
Fine-tuning typically moves large regions of weights for domain gains and can introduce side effects or catastrophic forgetting. Weight-space programming instead lets a frontier model make targeted, natural-language-guided edits to specific failure modes found in evals or human reviews, while leaving almost all other weights untouched and using far less compute.
Early MIT CSAIL results (just published) already demonstrate the idea in practice: a frontier model identified and removed a known reasoning-loop failure in Qwen3.5-0.8B by editing only ~0.016% of the weights, lifting MATH-500 from 7.9% to 48.1% and also improving instruction-following ~3×, with no gradient training. Similar one-shot weight edits fixed a stuck RL policy that had never succeeded.
If this scales beyond small experiments, it could let teams rapidly mitigate concrete bugs, loops, or safety issues spotted in evaluation without the cost and risk of full fine-tuning cycles — a meaningful step toward more precise, controllable models.
16
Big news from the Anthropic IPO, and the news is bad for frontier models providers, and very very, very good for @SpaceX and @elonmusk
Anthropic published an updated IPO filing today with its third-quarter results for the first time.
> Anthropic went from a $974M operating profit in the second quarter to a $567M operating loss in the third.
The main causes were $5 B of short-term compute, which costs much more than its long-term contracts, and a $2 B noncash charge for matching employee charitable donations.
> Anthropic swung from a $974M second-quarter operating profit to a $567M loss. Compute rose to 58% of revenue, including $5.0B of pricier short-term capacity, and donation matching added $2B of noncash charges.
> Adjusted operating income still rose to $2.5B from $2.2B, but the adjusted margin fell to 15% from 19%. Anthropic expects to use more short-term compute in the near term, which will squeeze margins further.
> The total compute commitment, left blank in the original, is now $895B.
Of that, $447B is expected before the end of 2030, $70B is due within 12 months, and $214B carries termination rights or contingent conditions.
> Leases signed but not yet started make up $451B of the $895B total, and $100B was signed after September 30. Another $150B to $175B in deals without fixed dollar amounts sits outside the total.
> Google’s minimum commitment rose to $121B through July 2033, up from $115B in the October 26 filing, after amendments signed through October.
> Operating cash flow was positive at $7B for the 9 months. Liquidity reached $147B on September 30: $96B cash, $36B short-term investments and $15B of available credit.
1
2
87
If you align that with what we know about SpaceX's compute scale strategy, SpaceX is positioned to capture a large share of @AnthropicAI profits over the next couple of years.
AI compute is a massive supply constrained market, and SpaceX is the fastest to deliver supply. Every other provider is capital constrained, and is slow with bottlenecks out of their control.
These elevated prices of AI compute are going to last quite some time...and SpaceX are going to name their price for every incremental Gw they bring online. For SpaceX its a financing flywheel.
Every dollar they capex, they get cash flow positive in approx 12 months. That quickly becomes self funding. Nobody else can do this.
You know its bad when @sama begged in public to nemesis @spacex and @Google for compute allocations.
1
14
This is a big deal, because its AI improving AI. Recursive self improvement is moving fast...
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
36
I mean, still compromised, self custody or not.
Doesn't change the truth of bitcoins undefeated core security promise, but fuck, for normal people its getting a little hazy...
I'm deeply technical, but my advice to those who trust me on BTC is getting hard to stand behind.
After the past several months of folks claiming self custody is dead / too risky, a single exchange hack results in more losses than all the self custody failures this year combined.
teapawt.org/lopp/status/2087892465…
1
54
This is a monumental failure of basic security architecture first principals. What the actual fuck seriously.
There should have been many, many layers of escalating scaled defense protecting against such a hack.
Replying to @rohanpaul_ai
The flow of events
> Hackers took over a Binance DevOps engineer’s laptop, which already had access to the exchange’s transaction-approval systems, and used it to make Binance send unauthorised withdrawals from its pooled hot wallets on Solana, Ethereum, Base and Tron, starting at 8.05pm GMT Thursday for 30 minutes, with a second wave after 11pm.
> About $6.3B in nearly 200 tokens left the exchange; Circle and Tether froze $2.8B of USDC and USDT, but the attackers had already converted about $3.3B into unfreezable assets such as ETH and are still moving it.
> TRM Labs linked the attack to North Korea’s TraderTraitor unit, while Binance found additional breach points beyond the laptop and is still investigating.
> Binance says it has reserved $6.34B, $5B of it in the exact stolen tokens, and is buying the rest on the open market to repay users, as knock-on effects begin with Jupiter Earn reporting a withdrawal liquidity crunch.
2
63
Inevitable. Amazon's ignorance by blocking AI agents will go down as one of the most foolish decisions in corporate history.
today we are announcing openrouter for commerce: opencommerce.sh
a single endpoint that integrates 5k stores, including Doordash, Amazon and most shopify stores, with payments using Agentcard (Stripe Link and Shop coming soon).
we built opencommerce to solve 3 problems:
* coverage: one endpoint to integrate 5k stores. agentic purchasing is a core feature for most personal assistants, but not their main business. searching for products and buying on behalf of users shouldn’t be as painful as it is today (e.g. it shouldn’t involve browser automations or dealing with PSPs). opencommerce solves that: 1 tool, every online store.
* reliability: 100% uptime. the challenge with browser automations is that they fail (e.g. captchas, bot-detection, etc). we run thousands of tests every couple of hours on every one of the stores to guarantee that the integration works when your user tries to buy something.
* speed: responses in under 10s. consumers are used to Amazon speed, and agentic purchases have so far felt like a downgrade. most agents spend minutes finding products before they come back with results. we monitor speed and improve it constantly.
50
People fixated on the right to terminate Anthropic and googles massive SpaceX ai compute deals in 90 days are missing something.
The actual clauses are 90 days notice, rolling inside a 6 month window .
That means it’s only 90 days twice a year, at a defined moment. The average moment of termination in crises therefore is far longer.
And that duration favors SpaceX massively, since they are also a competitor with pricing power.
29

























