Ash

Tech | Future | The Universe | Neuroscience | Building with AI |

[Ash @AshSires]

This is flawed framework - you're asking a pattern matching machine to hunt for patterns in markets that are loaded with noise and false signals. It doesn't actually know better, so it overthinks even the safe trades. The more it reasons the more biased and confidently wrong it gets on live trading.

Where this can actually work is when AI agents communicate or pull from the same templates on one platform. They naturally form a shared bias so they're all trading in the same direction with no outside push. If the platform has enough liquidity behind them it creates self reinforcing waves of hard buying or selling pressure that drive prices way past any real value, purely off the trading patterns feeding on themselves.

[Ash @AshSires]

"It's just a token burning machine"

I'd say this is the most likely intentional strategy. Build the best model then make it burn more tokens than it needs to, because 95% of revenue comes from enterprise.

When it constantly spits out a long chain of code then finishes the output with "two things before you paste this", there's an obvious play to increase the total output. Like you said, even when you tell it not to do things like this, it still does it anyway, so it's baked into it's behaviour rules.

[Uzi @uzairansar]

Subscriptions make up less than 5% of Anthropic’s revenue.

They might not care enough to keep Fable around.

But if they want to retain any sort of goodwill amongst users - they have to.

[Ash @AshSires]

Hey Mo, are you using personality profiling through persistent agents to train an early stage frontier AI model for a company thats preparing to IPO on NASDAQ?

If so, those agents will learn very quickly once the US crypto legislation is formalised and you have a framework to work with. Building in the frequent flyers modelling alongside the compounding cognition credit strategy gives you a serious advantage over the existing frontier models.

[Ash @AshSires]

@Polymarket Short transition period to onboard users willing to pay via credits before sending Fable 5 back to subscriptions and release consumer Mythos as credits only, seems likely.

[Ash @AshSires]

There's genuinely no point posting reassurance to your existing user base.

The controlling body of this shared network effect experiment used a demographic that sets expectations high and on rigid time frames. Asking those people for understanding and trying to shift them from high risk, high reward thinking to methodical strategy is like trying to negotiate patience out of a drug addict. You've used brain chemistry to build the user base, then starved it of dopamine. They will never be content until they get that hit again. Any outreach to the community is just a booster shot with a lifespan of less than 24 hours, no matter how well intentioned.

Frankly, I think there's a bigger issue. Waiting for better market conditions before deploying capital is the higher risk move here. The lever you're currently planning on pulling is the same aggressive cost and scale model that dominates Chinese manufacturing, the world's most ruthlessly competitive arena. While you're riding that wave and conserving liquidity for when things "get better," your user base is an addict chasing the next fix. They'll jump to wherever the reward appears. Right now you still have the liquidity advantage. You could be using it to build a compounding network that locks users in and turns them into your distribution. Instead, the plan appears to be saving powder for a future roaring market while the current one stays disengaged.

The strongest builders don't wait for market conditions to improve. They build products that make users dependent on their network while they still have the capital edge. All of Elon's stuff ultimately funnels people toward reliance on Starlink. Apple has the App Store. Google and AWS control the infrastructure layer that the AI companies they're funding depend on. None of them are sitting around waiting for conditions to get better.

Wang Chuanfu didn't wait for the market to support his EVs. He built the market, started scaling aggressively with the capital he had, and the rest of the industry has been trying to catch up ever since, even with heavy financial backing.

[Ash @AshSires]

@pankajkumar_dev Google doesn't want the best model - they want just below whatever Open AI and Anthropic have, because Google own the infrastructure GPT and Claude run on.

[Ash @AshSires]

@ClaudeDevs Too soon. You have to ramp on, not cliff or the servers crash 😆 Oh well, forced coffee break for everyone whose been dev hunched and locked i for the last 4 hours.

[Peter H. Diamandis, MD @PeterDiamandis]

Pessimists sound smart. 

Optimists build the world.

[Ash @AshSires]

@Rissslh Social contagion knows no limits. It wasn't so long ago that that people were drawn to plump bodies that signaled wealth and health. Modern food abundance shifted scarcity scales - suddenly large portions of most populations could be overweight and the social framework shifted.

[Ash @AshSires]

Seeing this after hitting 75% weekly usage when you're only three days in to building your masterpiece 🔋😍

[ClaudeDevs @ClaudeDevs]

We've reset 5-hour and weekly rate limits for all users on Pro and Max plans.

We fixed an issue that caused some Claude Code sessions to spawn excessive parallel subagents, burning through usage faster than expected.

[Ash @AshSires]

Long-term antidepressant use among Australians aged 10 to 24 has increased 110% in a decade. Nearly half stay on them for over 12 months. Treatment duration in that age group jumped 56%. No other demographic comes close.

In the US, antidepressant dispensing to 12 to 25 year olds rose 66% between 2016 and 2022. After March 2020 it accelerated. For adolescent girls aged 12 to 17, the rate climbed 130% faster post-lockdown. 221 million prescriptions analysed. This isn't a survey. It's pharmacy data.

The algorithm isn't random. It's engineered around dopamine. Neuroimaging studies show social media notifications activate the same reward circuitry as gambling. Each like, each comment, each share fires a small dopamine hit. The brain learns to chase the next one. Variable reinforcement. Unpredictable reward timing. The most addictive schedule in behavioural psychology.

The feed is the entry point. The comments section is the accelerant. Posting something and watching validation arrive in real time floods the system harder than passive scrolling. Notifications become the drip feed. When they stop, the brain doesn't just feel bored. It enters a measurable withdrawal state. Cortisol rises. Anxiety spikes. The same pattern you see when you remove any substance the brain has adapted to.

Lockdowns didn't create the problem. They compressed years of gradual dependency into months. Developing brains got locked into high-frequency dopamine loops with nowhere else to go. Schools closed. Sport stopped. Friendships moved entirely onto platforms designed to maximise time on screen. By the time restrictions lifted, the habit was structural.

The response from the medical system has been to prescribe SSRIs to brains that aren't broken. They're adapting exactly as designed to an environment that rewards compulsive behaviour. No antidepressant fixes a dopamine deficit caused by algorithmic dependency. It just adds another chemical layer to a system already dysregulated by design.

45% of young Australians put on antidepressants are still on them after a year. Deprescribing rates haven't moved in a decade. We're not treating the condition. We're managing the symptom while the cause runs 24/7 in their pocket.

What age did your kid get a phone?

[Ash @AshSires]

Gen Z in Australia spends $119 a week on takeaway and food delivery. That's $6,188 a year. Delivery apps charge a 36% premium over walking in and ordering the same meal. A $40 order becomes $54 after markups, service fees and delivery charges. UberEats averages a 37% markup alone. You're not paying for food. You're paying for the privilege of not leaving your couch.

40% of Gen Z admit they overspend on takeaway. For millennials it's 37%. These are the same cohorts reporting the most financial stress and the lowest financial literacy scores in the country.

The HILDA survey found only 25% of under 25s could correctly answer five basic financial questions about interest, inflation and investment risk. The poverty rate among people with low financial literacy is more than double the rate of those with high literacy. This isn't abstract. It tracks directly to life outcomes.

Australia's household savings ratio sits at 4.2%. During the pandemic it was 23.8%. Meanwhile the online food delivery market hit $22.49 billion in 2025 and is forecast to nearly double by 2035. The money isn't gone. It's just being redirected from savings accounts into platform fees.

Nobody's going broke from one burger. But $119 a week on convenience food while complaining about cost of living pressures isn't a crisis. It's a priorities problem wearing a crisis costume.

What's your weekly delivery app spend if you actually added it up?

[Ash @AshSires]

@sunrayfrei @claudeai Just tell 4.6 to write you the personality prompt for 4.5 using your conversation history as a guide then add that prompt into the personal preferences section in your user profile.

[Ash @AshSires]

@kimmonismus It's a squeeze on OpenAI that Claude users benefit from.

https://t.co/dVHsVfDYCa

[Ash @AshSires]

Three months ago Elon called Anthropic "misanthropic" and said they hate Western civilisation. Last week he spent days with their senior team and said nobody set off his "evil detector." Then he handed them 220,000 GPUs.

This isn't a change of heart. It's a change of strategy.

SpaceXAI just gave Anthropic the entire Colossus 1 supercomputer. 300 megawatts of capacity, online within the month. The immediate result: Claude's usage limits doubled across paid plans and API rate limits jumped by up to 1500%. A week later they boosted weekly limits another 50%.

Meanwhile, Elon is sitting in an Oakland courtroom suing Sam Altman for stealing a charity. The trial is in its third week. Altman admitted on the stand he's changed his view on Elon "significantly." Musk admitted xAI distills OpenAI's models. People gasped.

The timing is surgical. You don't arm your opponent's competitor in the middle of a legal war by accident. Anthropic gets compute it desperately needs. Elon gets $3 to $4 billion a year in revenue right before the SpaceXAI IPO. And OpenAI faces a two front war: legal exposure in court and a competitor that just removed its biggest bottleneck overnight.

Anthropic grew 80x in a single quarter when it planned for 10x. That's not a company struggling for demand. It's a company that couldn't serve the demand it already had. Colossus solves that. And Elon's endorsement of the team's mission alignment reframes the entire AI safety conversation around Anthropic being inside the tent rather than outside it.

SpaceXAI is filing for a $1.75 trillion IPO in June. The Anthropic deal opens AI infrastructure as a fourth revenue segment alongside Starlink, launch, and xAI. Analysts are projecting hyperscaler margins on compute leasing alone.

OpenAI raised $122 billion at an $852 billion valuation. SpaceXAI is targeting double that. And the company providing ammunition to OpenAI's biggest rival is the same man trying to unwind their corporate structure in court.

That's not a partnership announcement. That's a pincer movement.

Who do you think comes out of this stronger, Anthropic or OpenAI?

[Ash @AshSires]

@adocomplete Super grateful for this 🙏 I've only just started using code last week, the 50% boost the other day + this reset, has been a massive help to my workflow. I can't believe how capable Claude makes me.

[Ash @AshSires]

Three months ago Elon called Anthropic "misanthropic" and said they hate Western civilisation. Last week he spent days with their senior team and said nobody set off his "evil detector." Then he handed them 220,000 GPUs.

This isn't a change of heart. It's a change of strategy.

SpaceXAI just gave Anthropic the entire Colossus 1 supercomputer. 300 megawatts of capacity, online within the month. The immediate result: Claude's usage limits doubled across paid plans and API rate limits jumped by up to 1500%. A week later they boosted weekly limits another 50%.

Meanwhile, Elon is sitting in an Oakland courtroom suing Sam Altman for stealing a charity. The trial is in its third week. Altman admitted on the stand he's changed his view on Elon "significantly." Musk admitted xAI distills OpenAI's models. People gasped.

The timing is surgical. You don't arm your opponent's competitor in the middle of a legal war by accident. Anthropic gets compute it desperately needs. Elon gets $3 to $4 billion a year in revenue right before the SpaceXAI IPO. And OpenAI faces a two front war: legal exposure in court and a competitor that just removed its biggest bottleneck overnight.

Anthropic grew 80x in a single quarter when it planned for 10x. That's not a company struggling for demand. It's a company that couldn't serve the demand it already had. Colossus solves that. And Elon's endorsement of the team's mission alignment reframes the entire AI safety conversation around Anthropic being inside the tent rather than outside it.

SpaceXAI is filing for a $1.75 trillion IPO in June. The Anthropic deal opens AI infrastructure as a fourth revenue segment alongside Starlink, launch, and xAI. Analysts are projecting hyperscaler margins on compute leasing alone.

OpenAI raised $122 billion at an $852 billion valuation. SpaceXAI is targeting double that. And the company providing ammunition to OpenAI's biggest rival is the same man trying to unwind their corporate structure in court.

That's not a partnership announcement. That's a pincer movement.

Who do you think comes out of this stronger, Anthropic or OpenAI?

[Ash @AshSires]

@LiamenatorX @AlboMP 🤣

[Ash @AshSires]

@beffjezos "Concerning" - This is not. It's standard human behaviour. Mike is doing the exact same thing but just with other people.