tracking where the money goes in crypto and AI. ETF flows every morning. real numbers, no hype
following the money
Joined March 2025
- Tweets731
- Following184
- Followers150
- Likes2.3K
SolFylin retweeted
ANTHROPIC ENGINEERS JUST TURNED OPUS 5.5 INTO A REAL OPERATING LAYER
claude is usually treated like a chat window. this setup makes it the layer deciding what should happen next across an entire workflow.
state → opus 5.5 → tools → code → external systems → human review
pricing makes that architecture much easier to run at scale: $4 input, $20 output and just $0.20 for cache reads per 1M tokens.
one tester pushed a 680,000-line migration through in under a day, while another hit 39/40 successful load-time fixes across a web app.
that’s the part i care about: opus 5.5 starts looking less like an assistant and more like the reasoning core behind long-running research, coding and agent systems.
SolFylin retweeted
I’ve been accused of being a long only permabull a lot recently.
Well brother, I posted quite a few bearish posts now.
Does this mean to go out and short? I always say 95% of ya’ll should NEVER short.
BUT this does mean raise cash, take some profit perhaps, hedge, sell CC’s, don’t utilize leverage for god’s sake, and wait for the right opportunity to btfd.
Sometimes wealth preservation > wealth accumulation. Better than losing right?
SolFylin retweeted
$SPX $XLF One more for bear's 👀 only.
I mean the $QQQ I posted yesterday was pretty good I guess.
But here's another one.
They usually say Financials/Banks are leading indicators in the stock market. Well it has been just horrible lately. Yet the S&P500 has been just chopping at the same time.
Very reminiscent of Jan-Feb earlier this year ain't it?! Before finally caving in...
Let's see how this plays out.
SolFylin retweeted
Everyone told you frontier AI needs a data center.
This repo just called that a lie — and backed it with numbers, not hype.
colibri runs models up to 2.8 trillion parameters. On a laptop. No GPU required.
Here's how it does it:
→ Pure C. Zero engine dependencies.
→ Treats storage, RAM and VRAM as one single memory hierarchy.
→ Streams experts straight from disk instead of loading the whole model at once.
Nine model families run today — GLM-5.3, DeepSeek V4, Qwen3.8, Kimi K3 (2.8T params) among them. One C file each.
I watched it live:
A 744B parameter model answered a real prompt in 32 seconds, resident at 9.9GB RAM. On CPU.
Then it showed its own performance dashboard — 2.9 tokens/sec, 19 seconds wall time, disk I/O overlapped with compute, zero hidden magic. Every phase measured, nothing hand-waved.
And the part that actually stopped me: a "Brain" page that maps all 13,260 experts inside the model like a cortex — Python, Law, Medicine, Chinese, each its own region you can walk into and inspect.
This isn't a demo. It's an open research platform, and the maintainers are explicit about it: no SLA on speed, hard guarantee on correctness. Nothing gets faster by quietly changing what the model actually computes.
37.3K stars. +4,000 in the last 7 days alone. It's not slowing down.
The GPU tax everyone assumed was mandatory for frontier AI just quietly stopped being true — and almost nobody's talking about it yet.
Would you run a 700B+ model on your own laptop if you could? Or does "no GPU" make you trust it less? 👇
SolFylin retweeted
Raiffeisen is bringing crypto trading to 11 European markets through Bitpanda.
The partnership could give around 18 million customers access to crypto through Raiffeisen’s banking network.
This is the kind of adoption I’m watching.
Crypto becoming available through major banks is a very different story from people simply downloading another exchange app.
SolFylin retweeted
🇺🇸 NEW: SEC Director of Trading and Markets Jamie Selway says tokenization and crypto have become politicized, but building these markets should have bipartisan support.
SolFylin retweeted
🚀 INSANE: Bitcoin is now ALMOST projected to go above $100,000 this year.
PEOPLE ARE GETTING BULLISH!
🤝 Paid partnership
SolFylin retweeted
Meta, $META, has been testing a human concierge for its new AI assistant, Muse, which entails having human contractors quietly handle some of the phone calls for the digital agent, per Bloomberg
SolFylin retweeted
BREAKING: Grok Bot helped SpaceXAI handle a 175% increase in support tickets with zero new hires.
• Support ticket volume increased by 175%, but SpaceXAI did not need to hire any new support staff.
• Without Grok Bot, the company estimates it might have needed around 200 additional people.
• Grok Bot resolves 99% of refund requests without needing help from a human.
• Each resolved ticket can cost as little as $0.20 to $0.30, compared with $1 to $4 for traditional AI support tools.
• It was trained on more than 1 million customer interactions to learn how the support team communicates.
• It analyzes more than 20,000 pieces of customer feedback every day and turns them into useful insights for engineers.
SolFylin retweeted
Michael Burry still shorting some of the biggest tech names out there in $MU, $ORCL, $PLTR, $SOXX etf, $NBIS.
And is going long…. Build a Bear $BBW.
BREAKING: Michael Burry has disclosed updated positions.
He shorted more:
-Micron $MU
-Nebius $NBIS
-Semi ETF $SOXX
-Palantir $PLTR
And purchased more (long):
-QXO $QXO
-Build-A-Bear $BBW
-Sprouts $SFM
-Birkenstock $BIRK
-Mercado-Libre $MELI
He says he will look to cover his shorts if the market takes them to new highs.
SolFylin retweeted
$BTC Polymarket.com odds of hitting $95k, $100k, $110k before eoy.
I think $100k is very doable at this rate. Get + money on this.
🤝 Paid partnership
SolFylin retweeted
This is the first time a federally chartered, OCC-regulated, FDIC-insured U.S. bank has begun live card settlement with a stablecoin across Mastercard’s global payments network. Transactions are already running on-chain.On September 22, 2026, SoFi and Mastercard announced that SoFiUSD settlement is now live in production. SoFi Bank is migrating its entire debit and credit card program to blockchain-based settlement. Expected annualized volume: more than $25 billion
xitter.cf/i/trending/21023728707…
SolFylin retweeted
🇺🇸 AI’s next bottleneck might not be chips or smarter models. It might be electricity.
The IEA found data-center electricity demand jumped 17% in 2025, with AI-focused sites growing even faster, while global electricity demand rose about 3%.
It expects data-center use to roughly double by 2030, with AI-driven demand tripling.
The problem is everything connecting those servers to power.
Grid connections, transformers and turbines are already struggling to keep pace.
Morgan Stanley now projects a 15-gigawatt U.S. power shortfall relative to AI chip demand by 2027, widening to 30 to 40% by 2028.
Its shorthand for the gap: about “6 New York Cities” worth of electricity.
Roughly 43 gigawatts of AI data-center capacity is planned across the U.S., but interconnection queues in hubs like Northern Virginia, Phoenix and Dallas can stretch 4 to 7 years.
Much of that capacity is waiting on power, not GPUs.
Efficiency per task keeps improving, but usage growth and heavier applications like AI agents are outpacing those gains.
Turns out the smartest model in the world still needs a wall socket.
Sources: IEA, Financial Times, SiliconReport / Writer: Julie
