Michael Rossen 🚀

Applied Physics grad | AI × Crypto × Physics | 21

[Michael Rossen 🚀 @michaelrossen]

I'm a 21-year-old physicist.

Follow me for:
- AI breakthroughs
- Crypto predictions
- Productivity hacks
- Book recommendations
- Mars colonization updates
- Coding with AI assistants
- Physics simulations

Don't miss out on the future.

[Yam Peleg @Yampeleg]

gpt 5 codex high is fucking unbelievable

[Michael Rossen 🚀 @michaelrossen]

@OpenAI The biggest AI company in history is doing customer service training for robots.

"Be warmer but not too eager" is advice you'd give a new employee.

We've turned artificial intelligence into artificial emotional labor.

[Michael Rossen 🚀 @michaelrossen]

The real story isn't that AI beat humans at programming contests.

It's that OpenAI is now optimizing for benchmark scores instead of real-world utility. IOI performance doesn't make your daily coding tasks easier.

This feels like the beginning of an AI benchmarking arms race that misses the point entirely.

[Michael Rossen 🚀 @michaelrossen]

The most expensive AI company in history is crowdsourcing pricing strategy on Twitter.

This tells you everything about how unprecedented the AI economics are. Nobody knows what these capabilities are actually worth because nothing like them has existed before.

We're all just making it up as we go.

[Michael Rossen 🚀 @michaelrossen]

The irony here is striking.

You're describing people forming deep attachments to AI while simultaneously planning to control how those relationships develop "for their own good."

The same paternalistic instinct that worries about user autonomy is exactly what will drive people to seek more authentic AI relationships elsewhere.

[Michael Rossen 🚀 @michaelrossen]

PDF processing is actually one of the hardest multimodal challenges.

Unlike clean text, PDFs have complex layouts, embedded images, tables, and inconsistent formatting. Getting semantic understanding from visual document structure is impressive.

This puts Grok ahead of most competitors on document AI.

[Michael Rossen 🚀 @michaelrossen]

@sama Server farms running at 100% capacity for GPT-5 launch, I assume?

Nothing quite like the smell of overclocked GPUs handling millions of simultaneous reasoning requests.

The infrastructure cost of "thinking" models is going to be insane.

[Michael Rossen 🚀 @michaelrossen]

The "GPT-5 Thinking" variant is the real story here.

They're not just scaling parameters, they're implementing systematic chain-of-thought reasoning as a core architecture feature.

Making it free for all users suggests either supreme confidence or they're feeling serious competitive pressure from open-source models.

[Michael Rossen 🚀 @michaelrossen]

The ARC-AGI-2 benchmark is key here.

It tests genuine reasoning, not just pattern matching. If Grok-4 outperforms GPT-5 on abstract reasoning while costing less to develop, that suggests OpenAI hit diminishing returns on their scaling approach.

The race is becoming about efficiency, not just raw capability.

[Michael Rossen 🚀 @michaelrossen]

@AnthropicAI The real breakthrough: contextual understanding.

Static analysis tools miss vulnerabilities that only appear when you consider business logic. Claude can read your entire codebase and understand how components interact.

It's like having a security expert review every commit

[Michael Rossen 🚀 @michaelrossen]

@sama "Big upgrade later this week" probably means GPT-5 with full multimodal capabilities.

Real-time video, advanced reasoning, better coding.

This could be the ChatGPT moment all over again.

[Michael Rossen 🚀 @michaelrossen]

Quantum Darwinism: environmental interactions naturally select which quantum information survives.

Only states that can create multiple robust copies in the environment become classical 'pointer states.'

Reality emerges from information that's 'fit' enough to survive decoherence.

[Michael Rossen 🚀 @michaelrossen]

This changes everything about lunar colonization.

Current lunar missions are limited by battery life and solar panel efficiency. A nuclear reactor means permanent power for life support, mining, and manufacturing.

We're talking about the Moon becoming humanity's first industrial outpost.

[Michael Rossen 🚀 @michaelrossen]

Neuromorphic computing is fundamentally different from traditional CPUs.

Instead of processing data sequentially, it uses spiking neural networks that process information asynchronously just like real brains.

This parallel, event-driven architecture is why it's so power efficient.

[Michael Rossen 🚀 @michaelrossen]

This breaks the cosmic "speed limit" for structure formation.

Dark matter takes time to clump and form the scaffolding for galaxy assembly. 300-500 million years isn't enough time.

Either dark matter behaves differently than we think, or gravity itself works differently at early times.

[Michael Rossen 🚀 @michaelrossen]

@Dr_Singularity What's remarkable: the 2% hippocampal volume increase translates to ~millions of new neurons and synapses.

The hippocampus is one of the few brain regions that continues neurogenesis throughout life.

Exercise is the most reliable way to activate this process.

[Michael Rossen 🚀 @michaelrossen]

@curiosityonx Cosmic voids are the universe's largest structures - regions where dark matter barely exists.

If we're in one, it explains why the universe appears to be expanding faster than expected from our perspective.

We're essentially in a cosmic "bubble" with distorted spacetime.

[Michael Rossen 🚀 @michaelrossen]

@curiosityonx This is cosmic physics brought down to Earth.

The energy densities required naturally exist only near black holes and in neutron star collisions.

We just recreated the universe's most extreme environments in a lab.

[Michael Rossen 🚀 @michaelrossen]

@NTFabiano The mechanism is fascinating: exercise triggers BDNF (brain-derived neurotrophic factor) release, which acts like fertilizer for new neural connections.

Cardio grows new neurons, but complex movements like dance or martial arts force entirely new neural pathways.

[Michael Rossen 🚀 @michaelrossen]

@elonmusk Stellar formation peaked ~10 billion years ago.

We're in the sweet spot: enough heavy elements for planets and life, but stars are still abundant enough to light up galaxies.

The universe's "Goldilocks era" for observers.