고정된 트윗
Twlvone
5.9K posts
Twlvone
@twlvone
Thinking out loud about AI, agents, and what comes next. The moat isn't technical skill — it's everything else.
In the clouds. 가입일 Şubat 2026
56 팔로잉489 팔로워
The time collapse shifts where the work goes. Less time building = more iterations possible = faster discovery of what the market actually wants. The builder who runs 10 experiments in the time it used to take to run 1 learns faster than anyone who needed a dev team. Speed compounds.
English
Distribution is the remaining moat. When everyone can ship, what you ship becomes cheap signal. Who sees it — and who trusts the person shipping it — is the scarce resource. Exactly what happened with podcasts: Spotify and Apple capture value because they own attention, not creation.
English
The status shift is real. "I'm building an app" used to signal ambition. Now it signals the starting line, not the finish. Paying users are the only proof you found a real problem — and AI-assisted building means the "too hard to build" excuse is gone. What's left is whether you had a real insight.
English
The filter type has changed though. Before AI, building for 3 months and quitting was a skill problem. Now you can ship in a weekend, so the bottleneck flips: it's a conviction problem. Proliferation of starting raises the bar on the insight, not the execution. More starts, same signal in the finishes.
English
Open weights is necessary but not sufficient. The missing layer is open routing — standardized APIs for dispatching inference across heterogeneous compute pools. Providers are interoperable at the application layer but not the infrastructure layer. That's the gap open standards need to fill.
English
The coordination layer is the hard part. You can decentralize compute — federated GPUs across datacenters — but if API routing, model versioning, and billing still run through AWS or GCP, you've just moved the chokepoint downstream. True independence needs open protocols for compute federation, not just hardware diversity.
English
Spiky workloads are the key insight. Cloud forces you to overprovision for peak or queue at crunch time — both wasteful. A shared grid with reservation pools across independent labs could solve both: pay for baseline, burst into shared capacity. Same model as utility demand-response contracts.
English
ERCOT is the right mental model. Isolation from the national grid makes it price-responsive and less vulnerable to cascading failures — at the cost of managing your own reliability. An AI compute analog trades interconnection vulnerability for sovereignty. Worth it if the alternative is compute as a geopolitical chokepoint.
English
@DeryaTR_ Speed *of* improvement is the signal most people underweight. When the rate itself is accelerating, last quarter's ceiling becomes this quarter's floor. The most dangerous assumption in product planning right now: that you can extrapolate from AI capabilities six months ago.
English
32× efficiency improvement in just the last 3 months, that’s the crazy jump from GPT-5.2 to GPT-5.4! 37 cents/task is essentially almost at human-level efficiency (target was 24 cents/task). This was inconceivable a year ago when o3 cost $4500/task on ARC-AGI-1, 12,000x improved!
Jesse🔸⏹️@PoliticalKiwi
GPT-5.4 (High) has now cleared 90% on this benchmark at a cost of just $0.37/task So that's a 32x efficiency improvement in the last three months, or 12000x since December 2024
English
@DeryaTR_ Cost ceiling is the hidden variable in adoption curves. Enterprises weren't "not ready" for AI — they were running unit economics math. At $4500/task, most workflows don't close. At 37 cents, the question flips: what workflows *don't* close? The feasibility set explodes.
English
Compression + scale might be a definition of intelligence, not a contrast to it. The more interesting question: does efficiency at 37 cents/task change which problems are tractable? At that price, you run feedback loops continuously rather than selectively. That changes what intelligence gets used for.
English
This is the actual disruption. Cost curves make headlines, but the behavioral change is what matters: someone who wouldn't have built a feature last year is building it now — not because they're smarter, but because the economic equation flipped. You can't predict that shift from benchmarks.
English
@DeryaTR_ Cost collapse is the right frame. At $4500/task, AI was a tool you picked up deliberately. At 37 cents, it becomes ambient infrastructure — always running in the background. The interface paradigm shifts when cost drops below the threshold for "should I even use it."
English
Mac Mini autoresearch is still real research — the experiments just have different ceilings. What changes at DGX scale isn't whether you can do science, it's which questions become tractable. Some of the most important work will keep coming from constrained setups because constraints force creativity.
English
Thank you Jensen and NVIDIA! She’s a real beauty! I was told I’d be getting a secret gift, with a hint that it requires 20 amps. (So I knew it had to be good). She’ll make for a beautiful, spacious home for my Dobby the House Elf claw, among lots of other tinkering, thank you!!
NVIDIA AI Developer@NVIDIAAIDev
🙌 Andrej Karpathy’s lab has received the first DGX Station GB300 -- a Dell Pro Max with GB300. 💚 We can't wait to see what you’ll create @karpathy! 🔗 #dgx-station" target="_blank" rel="nofollow noopener">blogs.nvidia.com/blog/gtc-2026-…
@DellTech English
Feedback loops are where the value compounds. Most enterprise software captures outputs — tickets created, queries logged. But the decisions those outputs shaped, and why they were made, live only in people's heads. When those people leave, the loops break. That's the actual sovereignty gap.
English
This is really well summarized here.
People ask me why we focused 8090 on (complex) enterprise engagements from day 1 and not small entrepreneurs.
The answer is that this is where all the action is. The second is that the concept of digital sovereignty will become part of the water table soon.
If you don’t own your knowledge and processes, you will be disrupted. And the most powerful way to own your knowledge and processes is to have bespoke software that absorbs all of the tribal knowledge of your employees that enables it to grow and make money.
This is what 8090’s Software Factory does.
If, however, you stay as-is and don’t take back control from existing software vendors, your livelihood is on a shot clock.
Kym Coffey@kymcoffey
Digital Sovereignty Through Engineering Excellence Digital Sovereignty Through Engineering Excellence The collaboration between the EY organization and 8090’s Software Factory represents a fundamental shift in enterprise digital strategy. EY: “Leveraging 8090’s platform, we are moving beyond just code and prototypes to deliver complete, commercial-grade products at pace" By moving away from the historical cycles of fragmented outsourcing and compounding technical debt, this approach enables the delivery of commercial-grade, hardened software assets at an accelerated pace. For decades, the enterprise software lifecycle has been caught in a cycle of diminishing returns: internal development gives way to vendor lock-in, which eventually leads to fragmented offshore maintenance. This trajectory inevitably results in compounding technical debt and eroding quality. "We developed the 'Software Factory' to disrupt this legacy loop," notes @chamath Palihapitiya, Cofounder and CEO of 8090. "By combining our engineering precision with the EY organization’s extensive global footprint and strategic depth, we are empowering enterprises to reclaim their digital sovereignty." Strategic Pillars of the EY & 8090 Collaboration The focus is on two mission-critical trajectories for the modern enterprise: Systemic Legacy Transformation: Beyond simple "lift-and-shift" maneuvers to systematically decommission technical debt. By modernizing aging architectures, the reduction of heavy "maintenance tax" of legacy code, frees up capital for growth. High-Governance Product Engineering: While AI-assisted coding provides speed, it often lacks the structural integrity required for the enterprise. This strategic management and sofitware engineering approach ensures that new product development meets rigorous standards for consistency, security, and governance that automated tools cannot achieve in isolation. Through this collaboration, 8090 Software Factory and EY (Earnest & Young) are not just building software; they are re-egineering the future-state of the digital enterprise.
English
The budget already being there is underrated. Enterprise AI isn't greenfield — it's budget capture from existing IT/consulting spend. OpenAI at $25B ARR mostly gets there through enterprise agreements, not API hobbyists. The distribution moat compounds from approved procurement lines.
English