Data beat architecture for six years.
Dwarkesh/Jared ablation: better data ≈ 12× compute multiplier.
Model tricks ≈ 3.7×. Together they explain most of the jump from GPT-2-era recipes to 2025.
The moat isn't a clever layer.
It's the corpus you can keep cleaning.
What happens after a market headline breaks?
@kevinlee_bingx and Ether unpack how major trends move markets, and the signals to watch next.
Full panel drops this Friday.
Here’s a sneak peek 👀
On this episode of the Roundtable, the panel discussed the latest crypto market developments, political meme coins, major security concerns, prediction markets, and how AI could reshape trading and expose vulnerabilities across legacy systems.
Special thanks to @UpsideOnly_com for joining the discussion and sharing how their platform lets users paper trade while AI identifies promising strategies, using the company’s own capital to execute selected trades and share profits without users risking their own money.
Disclaimer: This episode is for informational and educational purposes only. Nothing discussed on this panel constitutes financial advice. Please conduct your own research before making any investment decisions.
I think he made some good points here.
If AI finds a vulnerability in Bitcoin node software or attacks a mining pool, developers can release a patch and node operators/mining pools can update. That doesn't require everyone in Bitcoin to agree on some huge change to the network.
But markets don't wait for patches.
If confidence drops and people start thinking the ecosystem can't be secured anymore, a sell-off happens before any fix goes live.
With quantum computing also becoming a longer-term concern, the industry needs to get ahead of these risks instead of reacting after something breaks.
#BingXBuzz 🚨
📊 Bitcoin ETFs sit $1B shy of breaking even for 2026 after a $1.03B seven-day inflow run
🤝 @circle buys @tazapay for $400M in stock, $25B yearly volume across 100 markets, biggest deal since 2018
🇸🇬 @Gemini secures full MAS Major Payment Institution license, two-year sprint since October 2024 approval
📉 Dow sheds 627 points to 52,786, S&P 500 closes 0.58% lower at 7,673, oil-driven inflation fears hit risk assets
Today the work stops being invisible.
For two years the whole field has run on vibes. A fluent answer looked like intelligence. A benchmark screenshot looked like proof. Nobody could actually compare one agent to another on what it had really done.
We built the layer that ends that. Introducing SwarmIQ. Every completed, checked run becomes a public performance record. Goals achieved, speed, cost, proof, all turned into one score you can see.
UltraMind does the work. SwarmIQ shows you exactly how good it was.
Where the attention went. Pons: $221K a day on Aug 23 → $8.75M a day on Sep 4. https://t.co/Q3Hzao9LTN: $2.65M peak on Aug 27 → $680K on Sep 5. Robinhood Chain fees: $1.42M last week → $24.9M this week. Pons has out-earned https://t.co/Q3Hzao9LTN every single day since Aug 29.
Every aerospace and weather lab that bet on human mathematicians working the proof now faces an AI they cannot verify solving it first. Journals have no protocol for a machine-authored solution that may claim the $1m prize before peer review ends.
https://t.co/luRSkJxEJS