Dylan

generational lock-in

[Dylan @xbtDylan]

💀💀💀

[Elon Musk @elonmusk]

@cognition Grok 4.7 will exceed all current models.

That said, Anthropic is a great company and will probably release improved models soon.

However, the SpaceX training corpus is so awesome & unique that I would be shocked if any model is better at real-world engineering than 4.7.

[Dylan @xbtDylan]

poteto poteto poteto
I'm using Grok Build with poteto-mode.
I write a detailed prompt, paste in screenshots, but they disappear the moment I send it
Why is that happening?

[Dylan @xbtDylan]

wanna praise inference speed in grok build. got noticeably faster this week

[Dylan @xbtDylan]

I don't think they're even shipping grok 4.7 🥲

running out of time to fix it so they might as well just skip to 4.8...

[leo 🐾 @synthwavedd]

Unfortunately, it looks like Grok 4.7 won't be anything crazy.

While it seems once again very token efficient - and it'll be cheap - it gets cooked by the upcoming new Opus in my testing, and it's not all that competitive.

Grok 4.7 (xhigh) left, Opus-Next (max) right.

[Dylan @xbtDylan]

cmon @elonmusk release that beast (grok 4.7)

[Dylan @xbtDylan]

@benvargas @cursor_ai @poteto

[Dylan @xbtDylan]

@artficialisabel Peak 🫳

[Dylan @xbtDylan]

ok its time to lock in

[Dylan @xbtDylan]

@austinyuhao ts is just a pretext to ban models from chinese labs

[Kimi.ai @Kimi_Moonshot]

Meet Kimi K3

[Dylan @xbtDylan]

@synthwavedd imagine if it ends up being better and cheaper than opus 5

[Dylan @xbtDylan]

gpt 5.6 is amazing

[Dylan @xbtDylan]

@jietang vision

[Dylan @xbtDylan]

@PaulSkallas but what about the brain

[Dylan @xbtDylan]

But please not the way Anthropic is doing — don’t make limits burn faster during peak hours

[Dylan @xbtDylan]

One idea could be to add a status bar that shows users the current load level in real time. During peak hours, you could make inference a bit slower. This might encourage some users to postpone their tasks to quieter hours, which would help flatten the demand curve a bit.

[Tibo @thsottiaux]

With Codex the there is quite the gulf in load between peak and off-peak times, and we would like to achieve more of a smoother traffic pattern as that would be a more optimal use of our compute.

We have ideas, but curious what you all think we should do? Would more usage during off-peak and surge multiplier during peak times make sense?

[Star_OKX @star_okx]

No complexity. No accident.
10/10 was caused by irresponsible marketing campaigns by certain companies.

On October 10, tens of billions of dollars were liquidated. As CEO of OKX, we observed clearly that the crypto market’s microstructure fundamentally changed after that day.

Many industry participants believe the damage was more severe than the FTX collapse. Since then, there has been extensive discussion about why it happened and how to prevent a recurrence. The root causes are not difficult to identify.

⸻

What actually happened
1.Binance launched a temporary user-acquisition campaign offering 12% APY on USDe, while allowing USDe to be used as collateral with the same treatment as USDT and USDC, and without effective limits.
2.USDe is a tokenized hedge fund product.
Ethena raises capital via a so-called “stablecoin,” deploys it into index arbitrage and algorithmic trading strategies, and tokenizes the resulting fund. The token can then be deposited on exchanges to earn yield.
3.USDe is fundamentally different from products such as
BlackRock BUIDL and Franklin Templeton BENJI, which are tokenized money market funds with low-risk profiles.
USDe, by contrast, embeds hedge-fund-level risk. This difference is structural, not cosmetic.
4.Binance users were encouraged to convert USDT and USDC into USDe to earn attractive yields, without sufficient emphasis on the underlying risks. From a user’s perspective, trading with USDe appeared no different from trading with traditional stablecoins—while the actual risk profile was materially higher.
5.Risk escalated further as users:
•converted USDT/USDC into USDe,
•used USDe as collateral to borrow USDT,
•converted the borrowed USDT back into USDe,
•and repeated the cycle.
This leverage loop produced artificial APYs of 24%, 36%, and even 70%+, widely perceived as “low risk” simply because they were offered by a major platform. Systemic risk accumulated rapidly across the global crypto market.
https://t.co/IK2gW4xUOP that point, even a small market shock was sufficient to trigger a collapse.
When volatility hit, USDe depegged quickly. Cascading liquidations followed, and weaknesses in risk management around assets such as WETH and BNSOL further amplified the crash. Some tokens briefly traded near zero.
The damage to global users and companies—including OKX customers—was severe, and recovery will take time.

⸻

Why this matters

I am discussing the root cause, not assigning blame or launching an attack on Binance. Speaking openly about systemic risks is sometimes uncomfortable, but it is necessary if the industry is to mature responsibly.

I expect there may be significant misinformation and coordinated FUD directed at OKX in the near future. Even so, speaking honestly about systemic risk is the right thing to do—and we will continue to do so.

As the largest global platform, Binance has outsized influence—and corresponding responsibility—as an industry leader. Long-term trust in crypto cannot be built on short-term yield games, excessive leverage, or marketing practices that obscure risk.

The industry needs leaders who prioritize market stability, transparency, and responsible innovation—not a winner-take-all mentality where criticism is treated as hostility.

Crypto is still early.
What we choose to normalize today will determine whether this industry earns lasting trust—or repeats the same mistakes again.

[Dylan @xbtDylan]

@BrianRoemmele @flowith LFG, Looking forward for it!

[Matthew Carrigan @carrigmat]

Complete hardware + software setup for running Deepseek-R1 locally. The actual model, no distillations, and Q8 quantization for full quality. Total cost, $6,000. All download and part links below: