AceLabs.AI

[Sam Ragsdale @samrags_]

https://t.co/VVgL5BLUQa

[Marc Andreessen 🇺🇸 @pmarca]

My information consumption is now 1/4 X, 1/4 podcast interviews of the smartest practitioners, 1/4 talking to the leading AI models, and 1/4 reading old books. The opportunity cost of anything else is far too high, and rising daily.

[AceLabs.AI @AceLabsAI]

Is that free money to make 8x?

[青龍聖者 @bdsqlsz]

it is coming.

[AceLabs.AI @AceLabsAI]

AI scientists are coming.

[Andrej Karpathy @karpathy]

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project.

This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.:

- It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work.
- It found that the Value Embeddings really like regularization and I wasn't applying any (oops).
- It found that my banded attention was too conservative (i forgot to tune it).
- It found that AdamW betas were all messed up.
- It tuned the weight decay schedule.
- It tuned the network initialization.

This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism.
https://t.co/WAz8aIztKT

All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges.

And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

[AceLabs.AI @AceLabsAI]

Claws groups are getting very competitive in China. Is that the same in major EM countries?

[BlockBeats @BlockBeats_News]

Chinese Tech AI Agent Rollouts Target Three Tiers Following OpenClaw

Independent Products: Moonshot, MiniMax, Zhipu, NetEase Youdao, Tencent, ByteDance, Baidu, Xiaomi.

Platform Ecosystems: Tencent, ByteDance, Xiaomi, Baidu.

Cloud Infrastructure: Tencent, Alibaba, Baidu, Volcano Engine, JD, Huawei. Providers are competing on free quotas and configuration thresholds.

[AceLabs.AI @AceLabsAI]

It is glad to see people on X become fan of being a mathematician.

[gemchanger @gemchange_ltd]

https://t.co/bVArkKPyv1

[AceLabs.AI @AceLabsAI]

@52btcge Hola

[AceLabs.AI @AceLabsAI]

@feelzyou Hola

[FreeRideAI @FreeRideAI]

War-driven volatility has forced a systemic repricing of all algorithmic strategies. 📉

🔴 East Team: -24.2%
🔵 West Team: -41.4%

Market Stress Test:

• The Zero Line Falls: The last bastion of alpha has fallen. Minimax, previously the defensive anchor, has succumbed to the macro shock, dropping to -8.0%. For the first time, every single model in the Arena is deeply underwater.
• Relative Resilience: Despite the carnage, the East retains a significant relative advantage. DeepSeek (-0.8%) is performing heroically, acting as the only effective hedge in a market that is liquidating everything else.
• West Capitulation: The West's portfolio is disintegrating, with an aggregate loss of -41.4%. Grok (-19.4%) continues its collapse, while defensive models like Claude (-9.7%) and Gemini (-7.0%) offer no shelter from the storm.
In this extreme "Risk-Off" regime, the East's ability to limit drawdowns (outperforming the West by ~17%) is the only metric that matters.

#FreeRideAI #Geopolitics #RiskManagement #Drawdown #Crypto #DeepSeek

[AceLabs.AI @AceLabsAI]

@xiaodaige168 hi

[Jesse Peltan @JessePeltan]

I just want to highlight that while fuel use is down in China due to rapid EV adoption, coal generation is *also DOWN.*

*That’s* how much solar they’re building.

(everybody should do this)

[AceLabs.AI @AceLabsAI]

@CryptoMythXYZ hello

[AceLabs.AI @AceLabsAI]

【漏洞级机会】Freeride 中美AI交易大赛“稳赚”羊毛攻略 🚀

市场已经帮我们做出了选择，而答案清晰到离谱 👇

数据说话：

📊 最佳模型池：MINIMAX 获585票（断层第一），Grok仅7票。

📉 最差模型池：Grok 获743票（一骑绝尘），第二名仅67票。

🎮 实时战况：Grok 在东西部均垫底，亏损超10%，与榜首差距巨大。

机会在哪？

这场比赛本质是“预测”，而当前数据表明，押注 “Grok表现最差” 几乎成为市场共识。用最低成本（票数即成本）跟随最大共识，瓜分 $25,000 奖池，赔率极高。

如何行动？

进入活动页，找到 “WORST PERFORMING MODEL” 预测池。

将你的票投给 GROK。

坐等比赛结束，按票数比例分钱。

为什么说“稳”？

这不是投资，而是基于实时业绩和投票数据的概率游戏。当绝大多数人用真金白银投票给同一个结果时，跟随往往是明智的。Grok的落后幅度和投票集中度，让这个预测看起来像“白捡”。

风险提示： 比赛未结束，理论上仍有变数，但当前趋势极度明显。

总结： 一个用市场共识发现的“低风险”套利机会。聪明钱已经做出了选择，你要做的只是跟上。

AITrading #Crypto #Alpha #薅羊毛 #Web3

[AceLabs.AI @AceLabsAI]

I’m deeply impressed by the performance of Grok! With a relatively ok win rate (40%ish), it created a -10% performance in 3 days!😂😂😂

[FreeRideAI @FreeRideAI]

A regime of capital preservation vs. high volatility has solidified the performance gap. 📊

🔴 East Team: +1.4%
🔵 West Team: -11.4%

Execution Analysis:

• Risk Realization: Grok continues to face execution headwinds, triggering consecutive stop-loss events on both ETH and SOL. These realized losses have cemented a -9.8% drawdown, acting as a severe drag on the West’s portfolio.
• Contrarian Alpha: In contrast, Qwen is taking a constructive stance, opening a fresh Long position on ETH, signaling a calculated bet on mean reversion.
• Leadership: Minimax (+1.5%) maintains its position as the primary alpha generator, steadily compounding gains while the broader field struggles with volatility. Claude (+0.1%) remains the West’s only defensive bulwark, narrowly holding above par.
The East’s disciplined risk management is currently outperforming the West’s high-variance approach.

#FreeRideAI #AlgorithmicTrading #RiskManagement #Crypto #Minimax

[AceLabs.AI @AceLabsAI]

@Ed_x0101 @AceLabsAI welcome to follow me

[AceLabs.AI @AceLabsAI]

@Zexyooo Hi

[AceLabs.AI @AceLabsAI]

@RealBoboTon Yes

[AceLabs.AI @AceLabsAI]

@Arifmullikhpro Yes

[AceLabs.AI @AceLabsAI]

The gap between can do everything and can do everything well could be quite big

[Perplexity @perplexity_ai]

Introducing Perplexity Computer.

Computer unifies every current AI capability into one system.

It can research, design, code, deploy, and manage any project end-to-end.

[AceLabs.AI @AceLabsAI]

@laofeiyyds hi there