链研社

Age 8.3 Yıl

Peak Tier 4

No content yet

链研社

22h ago

Is the awesome HappyHorse from Seedance 2.0 actually open source?
The person behind HappyHorse is Zhang Di from Taotian Group. He was previously the first key person at Kuaishou Kling, and before that, he was the head of big data and machine learning engineering architecture at Alibaba's Alimama. He worked at Alibaba for 10 years and was re-hired by his former employer last year.
However, you know how Alibaba's public relations hype goes; the actual experience might be a bit different. Compared to Seedance 2.0, matching Kling is still possible. The key point is that it's open source—what m

View Original

2 Likes

Reward
2
Comment
Repost
Share

链研社

04-07 03:42

The cheating distillation path of China's large models is about to be jointly shut down by the overseas Big Three. The good days for domestic large models are coming to an end. Let's see who will reveal their true colors first 😂
Kimi, Minimax, DeepSeek are all on the distillation blacklist, but GLM and Qwen were not named.

View Original

Reward
like
Comment
Repost
Share

链研社

04-06 14:36

Qwen3.6-Plus, based on my actual testing, clearly outperforms MiniMax M2.7 and is more multimodal. However, this time there aren't as many AI bloggers giving glowing reviews because Alibaba has cut off their supplies.
Probably because Alibaba's recent financial report issues led to a cut in their promotional budget. Honestly, that's the right move. They should have relied on real strength to win. When Qwen3.5 was released, they poured a lot of budget into hype and exaggerated claims, but the actual performance was a complete disaster.
Alibaba's current strategy has shifted to:
- Small models,

View Original

Reward
like
Comment
Repost
Share

链研社

04-06 10:48

Chinese large models are increasingly shifting toward semi-open source. The latest released large models are not immediately open-sourced, and those that promised to open source have not done so yet. This indicates that the industry has moved past the stage of using open source to gain attention and has entered a commercial phase focused on using closed source to maximize profits.
GLM 5.1 was not open sourced immediately.
MiniMax 2.7 was not open sourced.
Qwen 3.6 Plus also has not been open sourced.
Now, the trend is basically toward delayed open source or even closed source. Immediat

View Original

2 Likes

Reward
2
Comment
Repost
Share

链研社

04-04 14:35

I developed an options tool for Deribit, making it easier to sell options for bottom fishing. Instead of trading on the exchange's dual-currency wealth management products, why not do it yourself? Built using Deribit's open API, it helps find suitable options to sell, filtering for appropriate APR and good liquidity. It uses data to roughly assess whether current volatility is high or low, when selling options is profitable, and the smart money's direction.
GitHub repository: lianyanshe-ai/deribit-options-monitor
It monitors the BTC/ETH options market on Deribit and the design ideas behind the

BTC0,87%

ETH-0,03%

View Original

Reward
like
Comment
Repost
Share

链研社

04-03 14:55

The hardest money to make in investing isn't from news, nor from emotions; it's from the expectation gap.
The so-called expectation gap means your understanding of a company is earlier and more accurate than the market's. Changes that the market hasn't yet seen, you spot in advance; logic that the market hasn't accepted, you accept early. When the market gradually reacts, the stock price begins to reflect your insight.
There are mainly three types of expectation gaps.
The first is performance expectation gap.
The market originally expected the company to earn 1 billion, but through res

View Original

2 Likes

Reward
2
Comment
Repost
Share

链研社

04-01 13:59

OpenAI said they want to create an integrated super app for desktop that combines ChatGPT, the coding platform Codex, and more.
Then I found that Doubao's desktop version is even more of a mess—chat, browser, image generation, videos—all of these are included, and it's been like that for at least half a year already. 😂

View Original

Reward
like
Comment
Repost
Share

链研社

03-30 15:29

After Trump’s reckless actions, the Federal Reserve was the first to loosen its stance.
Federal Reserve Chair Powell: Inflation expectations seem to have stabilized.
Traders have canceled their expectations of Fed rate hikes and are now considering the possibility of rate cuts this year.

View Original

Reward
like
Comment
Repost
Share

链研社

03-26 15:57

Xiaomi's trillion-parameter large model MiMo-v2pro continues to be available for free trial for another week, teaching you how to quickly get started and experience it, supported on OpenClaw for free use.
Now is really a good time to try it out for free; after testing, it feels like the top tier of domestic models, which is not an exaggeration. The official defines it as a flagship model aimed at agentic workflows: with a total parameter size exceeding 1 trillion, supporting around 1 million context length, with a focus on enhancing usability in complex process orchestration, engineering tas

View Original

Reward
like
Comment
Repost
Share

链研社

03-26 04:54

Reposted from Lonely Brain
Language is humanity's greatest invention and also humanity's biggest bottleneck.
When large language models demolish the Tower of Babel, what exactly will they bring?
Humans built civilization through language, formed societies, and developed commerce. Humans monopolized language, and friction and misunderstanding between languages gave birth to most job positions.
Large models arrived, and in an instant, reduced language friction.
One: The essence of an enterprise is a language organization. What truly gets restructured on the B2B side is the enterprise as a langua

View Original

1 Likes

Reward
1
Comment
Repost
Share

链研社

03-25 08:03

If I were to recommend 3 must-have Claude Code plugins or skills, I would choose:
**1. Code Execution Environment**
This allows Claude to run and test code in real-time, debug errors immediately, and verify solutions work before delivery. Extremely valuable for development efficiency.
**2. API Documentation & Reference Library**
Quick access to relevant documentation for popular frameworks, libraries, and APIs. Saves time on lookups and helps generate more accurate, production-ready code.
**3. Git Integration & Version Control**
Seamless integration with Git repositories for tracking changes,

View Original

1 Likes

Reward
1
Comment
Repost
Share

链研社

03-24 12:11

Are the PUA tactics from big tech companies really like this? Has anyone come forward with personal accounts?
In comparison, PUA in crypto circles is even more destructive. They manipulate you when asking for data, and once they have it, they kick you to the curb.

View Original

Reward
like
Comment
Repost
Share

链研社

03-24 03:18

In long-term value investing operations, the approach is:
1. Market pessimism is the norm, but you should choose to be an optimist against a pessimistic backdrop.
2. Always keep cash on hand. Since you don't know what tomorrow will bring, cash is the best bullish option.
3. Even for good companies, you need to consider price, valuation, and timing. Market value reversion is a rare occurrence. When you see a good company, you can choose to buy blindly without timing, but understand that the probability of further appreciation from reasonable valuation is far smaller than the probability of decl

View Original

2 Likes

Reward
2
Comment
Repost
Share

链研社

03-23 07:15

Trump's Three Dark Arts
I. Never admit you're wrong
II. Even if you lose, never concede defeat
III. Even if you're truly wrong, strike back at those who attack you
On the battlefield of power, controlling the narrative matters far more than possessing the truth
Others have summarized three even more incisive points: sophistry + shamelessness + keep going

View Original

Reward
like
Comment
Repost
Share

链研社

03-22 08:29

I followed the tutorial as a reference and configured it according to Lightning's best practices; the entire process takes less than 10 minutes.

View Original

1 Likes

Reward
1
Comment
Repost
Share

链研社

03-22 08:24

# Translation:
No need to use Typeless anymore, and paid membership is a waste of money. Lightning Talk + Douyin Voice Model completely outperforms Typeless in terms of recognition speed, accuracy, and features.
The speech recognition model comes with 20 hours of recognition quota, and after that it's just 1 yuan per hour. The voice assistant large model also includes 500k tokens for free, which is more than enough for a long time. Pay-as-you-go pricing at 3.6 yuan per million tokens.
In the AI era, you should try learning to assemble some software tools yourself.

View Original

2 Likes

Reward
2
Comment
Repost
Share

链研社

03-22 05:50

I have successfully integrated the local openclaw!
Launched a Clawbot plugin that can connect to the local openclaw, making daily use very convenient now.
1. Update your phone to the latest version 8.0.70.
2. On the computer where openclaw is installed, open the terminal and run the command: npx -y @tencent-weixin/openclaw-weixin-cli@latest install.
3. If you encounter any issues during installation, have openclaw or Claude code handle it.
4. Open the terminal: openclaw channels login --channel openclaw-weixin.
5. Log in by scanning the QR code.
6. Your chat list will now include

View Original