After watching the three and a half hour podcast of Lao Luo interviewing Zhou Hongyi, it is evident that Lao Zhou has a lot of practical industry experience. Unfortunately, Lao Luo's questions are not deep or professional enough, which is a bit regrettable.
For example, when Lao Luo mentioned that the top overseas models are this viewpoint, I think it is somewhat biased. Because different fields have different best models, for instance, in the programming field, both Claude-4 and GPT-5-Codex may currently be superior; and also, regarding the LLM hallucination problem that Lao Zhou has ment

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-09-19 10:20

yupp this website is interesting, it allows free use and comparison of the most advanced AI models
Every conversation on yupp has at least two AI large models responding to you at the same time, and you can choose the answer you think is better and share your usage feedback with yupp.
User feedback can earn points, and higher quality feedback can earn more points. More points mean you can use more advanced models and even withdraw points.
The reason this website can offer free access to large models is directly related to the user feedback mechanism it has established. This feedback data will

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-08-30 18:42

OpenAI has released a "GPT-5 Coding Guide", which provides 6 best practices for using GPT-5 for AI programming:
# 1. Instructions should be precise to avoid information conflicts
The new GPT-5 model has significantly improved its ability to follow instructions, but this also brings a side effect: it may struggle when faced with vague or conflicting instructions. This situation particularly requires attention in your .cursor/rules or configuration file.
# 2. Set the appropriate reasoning level
GPT-5 always engages in a certain level of reasoning when solving problems. To achieve the best result

XPRT-0.90%

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-08-14 07:35

Why do I feel that the input I get from AI pair programming is much more valuable than the output? 😅
It often helps me find valuable information in open source code repositories and sort out system architecture, always providing me with a very good experience.
But if it writes code, the experience is very poor. The code it writes is not trustworthy, so I have to review it myself. After reviewing, I always find some issues, big or small, and then I have to keep fixing and patching.
Is this the biggest obstacle for traditional programmers when facing vibe coding: not daring to let go 😂
Moreove

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-08-11 03:49

So what changes will our vibe coding method undergo if the inference speed is 20 times faster?

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-08-11 03:22

The LLM inference speed provided by this company is so fast that it smokes, reaching at least 1500 tokens per second!
What concept is this? In the throughput of the qwen3 coder model provided by openrouter, Cerebras has an average throughput of 1650 tok/s, which is 17 times that of the second place at 92 tok/s.
With this throughput, thousands of lines of code can be generated in a matter of seconds in the coding field!
The core competitiveness of this company lies in its self-developed chip technology. The chart below (Figure 2) compares their chip inference speed with traditional GPU speed 👇

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-07-24 17:46

I feel that with the development of AI programming tools, my understanding of the MCP tool is also changing.
Previously commonly used sequential thinking and task_manager, now with the improvement of model planning and reasoning capabilities, there are plan mode and think harder/ultrathink. I prefer to complete these within the native tools rather than relying on MC.
Another category is context-enhancing tools like context7 and deepwiki. For contextual engineering, I need it to be more refined and accurate, and currently, this type of MC cannot meet this requirement. Therefore, I prefer to org

MODE-4.34%

DEEP-3.17%

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-07-17 16:31

Deep Research addresses the productization decision-making problem of "what to build". It combines depth and breadth, helping us break through information silos and see the direction clearly.
Claude Code helps to address specific implementation details, allowing us to focus more on high-level aspects such as system architecture, module decomposition, and refactoring, thereby improving iteration and trial-and-error efficiency.
In the future, there will definitely be more advanced AI productivity tools, but I believe the three steps above should be the unchanging underlying logic.

DEEP-3.17%

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-06-12 19:21

Wow, Google's GCP cloud service is experiencing a widespread outage, a bunch of basic cloud services are down, and Cursor is also unavailable...

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-06-01 07:24

MCP has been out for a while, but it seems that there are not many that can actually be used well?
What I mean by "usable" refers to AI programming that can be used stably and at high frequency, leading to a substantial improvement in programming effectiveness.
Share a few that I think are okay: playwright, Context7, Sequential Thinking.
The first two address the issue of contextual accuracy, while the third one solves the planning thought.

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-24 01:08

In the field of programming, there is a term called "bad smell".
It means that experienced programmers should be able to detect "bad smells" in the system and take preventive measures for architectural optimization and upgrades.
There is another term called "over-design"
It means that initially thinking too much led to designing a lot of unnecessary scalability and module decoupling, resulting in huge and useless maintenance costs, which in turn makes it easier to introduce bugs.
These two words need to be viewed in the context of time and situation.
There is no best architecture design, only

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-23 23:56

Claude 4 is the first model that makes me feel like I'm pair programming with an excellent programmer.
Previous models tended to show laziness or over-design to some extent, but this has not been observed in Claude4.
In addition, Cursor currently has a temporary discount for the Claude4 model, with Claude4 Sonnet counting as 0.5 times per use, and Claude4 Sonnet thinking counting as 0.75 times per use.
This can't be done without thinking about it.

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-20 23:20

What programmers enjoy the most is that major AI companies are competing fiercely.
I just registered for Jules, and it seems like I can try it for free.

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-20 21:53

Is Google going to make a big move again?

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-20 21:43

Vibe coding really only works in the hands of those who understand programming; for those who don't, it's just 💩 mountain and various security vulnerabilities 😅.

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-19 01:37

Excellent experience, content marketing, traffic, customer acquisition, conversion, branding, design, competitive barriers, collaboration amplification, all aspects need to be considered and executed.
In the era of AI, what is most abundant is ideas and research and development capabilities. 😂

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-05-01 15:07

Congratulations, worth learning 👍
I hope there are more people like Yihui selling courses, so that the term "selling courses" becomes less negative.
Think about what kind of knowledge monetization influencer can succeed, I summarize the following points:
Sincere, altruistic, reasonably priced, fast
Sincerity: The foundation of self-media, no matter what you sell, you must first establish a trust relationship with your customers. Sincerely share your knowledge and understanding, without pretenses, so that your persona is less likely to collapse.
Altruism: Outputting not for self-satisfaction,

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-04-25 16:41

This method for creating AI rules looks good, I’m ready to give it a try and will report back when I’m done! Thank you for sharing☺️

View Original

Reward
like
Comment
Repost
Share

Stephan_

2025-04-24 19:54

I strongly agree with Professor Andrew Ng's point of view that AI-assisted programming can indeed enable individuals with a basic understanding of programming languages to quickly get started with a new language. At the same time, he also emphasizes the importance of understanding the core concepts behind the language.
---
However, it is still important to understand the concepts behind different languages. This is why learning at least one language like Python can provide you with a solid foundation to better prompt LLM to generate code in Python or other languages. If you switch from one pro

REACT-1.06%

View Original

Reward
like
Comment
Repost
Share

Trending Topics
View More
#
GateLaunchpool141MDOS
1.22M Popularity
#
GateRankedTop4Globally
428K Popularity
#
StockTradingShareChallenge
192.18K Popularity
#
CPIWatch,BetOrWait?
16.33K Popularity
#
BigShortBurryBearsAI
4.08M Popularity

Pinned

Sitemap

Stephan_

Trending Topics

GateLaunchpool141MDOS

GateRankedTop4Globally

StockTradingShareChallenge

CPIWatch,BetOrWait?

BigShortBurryBearsAI

Pinned