Yusi Zh

Adding couple of more features at whattoolkit.com 1. Search and fav most frequently used tools 2. Auto reorder tool based on the usage frequency Again, all client side javascript, no server. Privacy first!

#Github #Copilot #Agent does something makes it a no-brainer for me to use it on my toy website whattoolkit.com. It's so intuitive to see the result with a screenshot. OpenAI's #CodeX lacks such product intuition.

The pretraining is usually self-supervised, which means the labels are created automatically from the inputs. This also means there is no label necessary in pre-train

huggingface.co/learn/llm-cour… Pretraining is the act of training a model from scratch: the weights are randomly initialized, and the training starts without any prior knowledge.

Stop getting single-perspective AI answers! 🛑 🌟 AI CLI = Multiple AI models collaborating: • GPT-4 generates ideas • Claude critiques them • Gemini refines solutions Like having an AI think tank in your terminal! `pipx install ai-cli-chat` #AI #DevTools

Language models work by being trained to predict the probability of a word given the context of surrounding words.

And i think that's totally fine. Just keep trying them out so you will not overhype or underestimate what AI can or cannot do for you today, so you know what to look ahead and expect in the future. Let's all start build that muscle.

two days ago, i am so amazed by what #claudecode can do, and i probably vibe coded it too hard. I build a cli package and that is ready to ship to pypi, but just i went through all the details, especially readme; only then i realize it has hallucinate so much stuff....

But my point is that this is going to be a new technique to acquire and i am very excited about it. The only way you stay ahead as SWE in this wave of AI disruption, is to get your hands dirty, trying these things even if it's all building "demo"-driven project.

what i am trying to say is not about CC is not amazing, trust me IT IS AMAZING. It just poses a new challenge to its user on context management, planning, or simply keep track of things. And yes, i totally believe there xyz tricks you can with with #claudecode to make it easier..

And certain features are built but later removed, except the removal are not complete. i.e. one feature was to move the prev chat context into the next prompt and it is exposed as an option. But later on this behavior becomes default, so whether the option is on/off makes no diff

The cli i'm building involves quite many parameters, or CC thinks it needs them. So while it builds them out, it start to "plan" things that are necessary, such as show timestamp, preserve cli chat history etc. But it lost track of them. i.e. some cli options exists but no affect

I gave a general AI agent 27 minutes to be my candidate sourcing agent. What I really measured wasn’t time—it was attention.linkedin.com/pulse/i-gave-g…

inherent complexity of your task being higher than the agent's capabilities rather than some simple mistake that can be corrected.

A common mistake for people who are new to using agents is that they commit to making an interaction successful, even when an agent's work is veering off track…you should be ok discontinuing that conversation or manually taking over. Sending more messages is more likely a sign

If using AI to aid in debugging, we recommend asking for a list of probable root causes rather than trying to debug and fix everything itself. Then, a human can decide based on their own experience which one is the real root cause.

Adding a new tool to render your Mermaid diagram. Everything is client side rendering, no data sent to any server. Your data is safe ;)