ferryzhou

if technical path is wrong

llm + prompt + tools have issues of

so the question is how do we ensure agent behavior is intended consistently?

So Google deepmind hosted llm chess competition.

Seems it can handle the situation which is not explicitly mentioned in prompts/instructions.

https://medium.com/better-ml/how-to-prompt-a-capable-llm-to-almost-gold-at-imo-2025-39b69e91d296

https://developers.googleblog.com/en/simplify-agent-building-adk-gemini-cli/

https://simonwillison.net/2025/Jul/7/

https://github.com/pig-dot-dev/muscle-mem

https://googleapis.github.io/genai-toolbox/getting-started/introduction/

1. Fix it. Otherwise, the same issue can happen again and again.

数据和代码都是知识编码

Llm breaks the barrier between natural language and computer language

Recent texas flash flooding killed 50+ people

following the last lesson

Given the same prompt, the LLM output could be random

so i want to create ferryzhou.github.io with my programming only posts on ferryzhou.wordpress.com

https://github.com/codelion/openevolve

https://simonwillison.net/2025/Jun/29/agentic-coding/

Tool: essentially a function, input is function args. out

I asked on June 9th that when Google release a product like Claude code: https://ferryzhou.wordpress.com/2025/06/09/a-conversation-on-claude-code/

See My First Open Source AI Generated Library https://share.google/HTMCcwNB9dmMuuBwf

Llm is good at understanding text, human input, prompts, planning, but lack specialized knowledge and capacities that’s not in training data.

Agent programming is a total different way than traditional programming

Last week I created an agent to do bug cleanup

https://youtu.be/Yf_1w00qIKc?si=D-Mrc1FvqHvE8OEf

继续研究claude github agent

claude code就是一个agent

最近几天在外面旅行

YouTube推送了一个原子能博主的视频，关于SQL的，https://youtu.be/IUUpxfa1SSw?si=WgsCMh4yPJ0cctkE

今天又一次觉得这个太重要了

https://github.com/ggtracker/sc2reader

all in division 3

now can beat all races veryhard level

see https://github.com/ferryzhou/sc2_bots/blob/a28d3dc7040e88fde5f31d1c1d8d2cc8cb69b0f1/han/han.py

之前一直想改sc2 bot han 的代码

早上半梦半醒中意识到简化的一种基本方式甚至普遍方式是config

前段时间看过一篇文章

https://github.com/ferryzhou/sc2_bots/tree/main/han

https://github.com/ferryzhou/sc2_bots/tree/main/han

得想个系统性的办法来提升sc2的战力

qin2’s elo dropped below 1500.

qin2 1553 https://aiarena.net/competitions/stats/1913/qin2-sc2-ai-arena-2024-season-3

ImportError: numpy.core.multiarray failed to import

https://github.com/AresSC2/ares-sc2/pull/209

Traceback (most recent call last):
File "/bots/qin/run.py", line 18, in
from bot.main import MyBot
File "/bots/qin/bot/main.py", line 6, in
from ares import AresBot
File "/bots/qin/ares-sc2/src/ares/init.py", line 3, in
from .main import * # noqa: F403 F401
^^^^^^^^^^^^^^^^^^^
File "/bots/qin/ares-sc2/src/ares/main.py", line 6, in
from cython_extensions import cy_unit_pending
File "/bots/qin/cython_extensions/init.py", line 6, in
from . import bootstrap
ImportError: cannot import name 'bootstrap' from partially initialized module 'cython_extensions' (most likely due to a circular import) (/bots/qin/cython_extensions/init.py)

上次手机丢了以后github一直没能login

• Installing jedi (0.19.2): Failed

smile. there’s some evidence that simply smiling can be as effective as antidepressant medications

use it or lose it