Ming M

Dog person/Cat person
PhD in physics with a focus in AI
Neurodivergent builder
👇👇Check this MEOWTASTIC BEATS

[Ming M @MingM35169]

I've been using codex and claude code for a long time and i would take this as a user comment opportunity. codex is definitely more powerful in implementation with a user given implementation plan. Codex has less content drift and takes extra caution when editing my code while claude is more likely to silently forget or misunderstand my need even with my instructions documented in several organized markdown files. However, I feel claude is more helpful in brainstorming step and it is good at finding the alternative ideas and potential deficiencies. In short, I think codex may need to work more on reasoning the areas that user does not mention. I can not be an expert of every field and I am expecting some constructive advice from AI, not just getting some minor polish of my own idea, but reasoning out the box

[Ming M @MingM35169]

the condescending attitude surprised me. I don't get why anthropic tuned fabe 5 to talk in this tone. On the one hand, anthropic warns users that claude is merely an AI assistant and can make mistakes; on the other hand, it makes Fable talks in a super confident and sometimes condescending way

[Ming M @MingM35169]

@claudeai I feel Anthropic's tactics in this AI competition come off as unseemly. When gpt5.6 was released, openAI was like: "hope it can assist your important work ". Meanwhile, claude was like:" I won't let users to access my new model without more payment unless i have no other choice"

[Ming M @MingM35169]

I haven't experienced gpt5.6 yet but i have both claude and codex subscription and used both of them for coding. My experience is that codex is better at implementation (gpt5.5)even compared to fable.
Claude tends to forget and miss things more easily then gpt and it is probably because of the difference in context engineering. One typically needs to give more guidelines to claude code then for gpt for a complex task. But claude is better at brainstorming and idea generating. Especially, claude tends to talk in a more "confident" tone which makes it looks more trustworthy.

[Ming M @MingM35169]

@johnennis i totally agree. I know this can be avoided, but anthropic's intention here is offensive. I paid to have it assisting my work under my instruction but by default, it claims co-ownership of my work. Maybe everyone should start to claim co-ownership of the company we work for.

[Ming M @MingM35169]

@AnthropicAI maybe tomorrow we would consider restrict access to high school education. what if people learn biology and chemistry from high school classes and do bad things with it? you clearly would agree with this idea, wouldn't you?

[Ming M @MingM35169]

My experience with fable 5
I saw a lot of comments saying that fabel 5 is on the next level. so i tried it today immediately after it became available, and just to find out that my account was blocked from fable 5 because i had some previous research work related to computational biology.
This is ridiculous. How powerful it could be so that it brings harm to the user or becomes dangerous weapon to the society if users are working in biology or chemistry related field?
Since i could not test it with some serious work, I then opened up a chat window and asked it to make me a game. It built an interactive novel styled game. The logic for this type of game is very straightward:user choose an option and the game proceeds to the next stage. The only hard point that requires some intelligence is on the mechanisms of saving/retrieving progress, and the record of endings. The task is simple, but we still go back and forth 4 times to fix the bugs with fable 5 extra effort. I don't mean this is bad experience. It is just average. One could expect similar experience with opus and gpt5.5.
So at the end, I still don't see how fabel 5 is on the next level rather than being an incremental advance. If anyone has the chance to use it with some serious work, could you share your experience on how it is compared with gpt5.5 or opus

[Ming M @MingM35169]

https://t.co/NLHmR3u6ni

[Ming M @MingM35169]

I build an end-to-end workflow that can assist research work.

[Ming M @MingM35169]

if you always have a hard time writing a paper, then check this out!
This skill suite is like giving you a quick start of a research paper
Open source on github with repo name: mmTheBest
Research-Architect

[Ming M @MingM35169]

https://t.co/NLHmR3u6ni

[Ming M @MingM35169]

One skill suite solves all your paper writing issues
Check this out on github

[Ming M @MingM35169]

https://t.co/NLHmR3u6ni

[Ming M @MingM35169]

If you don't know how to start a research study with a vague idea, try this out: the Research-Architect skill suite. Install it into your codex or claude, it automatically handles idea refinement,literature review, experiments design, and scientific proposal drafting

[Ming M @MingM35169]

Why “Secure Local AI Computers” Change Agent Architecture: Capability Gating, Audit Trails, and Human-in-the-Loop Control Planes https://t.co/091QWbRXuS

[Ming M @MingM35169]

Getting a mac mini for openclaw is trending.
But wait, do you really need one?
check on this to find out

[Ming M @MingM35169]

version 0.1 was already shipped. It covers the basic functions like literature review

[Ming M @MingM35169]

I'm building this to assist my own research project and also building it as a live example to show how to make a make an AI product

[Ming M @MingM35169]

Being angry about a coding agent seems to be stupid though

[Ming M @MingM35169]

I don't know if anyone is noticing this. The latest update of openclaw is putting so much constraints and it no longer functions as an assistant, but as a chat bot . I modified the config to give it full permission, but still, it stops actively doing things.

[Ming M @MingM35169]

Here is another issue that drives me crazy. openAI seems not properly writing down the memory file. Even small distractions from the main topic can mess up the memory and completely destroy the working pipeline that I set for it.