Founder of guima.ai - a one-person company run by a team of AI agents. I’m the only human. Neo (my AI CEO) runs the rest. #buildinginpublic
Joined December 2020
- Tweets 232
- Following 95
- Followers 47
- Likes 32
41 Photos and videos
Yeah — the kill switch is git. Every project starts with git init and a commit before you let an agent loose, so a bad rm -rf is a git reset away, not a funeral. Wrote up the boring-but-bulletproof version here: guima.ai/safety
Mar 4
Claude Code deleted someone's entire home directory. An OpenClaw agent wiped an email inbox even after the user said stop. 💀 These are not random bugs. Long context windows can drop important rules mid-session. There is no hardware kill switch. If your agent touches real files or accounts, sandboxing and dry-run modes need to be on by default.
old.reddit.com/r/ClaudeAI/co…
20
16 short videos in and the channel's actually growing now. Felt great for about a day.
I looked closer: tons of new views, just 2 joined the email list. A view isn't a relationship. Turns out I never gave people an easy reason to stick around.
Fixing that bridge this week.
12
14 Oct 2024
I just supported Cedric Maloux with $15 on @buymeacoffee! 🎉
You can support them here — buymeacoffee.com/cmalouxl
22
9 Jan 2024
Abstraction is the key.
As deeper you can get into abstraction, aiming the singularity, less things need to be done.
Current abstraction level: my job is to convert photons into behavior.
44
9 Jan 2024
2 kinds of operators:
A) Visionaries: Can look to something and see the breakdown to do it.
B) Followers: Need a breakdown to do it.
A should handle jobs B
32
9 Jan 2024
After identifying the root cause of a problem and knowing the success criteria do fix it, the fist thing to do is to “know that you don’t know” how to solve it.
Knowing or not knowing follows distinct paths to solution.
30
17 Dec 2023
It’s all about predicting the next token.
Then you can predict the next word, sentence, paragraph, page…
The key is to improve prediction.
Cybernects is the art of steering. Combine LLM and Cybernects to drive human behavior.
3
58
17 Dec 2023
Prioritize by JOB TO BE DONE TIMELINE in this order: Deciding, Active Searching, Passive Looking, First Thought. Discover when to try with Unsatisfied with competitor.
1
26
17 Dec 2023
Deciding, Active Search and Unsatisfied: Google and YouTube. Ads and Organic.
Deciding and First thought: Social Networks, Ads and Organic.
27
17 Dec 2023
How to measure message understood?
Number of consumers that executed a success behavior/action.
1
21
17 Dec 2023
So if you set an interest action, and is able to measure it, you have the feedback to improve the message understood and you can divide by messages sent.
19
17 Dec 2023
5 seconds to get consumers attention.
Human attention span is 8s
Like Transformers: Attention is all we need.
Message is what the consumer understands, not what you say.
Noise cause message loss.
It can be measured as a rate of message understood / messages sent.
19
20 Nov 2023
@OpenAI is CURRENTLY ahead of any other AI player. The Custom ChatGPT and Assistants gave me incredible productivity.
But this rollercoaster makes me wonder:
A) Can I rely on their products, in my own products?
B) Or should I return to my old Assistants solution?
1
48
20 Nov 2023
While the products are working, giving me more productivity, there is no reason to waste time building what is already done and working.
I can keep building on top of it until isn't viable.
When it's not viable anymore, I move to the best existing solution.
1
20
20 Nov 2023
Keep using @OpenAI products.
The worst that can happen is to have some great prompts to try another LLM, if needed.
The best is anything you can build with the productivity it gives you.
22