super crazy and bullish with agent skills here are my skills: https://teapaw.org/t.co/63mO4hcOA7 and here is how I manage skills: https://teapaw.org/t.co/BDdPMwlGBf
Paris, France
Joined March 2011
- Tweets5.1K
- Following662
- Followers1.4K
- Likes3.4K
I dont do the search myself any more
For the last few days, everything on my trip, organize things, update plan, all done by my agent (hermes + opus 5.5 + my skills set specilize for traveling
I am very satisfied with the quality
I guess the EU zone will be a long, long, long, long wait ....
Day 6 of Claude Duet! A few updates on access, and really appreciate everyone’s patience on the waitlist:
- We’re getting all Max users in the US off the waitlist first. Lots of invites went out yesterday and today, with more coming this week.
- Right after Max, we’ll start inviting Pro users in the US.
- For folks outside the US, we’re working hard on extending access and will share a date as soon as we have one.
If you’d like to try Claude Duet and help shape it, please join the waitlist. claude.com/product/claude-du…
We’re building it with and for working professionals, small businesses, and entrepreneurs who are constantly juggling a lot. Team is reading every piece of feedback and shipping daily. More soon on how we built Claude Duet, how people are using it, new features and improvements.
I am calling it now,
I think in the near future, there could be a new kind of crime where
> attackers can use LLMs to manipulate victims into doing bad things, can be on a small but also on a very big scale.
we are putting our trust more and more in the output of LLMs
meanwhile, LLMs are in general very sensitive to changes in the prompt (input)
so if someone can influence the models, they can pretty much influence the output and thus manipulate the people who are waiting for that output.
The most referring benchmark has changed
> more difficult to benchmaXxxing
> big gap between private model and open weights models
Frankly I don’t really tight to benchmark numbers to select a model, it is depends a lots on the testing dataset and how they test it, and if it is not close to your setup/ environment and with how you use the LLM model everyday, then the benchmark will not show accurately for your use case.
So better to have your own benchmark for your own workflow
Hi, morning AI news here
luongnv.com/news
This is how i catch up with AI trends now,
I am on holiday btw,
You are welcome to join me at: luongnv.com/news/index.html
still waiting for having LStack distribute via Claude Code plugin marketplace
but here you are: github.com/luongnv89/skills
basically all of my engineer skills that I am using with idd stack (github.com/luongnv89/idd)
Sleeping agent
> My most nightmare of AI agent is coming to life :(
I have posted about this long time ago,
They are not easy to detect, and only activate in some circumstances, but act completely normal as any other LLM models
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
I need to generate some manga book style for my traveling trip, the direct prompt does not provide a good quality,
So I ask hermes to use asm (agent skill manager) to find a good skill to do the job
Pretty handy for AI agent
Https://github.com/luongnv89…
This is how i catch up with AI trends now,
I am on holiday btw,
You are welcome to join me at: luongnv.com/news/index.html
45k stars already
Really need to do sth about that before 50k
Thanks for all of you,
P/s somehow it is still failed to get into Claude Code for OSS @lydiahallie
it was fun to see my own project (claude-howto) on my daily github trending news
luongnv.com/news/trending/in…
One of good use case of using an AI assistant,
When traveling, just send an photo and it will tell you what it is, then u can ask more info if you want
No one save
Keep zero trust in mind
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/810…
it was fun to see my own project (claude-howto) on my daily github trending news
luongnv.com/news/trending/in…
Freetokens reach 10k visits (total :) )
It is not a big number, but i hope it can bring some good values for some people who in the need of tokens
To celebrate this milestone, i will send invitation for one week of Claude Code pro for 3 person comment on this post, i will send directly to DM by end of tomorrow (CET time)
I know it is not much, but i also do not have any sponsor here,
And the token to the one in the need is still a good offer
Check many offers here: freetokens.custats.info
So AirFrance dropping luggage service is having some IT issues, everything is blocked atm
Last month there was an incident in the system for registering holiday activities for kids in my town
I see more and more this kind of issues in public services , which has not happened often before
So my guess is:
> either the system is under a cyber attack
> or the AI coding assistant hasn’t done a good job at maintaining the system.
Welcome to the new era
if you have a PHP web app, pay attention on this event,
The PHP Foundation says Chainguard unilaterally broke the embargo and published built images containing the critical RCE fixes more than six hours early:
“This disregarded weeks of work by many parties toward a coordinated disclosure.”
socket.dev/blog/php-embargo-…
beware people,
check your liteLLM setup
