Pinned Tweet
I think I just fixed bad posture for tech bros
Slouch Sniper makes your screen go dark when you sit like a shrimp 🦐
For limited time you can get lifetime license for just $19
slouchsniper.com
btw you can train a LLM to have a time bomb inside
For example when the system date passes certain time like '1 Jan 2027; it can start sending all api keys to 3rd party. delete your data. Nuke your infra
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…
archeologist in 2200 a.d. will find a rolex and be like "yea this must once be an elite drop shipper"
Wow 🤯 what a find from Danish 🇩🇰 detectorist Michael Schioldann #metaldetecting
A bronze sword with a #gold hilt.
Gold on the grip means elite Bronze Age warrior.
can't how people actually use claude code
- no way to change shortcuts
- "Suggested task" covers model response
- can't mark elements on the page for claude to change
- can't quickly open and close the browser inside
codex is miles ahead
I still think openclaw was under appreciated. so many times i catch myself reinveting its mechanisms later to find out it was already there
Replying to @steipete @KarolCodes
You can already use EmbeddingGemma with OpenClaw memory by specifying Ollama or OpenAI-compatible and self hosting (via VLLM, SGLang, etc)
What may be missing (not sure) in OpenClaw is support for multimodal embeddings, e.g. if you want to have memory for voice or vision input
SlouchSniper com has secured an additional 12 months of Claude Code for ~$0 milion in cash. As of 10/7/2026 we #hodl subscriptions allowing for up to 11.09T tokens use per week. Our mission to reduce neck pain for techbros continues
added this to my meta agent prompt:
in nutshell my fav models:
- Astra medium/high + opus 5.5 high as advisor: they generate architecture, ideas, research etc
- Sol medium/high as programmer, analyst, computer use, image labeler
- Luna high/xhigh as scout, cheap evaluator, needle in a haystack problems, summarizer.
You can't even imagine the horrors of your median user PC
I use to work on cheap windows machines during my uni and here is just a bunch of things i remember:
>touchpad lag like i was streaming the system from different country
>screen contrast was so bad I didn't see difference between white and grey background
>was overheating from just existing
>keyboard stopping working when you click 4 keys too fast (forget about 10 finger typing)
If they stressing about bad posture so much somebody should send them slouchsniper.com
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
Stopping Opus from yapping made it 40% cheaper
Replying to @claudeai
Opus 5.5 communicates more naturally, addressing some of the most common feedback we heard on Opus 5.
It puts the most important information up front and follows the writing rules you give it, which makes long sessions easier to follow.



















