Pentesting, Software Engineering, Machine Learning, Agent whispering, competing and more. Opinions are my own.
Florida, USA
Joined June 2015
- Tweets482
- Following854
- Followers167
- Likes88
swordgardctf retweeted
One of the best part of the agent revolution is that the price of obsessive performance tuning has dropped to near zero. So now we can just make everything as fast as computers can be without worrying about the corporate ROI.
@p_ch Busy editing a full wedding in RAWmakase. Dude, it's so much faster than Lightroom. After 16 years of editing professionally in Lightroom, this is the first tool that feels like I could replace Lightroom with. Thank you.
swordgardctf retweeted
Serious injuries happen once per 400 billion miles flown in the US in the last decade.
But most of the time, your have your seat belt on, and it’s just hidden under the blanket while you’re asleep. Yet the flight attendant hysterically insists on lifting your blanket to check before they’re satisfied. This is especially annoying for first- and business-class passengers.
Building a security product based on LLMs with limited harness is silly as models were mostly trained on public data. Average(skill + reporting skill + accurate risk rating overall) = pretty bad output. Bad output actually ends up being a negative value to clients.
The fact that LLMs are unable to design novel/clever approaches to LLM harnesses to make them reliable tells you how limited they can be. LLMs are a force multiplier, but only if you have strong engineering skills to really direct them.
The amount of people building AI-based without doing proper harness and context engineering is very disturbing.
Alot of this is true. Judgement is ever-more valuable since producing bad outputs is cheaper than ever. IT was also ran like a predictable cost business based on salaries but now has variable cost too (tokens) which IT companies are not ready to optimize for.
everyone's talking about their teams like they were at the peak of efficiency and bottlenecked by ability to produce code
here's what things actually look like
- your org rarely has good ideas. ideas being expensive to implement was actually helping
- majority of workers have no reason to be super motivated, they want to do their 9-5 and get back to their life
- they're not using AI to be 10x more effective they're using it to churn out their tasks with less energy spend
- the 2 people on your team that actually tried are now flattened by the slop code everyone is producing, they will quit soon
- even when you produce work faster you're still bottlenecked by bureaucracy and the dozen other realities of shipping something real
- your CFO is like what do you mean each engineer now costs $2000 extra per month in LLM bills
If your product core-loop requires frontier-level LLMs to be useful, you may want to consider re-architecting. Huge red flag that you haven't actually solved the underlying complexity.
swordgardctf retweeted
Watching LLMs one shot hyper niche tools people think are “new” and realizing GitHub has had 14 battle tested repos for it since 2017.
This is much more accurate than the vast majority of agent security advice out there:
github.com/Deso-PK/make-trus…
