I'm a geneticist, so it's nice to see, but it doesn't seem to belong on HN to me either. Maybe because someone thinks this information should influence their use of the hot-new pseudoscience cureall - "peptides"?
The job that a compiler does is just a boring mechanical process, which makes it well-suited for a computer to replace. Also, I can use a compiler for free without relying on an evil company.
Personally I just accept that all technologies are great and must be embraced! This way I do not have to think about ethics and potential implications for society.
it's mind boggling. personal mobile phones have potentially anyone's software running on them and that can connect to the internet means that literally anyone could be tracking and gathering who knows what data from your operation. it's an indication of the greatest unseriousness.
the location of the ship of course is not secret. but there is finer grained data about the people, the devices and what they're doing that could be gathered. and inferences made from that data. i would only allow this data to leak out if i could somehow use it to deceive my enemy.
"just" is doing a lot of work there. There's a stack, and everyone is fighting and racing to figure out which parts of the stack are commodities and which have protectable IP.
For example, years ago someone might have said: "Software programs are "just" some instructions on top of a computer, which does the actual computing work. The software companies don't own the computer. The next version of the computer might not even be capable of running the software." And for some kinds of software, they'd be right. Some programs turned out to be more replaceable than others.
I assume the GetViktor folks are hoping that some combination of know-how and learnings from real customer interactions and data will help them build or find a sustainable competitive advantage in some niche.
> They don't own the LLM model. You're just one silent update from not delivering what you promised.
This is true but a small risk IMO. Worst case they can shift to open-weight LLMs for inference. I would bet on LLMs being an increasingly commodity part of the stack.
u mentioned weighing OpenViktor against staying on Viktor for the security tradeoffs - did you happen to clone the repo before it went down? I'm trying to get a copy to evaluate it myself. would really appreciate it if you could share - openviktor@proton.me
I doubt it makes a difference. The primary risk is the agent exfiltrating your private data. That's going to exist either way.
Essentially anything you give it access to should be considered inside the same security boundary. Which is quite unfortunate if you want it to respond to emails for you and also query the internet at large.
Strongly agreed. However, some developers have trouble writing clearly and reading lots of text, and therefore prefer oral and interactive + real-time transmission of the information. Those developers, I suppose and hope, are discovering that they can talk out loud to their agents, explain everything interactively, and then the agent can create whatever longer-term artifact it wants to record the understanding. Multi-modal interfaces FTW?
Users are "dumb", and it's a dumb _system_ and dumb business that doesn't plan for that in terms of FTUE, business model, support model, and product flows.
We product makers get to think about our one little product all day, and it's our job to make our product work for the "dumb" users. It's not their job to adapt to us.
Cool question. What form would an answer take? We need some detection benchmark data thats invariant over the period of interest. I hope the data exists but I would be surprised.
Another way to come at it would be mortality data. But that has a bunch of its own problems.
Everything is changing at once, it makes this kind of science so hard.
reply