Hacker Newsnew | past | comments | ask | show | jobs | submit | ymir_e's commentslogin

I do like the idea of getting everything into one place, but I'm a bit concerned about how it gets the data from each platform.

The platforms have a pretty huge incentive to keep you on their platform, so any top of automation/botting is actively something they try to shut down, or am I incorrect about this?

For example LinkedIn needs to serve ads, if you're able to bypass their UI entirely, you're not getting ads. Or am I missing something here?

Would be cool to hear your thinking around this as well as what technical approach was taken.


I agree with all that Joseph commented above. I'd add that we're not removing users from the LinkedIn or Instagram scroll - if you want to view content on LinkedIn or any other social platform, you still need to go to the platform itself. We just wanted to make it easier to stay on top of the 5+ different messaging platforms we're all expected to check.

I think about it like an inverted WULPH.com (the office). I just want to have 1 app that is the source of all my comms.

- Julius


yeah, great question.

Many platforms right now are making the difficult decision whether to: (a) wall off their data from programmatic access (agents) altogether in service of keeping real humans on their site, or (b) allow programmatic access for the increasing number of users who are demanding their agents interact with the data

We obviously believe humanity and society and cyberspace should go in the direction of (b)

LinkedIn is an interesting case because their policy, so far, has been to aggressively restrict agents from interacting with their site. But this is because they serve human-centered ads which most of us find aggressive and annoying (I'm thinking of messages appearing in my DMs that appear real but are in fact sponsored). So you would still see that "ad" even if you were using Attention. It would just be easier to label it as "promotional" and i would prefer my agent does that so I do not even have to see it.

I want linkedin to serve my agent ads that are so interesting and personally compelling that my agent thinks i should hear about it.

Just my 2¢ - perhaps my cofounder will chime in with his own thoughts

- Joseph


A tmux plugin that makes is better to work with many agents:

https://github.com/Ymirke/tmux-agent-switcher

You press ctrl+n and you get a sidebar showing all the tmux tabs you currently have and the statuses of the agents.

The sidebar then has a couple different navigation modes based on what floats your boat, toggle by tab:

1. Using vim bindings j/k, and supporting number -> j/k to go up or down quickly.

2. Using number -> number. Meaning you press the number of session then number of tab.

3. Using search, useful if you have a ton of sessions open.

I really like what herdr, cmux and others are doing in this space, but I find it hard to get myself away from my core tmux setup, so I made this to not have to change too much of how I work.


Edits for clarity: seems like this was a controversial take, which I find pretty interesting.

I believe people took this comment as LLMs being better than security aware engineers who have the time to spend building solid systems.

This wasn't the point, the reality outside "established" tech companies is that software security can be lackluster.

There often simply isn't enough resources to check old software for basic vulnerabilities, outdated packages with known issues, there might be a manager who insists on a certain solution. Or that certain services are on "maintenance mode", but rarely get checked since they're on a certain part of the internal network.

If they are able to properly scan their full software stack for CVE issues, they cannot deal with a flood of CVEs.

----

We're in a transition period where AI will eventually make software much more secure than it ever was.

These noisy CVEs will probably lead to agents verifying vulnerabilities before humans review them.

The problem with agent reviews from what I can think of is:

- cost to use LLMs to review things

- not necessarily easy to plug-and-play in repos: (domain knowledge + vulnerability knowledge)

- especially with anthropic: able to use models defensively, without hitting guardrails

The last one is the most interesting one to me. How does the AI providers know if you're a "good or bad" guy? And does it matter if open source models is catching up?

We're in a kind of cyber arms race wether we like it or not.


Currently we either get AI promo vulnerability dumps like from Chrome with pretty graphs that no one checks or false positives.

Where is this one now that was hyped everywhere?

https://news.ycombinator.com/item?id=49133889

The GitHub submitter could no longer reproduce the issue and the LKML post has no replies:

https://lore.kernel.org/all/CALCETrXbj__SFQMzPZhES5y6-sh4np-...


All the big maintainers I know of disagree. Curl for example is getting overwhelmed with real issues that LLMs are finding. A year ago most LLM found issues where false positives (the code is correct), or hallucinations (the code in question doesn't even exist) - but that has changed and now LLMs are finding real issues in code. Many different projects have reported the same rules.


Absolutely. I'm not sure how universal this is, but I saw George Hotz tweeted something about Mythos not being such a big deal since he's able to find security issues without a claimed $20,000 in token spend per vulnerability.

The reality is that humans aren't perfect, so we do not make "perfect programs".

It is also possible to test it yourself to see how good LLMs are at finding software issues:

Find any open source project that isn't huge, then have Kimi 3 review it for potential issues.

You will find code "issues".

Most of the time it won't be detrimental for security, but often there are code paths that are not "accounted for", or you're able to bypass a type of check by interacting with the software in a strange way.


So we just need a redundant array of George Hotz, I wonder how much that costs?

This is where quantity becomes a quality. Actors with huge budgets and at least a small team to keep the AIs on track now have an immense amount of potential power. Anthropic has somewhat resisted in giving state actors like this full access to 'dangerous' models, but it appears other companies like OpenAI have no such reservations.

Before such states may have difficulty hiring enough security researchers/hackers. Now you can spin them up. Someone with a few data centers of GPU compute and enough proxies could probably blitzkrieg the internet in a day and capture ungodly amounts of information.


Not just the cost, also boredom. I have used AI to trace down some tricky multi-threaded issues (someone has to write that cross thread framework, and it was me - fortunately most of our code isn't this complex). I already knew the code well, but it would have taken me a long time to find the same issue - the more likely thing is I would give up. Even if I don't give up I'm likely to get bored/frustrated. A month worth of tokens replaced what I would have needed a year to do if I was allowed to spend a year on it.


I'm sure if somebody hired him for a pentest it'd cost more per vulnerability found.


I think his point was more that he didn't believe it was difficult to find these issues.

Could be Dunning-Kruger, but if AI models continue to get better this trend will surely continue.


1. George Hotz exhibits strong signs of Dunning-Kruger in most things he does nowadays.

2. Even in exploit development, his original field, he might be able to find vulnerabilities in mature, well-maintained software with security programs. But would he charge $20k or more for that service? Say it with me, Claude: “Absolutely!”


This was my first thought, this could be terrible if used offensively.

The best defense I can imagine is to have an agent reproduce the issues before a human sees it, but even that will cost money.


If by "terrible" you mean "amazing". Imagine being able to tear down an entire Jenga tower of bullshit corpospeak, and replace it with actual security.


You mean wezterm right?


Woosh


Lately I've seen a lot of people complaining about GitHub downtime, performance and overall quality.

Happy to see something in the right direction. I think they've woken up a bit. Still surprising how slow things can move at big companies.

Companies like Linear, Vercel, Zed and Cursor all seem to be looking at GitHub more aggressively though. I do suspect there will be more competition shortly.


I've tested out just about every tool like this, and ended up coding my own, it is really minimal though.

The primary reason people reach for these tools are two to three reasons:

1. Tmux does not natively show agent statuses of agents / notify you when one needs input. Helpful when you have a huge list of small things to fix: I just spin up N agents in parallel to handle all of them, then I go over and review.

2. Tmux does not handle worktree handling. If you wanna make changes in parallel you cannot have two agents make db migrations at the same time. The way to solve for this is to have them work on two different worktrees with separate environments, ports etc.

3. Tmux tree view is not super beautiful, especially for viewing agents.

I built this as a tmux plugin since I don't really want to opt-out of tmux as a whole, especially since I like my keybindings there.

https://github.com/Ymirke/tmux-agent-switcher


This looks nice, but I think a screenshot or three in the README would help bring it to life better


Great input. Added a screenshot to the readme:

https://github.com/Ymirke/tmux-agent-switcher


In your experience using your plugin (and the other tools that do something similar like herdr), does it make doing a larger volume of work feel more manageable? In my experience thus far, surfacing/managing session state is only part of the issue, context switching between them is another entirely and feels like the real productivity bottleneck for me


You're spot on here.

If you're truly multi-tasking with this type of setup it feels more like "tiktok coding", "brainrot coding", or any other term that should be used for it.

I'm pretty sure we're moving away from it, the agent you interact with via voice or typing will go to a higher and higher abstraction level, and you won't have to effectively micro-manage them.


> The way to solve for this is to have them work on two different worktrees with separate environments, ports

What are you using for the environment part? Containers?


Yes! The project I'm working most on has a dev environment that runs a compose witin two scripts: a setup.sh and a run.sh script.

setup.sh figures out which ports are available and changes .env files accordingly. Run.sh runs the actual setup.


Please continue developing this I’m sick of all these one-offs that escape the mature good tooling we already have.


FYI: Kimi K3 is relatively expensive on open router API pricing for agentic tasks, or at least that's been my experience playing around with it.


They just opened the weights. I expect competition from various providers will drop the price a bit. But you're right. I'd hope a model closer to GLM 5.2's price would be sufficiently useful.


A license and presumably revshare is required for large scale inference as a service of Kimi K3. There's clearly something going on right now with every router at the same or higher price as the Moonshot list price.

So I wouldn't count on competition if $/mil token is actually set by Moonshot, but I would expect $/tok to drop when there's a more competitive frontier open weight model.


Where did you get that from? That‘s not what the license says: https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE

The current price is likely a result of the high demand and the high requirements of this model.


Did you actually read the license?

> If the Licensee or any of its affiliates operates a Model as a Service business, and the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars (or the equivalent in other currencies) in total over any consecutive 12 months, the Licensee must enter into a separate agreement with Moonshot AI before using the Software or its derivative works for any commercial purpose.


I stand corrected. I read over that multiple times somehow.

Looks as if these companies could wait until they reach $20 million of revenue with Kimi K3 until they enter a separate agreement.


if you’re hosting it then you have to have an agreement with Moonshot, which presumably includes terms about pricing.


Only if "the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars (or the equivalent in other currencies) in total over any consecutive 12 months"


What a fun website.

This feels like a sort of flashback to how the web used to be in some ways.


Can Europe train a frontier AI model?

It seems unlikely at the moment. The writeup you have is quite interesting, but notably distributed compute for frontier is at least a bump in the road.

As other's have pointed out, if resources across Europe came together there could be a good amount of resources today to do so, but it is fragmented.

Project Glasswing, and most recently the blocking of Fable 5 potentially only for non-US citizens have sparked more attention in the political spheres regarding sovereignty/competitiveness of European AI.

There as also this great write-up which explains the trajectory Europe currently has when it comes to AI compute here: https://europe2031.ai/

So theoretically: yes, but there doesn't seem to be a big enough will to attempt to catch up with the trajectory of American hyper-scalers.


Thanks! Will give it a read :)


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: