You may be new here. This is Simon’s de facto benchmark for models. I happen to find it a really good one.
Small aside: It’s crazy to me that while it’s improved over time it does seem like most of the models haven’t been trained specifically to defeat this one.
I’ll never forget watching Gary Bernhardt give his talk on JavaScript.[0] Was my introduction to asm.js, and the rabbithole associated with compiling code to run in the browser.
12 years on, it’s shocking how much of his fiction became reality.
Do you foresee this changing anytime soon? Would love to contribute but also I think community adoption and contribution would go along way in terms of businesses less worried about single points of failure.
It’s hard balance to strike for sure. And it’s getting weirder by the day with agents.
This is nuts.
But give me some rope here, how much of react renders in Vello. Like is Vello taking the place of the shadow DOM here or is the entire DOM being render on a WASM thread somehow.
Now do this in containers with gMSAs. It eliminates the need of passing around Admin creds. Which I cannot stress enough. You shouldn’t be throwing your DA credentials into your random Linux machine’s Kerberos cache.
Amazon open sourced a project trying to solve similar problems.
UV and the crew at Astral really moved the Python packaging community forward.
I would love to see them compete with the likes of Conda and try to handle the Python C extension story.
But in the interim, I agree with everyone else who has already commented, Pixi which is partly built atop of UV’s solver is an even bigger deal and I think the longer term winner here.
Having a topologically complete package manager who can speak Conda and PyPi, is amazing.
Depends on the industry. But many large enterprises in the Fortune 500 are actively trying to move away from your traditional VPN. (F5, Pulse, Cisco, etc).
Even with VPNs the question should be, what are we gating behind that VPN anyway. Does it actually give us the granularity of controls we want or is this all security theater. (Also what about hybrid infra, between the datacenter and cloud)
FWIW, my ideal architecture is Wireguard into Corp. (Ala CloudFlare Warp, Tailscale, etc) Corp doesn’t hold a ton of sensitive assets. Or put another way, it’s a lower trust tier.
And then using something like Teleport, Octelium, etc to reach production assets.
Admittedly no vendor product I’ve come across yet has bridged this gap nicely. The überProxy tend to focus on the application protocols they support. While the wireguard clients cares more about session control of the tunnel.
Reading through the docs. I feel like a lot of people are missing the value here. This could be a diamond in the rough if it actually delivers on its docs.
What enterprises want is to move away from perimeter based security models towards the promise that Google überProxy/BeyondCorp popularized many years ago. Which has been lost in the buzzword soup. It’s very simple.
1. A clean separation between Prod, Corp, and the public internet. And the UX to hop between them as an employee is as transparent as possible. (Often times network segmentation comes with additional painful friction for engineerings.)
2. One pipe to observe, and clearly attenuate permissions as traffic/messages flows between these boundaries.
3. Strong proofing of identity for every client, as an inherit requirement.
The problem is everyone outside Google has incredibly diverse protocol ecosystems. It makes those three promises incredibly difficult to deliver on as a vendor. (I’ve evaluated many)
To build a proxy that is protocol aware, only solves half the problem. It gets you some coarse grain decision making and a good logging story.
To build a proxy that is also able to perform type-inference at the request layer, allows for a much richer authZ story. One where businesses can build an authorization layer at the proxy better than their in-house apps could even do natively. (As it turns out, having all the predicates of the request available to a policy engine is super useful).
The docs are a little verbose, the marketing maybe isn’t amazing. But this is inherently a complex problem. No one has fully solved.
Teleport was first to the market to OSS and commercialize a lot of these ideas.
StrongDM also is doing really interesting work in this space.
I wish Hashicorp had invested more in this space.
Small aside: It’s crazy to me that while it’s improved over time it does seem like most of the models haven’t been trained specifically to defeat this one.