nijave

8 ms·

121.
by nijave 3mo ago
It's both. Some models, given web search and web fetch may only run a search and assume the summary text is correct and blindly return it. Others will validate by running a web fetch and checking the whole page contents. Even better, t
122.
by nijave 3mo ago
True but you can have cert-manager issue public certs then create service accounts for off cluster things to be able to pull the cert from the Secret so k8s+cert-manager acts as a local broker that handles renewal. You can also invert and h
123.
by nijave 3mo ago
We have a few subdomains for white labeling 3rd party SaaS where we do what is basically the AWS ACM equivalent and add a persistent record from a vendor. With this setup, I don't have to grant 3rd parties DNS access. I actually made a
124.
by nijave 3mo ago
On k8s, there's cert-manager but also you need k8s... Most browsers support trust on first use for leaf certs
125.
by nijave 3mo ago
My IaC is on public GitHub. They could do a network scan to find software then fingerprint to find version anyway. Removing attack surface is better than trying to hide it.
126.
by nijave 3mo ago
Arguably it's 1/2. You can put public certs on proxy then give proxy private CA to backend services. Then you don't need public certs for all the private stuff nor need to trust the private CA on all your devices.
127.
by nijave 3mo ago
I didn't realize the prevalence until I worked at a company trying to get enterprise market share desperate to "add logos" (the customers were always referred to as "logos" if that tells you anything)
128.
by nijave 3mo ago
Arguably I'd call that the 90%. In my case, answering the restaurant question correctly with "Rishi" in my tests was the sole intent and 90% of the problem. All the models "helpfully" added extra junk about the clos
129.
by nijave 3mo ago
My gut feel is Anthropic is very technical and pedantic which makes their models really technical and pedantic. They're top at code and technical benchmarks but anecdotally I've found OpenAI to be significantly farther ahead for g
130.
by nijave 3mo ago
Yeah, just another example of "connections pools are hard in general"
131.
by nijave 3mo ago
Can also use different database names since pgbouncer lets you remap that
132.
by nijave 3mo ago
You can rename the tabs in tmux but not sure if there's a way to do dynamic status (I suspect there is but it's probably non trivial to get it to play nice with the harness)
133.
by nijave 3mo ago
Everyone does that. Some random startup "We're used by Google!". More often than not some random engineer signed up for a trial or some tiny department somewhere is using it for some internal thing
134.
by nijave 3mo ago
tmux has a mouse switch--does this do more than that?
135.
by nijave 3mo ago
pgbouncer has different pool modes you can pick from that have some impact over what's possible to leak https://www.pgbouncer.org/features.html
136.
by nijave 3mo ago
Recently ran into a related bug in duckdb. They implemented a basic http connection pooler (described as a parking lot) but there are edge cases where broken connections get returned to the pool then the next thing tries to use a busted con
137.
by nijave 3mo ago
Would be interesting to see huge pages and io2 impact. I did a smaller version on Azure and disk latency had a massive impact much more so than max IOPs (although their crappy storage offering needed like 64-128 iodepth to get advertised io
138.
by nijave 3mo ago
x86 doesn't have a firmware hypervisor that allows splitting into LPARs. That's showing 3 layers of nesting: LPAR, hypervisor, nested hypervisor
139.
by nijave 3mo ago
No it's not. The poster is advocating replacing technology to share a single system with dedicated, non-shared systems. Even the most basic business app has an app server and a database server. If they have 6 business apps, they'd
140.
by nijave 3mo ago
I'm a bit confused. So you're saying instead of running 20 containers for an app that are 1-4Gi and some odd CPU I should replace them with 20 dedicated servers? By your own numbers that's 200x+ as expensive. Really 20 contai
141.
by nijave 3mo ago
IBM mainframes might but I have a very shallow understanding. Found this https://sixe.eu/news/kvm-nested-vms-ibm-power-linux-lpar
142.
by nijave 3mo ago
Proxies are still 3rd party hosts. I've ran into issues in my very limited go experience with them missing things and having hash mismatches (maybe someone changed a tag and it was working as intended? I don't remember the exact c
143.
by nijave 3mo ago
And Chinese don't use American models because they're American. What's your point?
144.
by nijave 3mo ago
Not sure if it qualifies as "swarm" but they're useful for controlling context size. You can spin them off for context heavy things like research and analysis then only a much shorter summary gets passed back into the main co
145.
by nijave 3mo ago
Not sure relying on a bunch of various VCS to stay online is necessarily a great approach either. I think go is also a little more amenable to source library distribution since there's a pretty broad pure go ecosystem. For interpreted
146.
by nijave 3mo ago
The safety filter rejected or the model was down?
147.
by nijave 3mo ago
Yeah, I really wish articles and comments about "<model> running locally" also reran the same common benchmarks published to compare the results.
148.
by nijave 3mo ago
Fuse overlayfs was noticeably slower than Docker's overlayfs setup. Maybe there's a way to reconfigure but I haven't checked lately. I think it was Zig (building ghostty) was broken in podman with obscure "unknown file&q
149.
by nijave 3mo ago
Yeah MCPs are easier to distribute. In my experience, they also work better with dumber models than CLIs (which saves money)
150.
by nijave 3mo ago
Tragedy of the commons

More ›