AtNightWeCode

Sounds like something a buyer would say. Surely Netflix can handle HBO traffic better and cheaper. Maybe HBO are stuck in some deals. But it is a no-brainer to trash the HBO backend over time.

Not missing working with LUA in proxies. I think this is no big thing. They rolled back the change fairly quickly. Still bad but that outage mid November was worse since it was many bad decisions stacking up and it took too long time to resolve.

The cost of tokens in the docs is pretty much a worthless metric for these models. Only way to go is to plug it in and test it. My experience is that Claude is an expert at wasting tokens on nonsense. Easily 5x up on output tokens comparing to ChatGPT and then consider that Claude waste about 2-3x of tokens more by default.

Maybe I am too cynical but the story sounds fake. And I would not be surprised if this is some way of manipulating the market.

Sounds like some sort of scam to me. The item is probably correct but the price seems way too high.

Exactly. The newbie mistake in SQL is also way worse than this. But the whole design is also bad. Clearly implementing things at the wrong place.

And, it took like over an hour between the problem started til my sites went down. That is just crazy.

So they made a newbie mistake in SQL that would not even pass an AI review. They did not verify the change in a test environment. And I guess the logs are so full of errors it is hard to pinpoint which matters. Yikes.

Does it even help? Get the name of some person => {"name":"Here is the name. Einstein." }

I care less and less about things like this. At some point you will write code in some lang that have fancy keywords and stuff gets mutated anyway. Also, what people tends to do if stuff is immutable is that they hide mutation by doing deep copies with changes.

This battle was lost a looong time ago. The effort it takes to keep up with all the shenanigans of Google and that play store is way worse than these new changes.

I looked into this before and the stocks of these large corps simply does not move when outages happens. Maybe intra-day, I don't have that data, but in general no effect.

From Azure status page: "Customers can consider implementing failover strategies with Azure Traffic Manager, to fail over from Azure Front Door to your origins".

What a terrible advise.

Earnings report today. A coincidence?

I can at least login to Azure. But several MS sites are down.

Still don't get why on earth anybody would run a Docker version of MinIO in production. And why is this even a problem. Not like you put a private storage service on the Internet? Or do you? The incompetence of the average HN user is just mind blowing.

It is a test to see if you can break out of the prompt. You have a system prompt like. Bla bla you are a pro AI-translator bla bla bullet points. But then it breaks when the context is like "name my cat" or whatever. It follows those instructions...

How about fixing the most basic things first? Claude is very vulnerable when it comes to injections. Very scary for data processing. How corps dares to use Cloud code is mind-boggling. I mean, you can give Claude simple tasks but if the context is like "Name my cat" it gets derailed immediately no matter what the system prompt is.

I used MinIO for local dev. I can use S3 or R2 in some cases instead. Kinda crazy to find out that people use these Docker images in production. Why on earth would you do that?

It is funny that only morons can down vote on HN. I'm pretty sure I am better than any random idiot on HN on this topic since I worked DAILY with these issues for a couple of years. Years before there even was an official "AI browser".

No, they did not breach anything through SharePoint. The flaw is that IDIOTS exposed these servers to the Internet. I am very pro holding vendors accountable but this is just stupid. "Pro-tip" btw. SharePoint installations often have the pw sharepoint, sharepoint123, sharepoint-123 and so on in various casing and delimiters.

It is very basic. I have built my own version of this based on Chromium that integrates both Claude and ChatGPT in the browser. It can do a lot of tasks like translate or shorten the text I selected and so on. It took me like a couple of hours to build. The problem is the cost of using the LLMs, especially since they are still pretty stupid and requires huge prompts.

EDIT: I think I misunderstood your Q. Sorry. You can take a screenshot and post it to ChatGPT and get back what it is seeing, in theory. I mean, I use ChatGPT to post screenshots of my sites to get feedback on my layout and designs...

I actually built my own browser on top of Chromium to be able to use both Claude and ChatGPT ad hoc for various tasks in the browser. Fact is that I use my OS native text to speech more than the LLMs...

Anyhow. The problem is that the LLMs are simply not good enough. And as someone who processes a lot of data through LLMs daily via APIs. The quality is just poor. Clearly LLMs does not work as stated. The fluctuation in the quality of the responses is just silly and very undocumented.

Not uncommon that machines get exposed during trouble-shooting. Just look at the Crowdstrike incident just the other year. People enabled RDP on a lot machines to "implement the fix" and now many of these machines are more vulnerable than if if they never installed that garbage security software in the first place.

You wrote "You’d need to have every single service you depend on, including 3rd party services, on multi tenancy.". This is highly incorrect. I worked at several companies that have a multi tenancy strategy. It is:

* Automated. * Scoped to business critical services. Typically not including many of the 3rd party services. * Uses data replication, which is a feature in any modern cloud. * Load balancing, by DNS basically for free or a real LB somewhere on the edge.

If you fail at this you probably fail at disaster recovery too or any good practice on how to run things in the cloud. Most likely because of very poor architecture.

You don't need expensive SLAs to do data replication or load balancing in the cloud. It is pretty basic.

In general it is not expensive. In most cases you can either load balance across two regions all the time or have a fallback region that you scale out/up and switch to if needed.

Considering the history of east-1 it is fascinating that it still causes so many single point of failure incidents for large enterprises.

I think you are bit wrong. Once someone close to you dies you remember them by their legacy. Also, you just have to laugh at some of the chaos these elderly cause. They call you in the middle of the night being lost somewhere and you have to guide em home. Or help the cops guide em home.

I am pro assisted suicide. Not sure about Switzerland but some countries allows it for young people with mental health problems. That I can't accept that.

exactly, cost per token is higher but it also uses tokens like a chipmunk on steroids

Sonnet is just so expensive comparing to other competitors. Have they fixed this?

Historically. If you are logged in you get better results. I got no ads when logged in. Not logged in I got 8 ad links in total before the real site.