ic4l

The Backblaze Restore Application does not work reliably when restoring a large number of files. On an M4 Mac, it repeatedly crashed and ended up in a non-recoverable state, forcing me to recover the files another way.

Some restore options also do not let you move data to their cloud for direct download, which makes it feel like they do not actually want users to fully download and recover all of their files.

We tried this initially

  FROM mirror.gcr.io/{image_name}

We received

  failed to resolve source metadata for mirror.gcr.io/

So it looks like these services may not be true mirrors, and just functioning as a library proxy with a cache.

If you're image is not cached on one of these then you may be SOL.

This broke our builds since we rely on several public Docker images, and by default, Docker uses docker.io.

Thankfully, AWS provides a docker.io mirror for those who can't wait:

  FROM public.ecr.aws/docker/library/{image_name}

In the error logs, the issue was mostly related to the authentication endpoint:

▪ https://auth.docker.io → "No server is available to handle this request"

After switching to the AWS mirror, everything built successfully without any issues.

Thousands of large Facebook groups appear to have been taken down, with many admins reporting a wave of rapid content reports from a single account targeting the group's first post. This often triggers the group name to change to "Group title pending" just before the group is removed.

> https://techcrunch.com/2025/06/24/facebook-group-admins-comp...

Yeah, thats one of the best ones I have seen, and it popped up a while ago.

That was fast :o

Did they already disable it?

When using `gpt-4.5-preview` I am getting: > Invalid URL (POST /v1/chat/completions)

`Especially in a database that doesn't allow wildcards`

I agree with you. Knowing the exact column names can speed up an attack and, in some cases, make it more feasible.

Why don’t they just request disclosure of what’s actually stored and allow renaming of the columns? It seems odd that knowing the exact column names would be necessary if the goal is simply to understand what data is being stored and its intended purpose.

Great answer, thanks.

LXD seems like an unusual choice when Kubernetes already has cadvisor and strong monitoring integrations. Avoiding extra agents is nice, but does this really scale better than existing solutions like Prometheus and OpenTelemetry?

What’s the advantage here beyond keeping things lightweight? Feels like this could hit limitations as complexity grows.

For me the o models consistently make more mistakes for me than Claude 3.5 Sonnet.

Most browsers won’t report anything higher than 8, even if your system has more memory.

I think its related to fingerprinting attacks.

In case you are wondering this actually beats the game:

    setInterval(() => Array.from(document.querySelectorAll('.main-btn, .upgrade-icon, .press-collect, .press-btn, .collect')).forEach(item => item.click()), 50);

After around level 40 things get very very slow.

HN stripped the query param for x/twitter.

https://x.com/search?q=%22%40cursor_ai%22&src=typed_query&f=...

Account settings not accessible, which means any pro users cannot use pro features.

Yeah I wasted 10 minutes on this, and even contacted support.

Their status page is misleading and the only way to see it is to look at the most recent tiny sliver of their chart.

Thats one of the worst CAPTCHAS I have ever had to deal with.

After passing it you end up getting a 503 ERROR against https://tcr9i.openai.com/fc/gfct/, and it shows no actual fail notification, and prevents you from logging in.

Started around 40 minutes ago.

The API appears functional but any playground login, or mobile sessions seem to be invalidated without the ability to log back in.

Heres a twitter/reddit thread confirming multiple people having issues:

https://x.com/TheWebLifestyle/status/1826324989576184250

https://www.reddit.com/r/OpenAI/comments/1exx0vb/cant_log_in...

How many people even have access to the gpt-4-32k, or gpt-4-32k-0613 models.

I think they give everyone access to the gpt-3.5-turbo-16k, but I have not found a way to request access for the 32k model.

There does seem to be an option through azures openai service: https://azure.microsoft.com/en-us/products/cognitive-service...

Absolutely, Zack. The work you've done with PAKman is seriously impressive. The efficiency you've managed to achieve with Alpine packages and LXD seems to be a real game-changer.

I'm really curious about the team behind this. Is this a solo endeavor, or do you have a team backing you up? Handling everything from building the system to managing the lifecycle with LXD - that's a lot of ground to cover.

If this is a solo project, it's nothing short of astonishing. The innovation, the effectiveness, and the potential of PAKman - it all points to an immense amount of work and dedication. Hats off to you!

The idea of instellar.app serving as a package repository is very interesting as well. It makes perfect sense that you'd initially want to provide a fully integrated solution. It fits right in with the efficient, streamlined approach you've taken with PAKman.

Either way, I'm definitely looking forward to seeing what comes next from your work. Keep it up!

This is neat stuff. You went against the grain by ditching Docker and K8s, and it's paying off. Alpine packages and LXD feel like a power move that's flying under the radar for many of us who are stuck in the Docker/K8s world. Kudos for hacking together something that works so well for your needs.

I'm curious about how you handle networking, security, and any other OS-level configs that Docker/K8s ecosystems typically manage. How has your experience been in these areas with the LXD/Alpine combo?

Also, I hope you're considering expanding PAKman's support for other build environments. The simplicity you're describing is super tempting! Keep up the good work.

In Thailand most in person transactions are done by bank transfer via QR code:

https://www.bangkokbank.com/en/Personal/Digital-Banking/Prom...

The local sellers pay little to no fee, and the transfer is instant. All of the popular banks here support this form of payment.

It's also very common for people here to do bank to bank transfers, and they are all instant.

The US is behind in many ways.

He was training AI models, and needed some test data sets... Of course he upset many people .

You also can use this to meet credit card minimum spending for credit card bonuses.

Why are these not being marketed to do something useful.

I mean a $3,000 water bottle holder...

Even if they fixed it in 20 minutes the chain reaction caused by Fastly being down took much more than 20 minutes to resolve itself.

An example is imgix:

https://status.imgix.com

It took them 11 hours to recover from Fastly going down for their claimed 40 minutes.

imgix is also having a major outage

I second this, Hasura is a solid choice.

If anyone wants to read the first HN thread 6 months ago about the termination it's here

https://news.ycombinator.com/item?id=24988301

I opened my 5 cards across the span of 13 years, and was advised to have more credit cards as only having 1 was negatively impacting my credit score. According to Experian the average American has 4 credit cards.

Also chase incentivizes you to open up more than one card with their signup bonuses.