koyote

Yep, same here but with xfce4-terminal.

I've just reverted back to using that as it's faster to load and has all the features I need.

I don't think that's a good comparison.

I've often had new phones that are 'faster' but I barely notice the fastness until I go back to my old phone and realise that my finger-memory needs adapting because things take just a bit longer to load.

I'd assume it's similar for screen refresh rates.

Microsoft has abandoned Azure Devops so we've had to move to Github for any 'new' (read AI) features.

Github is so bad I don't even know where to start. It's fine for small personal projects and the odd project with a few collaborators. But it's just not built for managing large software projects. Even the PR process is awful compared to Azure Devops (and Azure Devops was, in my opinion, ok bordering on good). Diffs are somehow worse than any tool I've ever used.

I am now questioning what Microsoft has been doing all these years they've owned it.

> Does the fact I have written this resolver make me liable should it not block the domains they listed which may contain copyright infringing content?

I use unbound and have a similar question. Shouldn't we just encourage everyone to run their own resolvers (it's turnkey for the most part!)? Will Sony/beIn start going after the root servers and enforce a global block?

This is the only way to work with AI and ensure your code is still (somewhat) maintainable in a year+.

Anyone who's ever directed AI at a very large legacy code base will have noticed that AI just gets dumber and slower the worse (larger + inconsistent) the code base gets (just like us humans do). The big difference is that with AI you can get to the point where the code base is an unmaintainable legacy mess in a fraction of the time that it used to take. Directing architecture is the only way to mitigate this to some extent.

Definitely agree. Sol 5.6 was great when it was released and for nearly a month.

Since then they cut the price but the intelligence (of 6 and 6.1) has gone down to Sonnet 4 levels. I am back in a world where the model is lazy, cuts corners, ignores existing coding standards and context, makes stupid mistakes etc.

I am not sure how those models still rank so high in benchmarks when real-life usage is so obviously bad.

I have a similar amount running on a 16GB NUC in proxmox LXCs and it usually sits at less than half memory usage.

The only big memory guzzlers are generally proprietary software, like the unifi network controller software which basically requires an entire room of Cray supercomputers just to view your Wifi AP status.

Do you have any insight into how your management tracks the value it is providing?

This is something we've been trying to do for a while now and have not found a convincing metric.

I have only once in my life had to request a chargeback for some random charge I had never initiated.

Where do you live or shop that you are happy to take what is effectively a 3% pay cut for the privilege of having a slightly easier option to get your money back if someone charges you the wrong amount?

Great work!

I assume because it has Python wrapper, the HomeAsstant integration can make use of it soon?

The US is really weird with flight numbers and I don't think any other country in the world does it the same way.

You often get flights sharing the same flight numbers such as JFK-SLC and SLC-SFO being the same flight number, even if it's a completely different plane doing the route!

Also super confusing when trying to search for your (or a loved one's) flight online as you'll get multiple matches.

I think the fact that Sol 6 appeared higher than Sonnet 4 on benchmarks shows that benchmarks are completely rubbish and useless.

I've never seen such a large degradation in intelligence in a model until I tried out Sol 6 after having used 5.6 almost exclusively for several weeks.

I am glad I am not the only one to notice. I feel like I've gone back to Sonnet 4 levels of incompetence!

With Sol 6 I am back in a world where the model writes bad code because it is lazy ("You're absolutely right, I did not [do it properly] because I did not want to edit [a normal amount of files]").

And yet they have only improved marginally in my use cases since around Opus 4.5.

The harnesses have improved somewhat, but the code produced on large or legacy code bases is still very average and I still see similar mistakes made that I saw back a year ago (although less now that harnesses have become better at steering).

For my use cases, we are definitely on the flatter part of the curve at the moment.

The non-ducted systems will usually have an IR remote. You can buy devices to blast IR at it and HA has good support for that. This is what I use.