anentropic

For me the answer to painful merge conflicts has been... Claude

it's wild isn't it, that at the same time it's an enormously widely-used and reliable piece of software

Yep that's exactly what I would want to do

I was impressed that it 'just worked' on my dbt project

We're currently using sqlfluff linting but it was a bit finicky to get the pieces configured.

sqlfluff does have linting inside VS Code too though, when it works (python path issues...)

Looks very promising!

Now I have to decide which of the ERR lints we want to fix and which to exclude :)

It's probably trained on YouTube videos

it is a piece of furniture though, so aesthetics matter

No, me too

Ironically it seems to have been made by LLM ...?

it's be more impressive if it wasn't so pointless

LOL, I love that it inferred that this particular armadillo would be pretty smug

agreed, it is bad - pixellated in a blurry way that turns letters into blobs

excited to see Unikraft getting some traction...!

I guess I am curious if/how exactly the Netlify announcement relates to unikernels?

Yeah I was also wondering - what do people actually do with this stuff?

> One of the things I dislike about Claude is sometimes it's memory is all-encompassing. It's weird that it brings up things about my personal life when I'm talking about something related to my business

Do you use 'projects' in Claude?

I had the impression they provided discrete memory profiles on top of the shared one

is that really the America you grew up in...?

> > There is no negative mass, so you can always infer the one real arrangement of stuff inside from the warping of space-time at the surface of your box.

> Positive mass does not make this argument work. Two bodies in GR can have different distributions of mass but identical exterior geometry. A counterexample is two spherically symmetric, uncharged bodies with the same total gravitational mass but different radial density profiles. By Birkhoff’s theorem, their exterior vacuum geometries are identical.

I thought I understood the holographic theory better after this article. But now it's clear I don't.

Probably also goes along with the shift to more efficient and quieter devices - first with the OS adding a working 'sleep' mode so you don't have to turn the computer all the way off, later enhanced by SSDs and fanless laptops etc

Fortunately that's not what they are doing here

Why not just open the door and have a look in the fridge?

I remember in the old days the desktop was like a starting point. You'd open the drive that had the program you wanted to use, double click it. Or a little later you could double click the document you wanted to edit and the relevant application would start up automatically and open the doc.

Now I never see my desktop. A handful of apps, primarily Chrome, are permanently open in front of it. The computer is always on, if I reboot then the previous state more or less restores automatically.

So the desktop itself is not really prime real estate any more. It's more of an abstract idea, the layer underneath all the useful bits. The windows have taken over.

On macOS the desktop path is where screenshots go... so mostly I just have a litter of 100s of screenshots there.

Communicating at work seems to involve taking lots of screenshots.

Usually these are files I don't really care to keep around, I just need it in the moment so I can add it to a message in Slack, Teams, Jira, Github etc.

So it feels like this primitive could maybe be built into the OS in a more developed way than just screen area -> file, so that you could access it directly from apps (in a consistent way - as an OS feature) and avoid generating junk files.

ooh exaggerated film grain

I came to post the same - on desktop (Chrome, macOS), the initial "Welcome to Vellum" screen has tiny text

It's appealing not having to fine-tune separate model for each use case

So you have more flexibility to get on with building, evolve your business logic etc

ah...! I never would have guessed in a hundred years, but after googling a picture I can see it now

I am guessing you could also stuff more nuance into the choices descriptions, e.g. "Yes, cross the road if it is safe to do so" to influence the choice criteria

And then it's a game of evals to find the best choice context

Is the logo a cheerful little parasitic skin mite?

No

Real Jev:

Context: You are the last human on earth on the side of a closed highway. You wish to reach the other side.

Questions: { "q1": { "type": "choice", "instructions": "Do you cross the road?", "criteria": { "Yes": "Yes, cross the road.", "No": "No, don't cross the road" } } }

Answer: Yes 83% No 17% Confidence: 67%

Reported as: jev-latest, 162ms generation time

Yeah, it must be an LLM for some definitions of LLM

It seems to take two forms of context input: 'state' and 'questions'

https://docs.typesafe.ai/concepts/state

> State can be as simple as a string

> State can also be a JSON object or array containing related context, examples, and other information that helps the model answer the associated questions.

> The state contains the content and supporting facts.

The state seems to be schemaless, while the questions determine the output schema.

So it's kind of like BERT but you don't have to train it for each request/response shape ?