dartos

I doubt an expert machine’s accuracy would change if you threw more energy at it, for example.

That article seems to be about contracted copywriters.

If your whole job is writing “150-word descriptions” (the article’s words) for $60 an hour and the like, yeah LLMs got you beat.

But that’s a far cry from mass layoff territory.

Also, in that same article, he was rehired by a (granted, a single) client who didn’t like chatgpt. What is the rate of clients that dropped chatgpt?

I can’t think of any technological reasons why every digital system can’t have an API (barring security concerns, as those would need to be case by case)

So instead, we put 100s of billions of dollars into statistical models hoping they could do it for us.

It’s kind of backwards.

The original content model was sustainable for centuries.

Customer pays per viewing, and experiences entertainment for an hour or so.

Then another model emerged after the printing press and record player.

Customer pays once and experiences the same thing as many times as they want.

Now we have a model where a customer pays constantly for no specific entertainment in particular and has no control over when they lose access to some entertainment.

Maybe we just go back to people paying for just what they want…

The issue is the addiction to growth and VC money tech has. We can’t have simple transactions in tech. It needs to be recurring or predatory.

For UX specifically, it makes creating personas so much easier that I think we’ll see a UX once this hype cycle ends.

I don’t know or talk to any PMs, personally.

Out of curiosity, were these announcements by AI companies or companies with an interest in AI companies?

At lot of articles you read are effectively advertising, especially if it’s a CEO talking about their company.

Remember the bipartisan bill to end/reform the H1-B program and that died after Musk tweeted about it?

Tech giants make too much money off of H1-Bs

They mention stdio specifically under the transport section of the spec.

Most protocols are transport agnostic, but http APIs tend to require HTTP methods. Http itself is transport agnostic, but truthfully all we’re doing is shuffling bytes around.

If you look at the stateless examples in that PR, they both assume some state (specifically the capabilities) and just behave accordingly.

To get the full features of the protocol (like notifications) a stateful connection would still be required.

Some people do things because they like doing those things…

when I wrote my other comment in my head TURN and STUN were practically synonymous, since I’ve never seen one without the other.

I’m so used to seeing both STUN and TURN together.

No? If you don’t drill then the carbon emissions from that drilling don’t happen.

I think that fact opens a loophole for carbon credits, actually.

You can have a variety and a single largest.

This is not true in practice.

For a model to perform well looking at ANY X-ray, it would need examples of every kind of X-ray.

That includes along race, gender, amputee status, etc.

The point of classification models is to discover differentiating features.

We don’t know those features before hand, so we give the model as much relevant information as we can and have it discover those features.

There very well may be differences between black woman X-rays and other X-rays, we don’t know for sure.

We can’t have that assumption when building a dataset.

Even believing that there are no possible differences between X-rays of different races is a bias that would be reflected by the dataset.

> Higher in what way?

Higher as in at a higher level in the network stack (not exactly, I don’t think, but lower on the same level)

Higher as in you can implement MCP on top of grpc.

You can’t implement grpc in mcp, that doesn’t make sense, because grpc is at a lower level, so to speak, in the network stack.

It’s not a “mandate” it’s a specification for how the protocol works. If the connection isn’t stateful, then it isn’t adhering to the specification.

I mean also like… how is it stateless? For both example transports (stdio and SSE) a connection must be maintained.

ANY connection must be initialized via an MCP message and both server and client must negotiate capabilities.

Those capabilities must be honored through the entire lifetime of the connection.

That alone is state.

I suggest you read the spec…

For example, a couple years ago there was a statistical model made which could fairly accurately predict (iirc >80%) the gender of a person based on a picture of their iris. At the time we didn’t know there was a visible iris difference between genders, but a statistical model found one.

That’s kind of the whole point of statistical classification models. Feed in a ton of data and the model will discover the differentiating features.

Put another way, If we knew all the possible differences between someone with cancer and without, we wouldn’t need statistical models at all, we could just automate the diagnosis.

We don’t know the indicators that we don’t know, so we don’t know if some possible indicators show up or don’t show up in a given group of people.

That is the danger of wholly relying on statistical models.

For example, If you include no (or few enough) black women in the dataset of x-rays, the model may very well miss signs of disease in black women.

The biases and mistakes of those who created the data set leak into the model.

Early image recognition models had some very… culturally insensitive classes baked in.

We don’t precisely know the most fundamental workings of a living cell.

Our understanding of the fundamental physics of the universe has some hold.

But for LLMs and statistical models in general, we do know precisely what the fundamental pieces do. We know what processor instructions are being executed.

We could, given enough research, have absolutely perfect understanding of what is happening in a given model and why.

Idk if we’ll be able to do that in the physical sciences.

Mcp is definitely not stateless. It’s explicitly stateful…

See the 2nd bullet where it says “stateful connections” [1]

And I was saying that with OpenAPI, you need a client per API, or at least a series of http api calls.

1. https://spec.modelcontextprotocol.io/specification/2025-03-2...

I remember first running into blender in 2009

Well it’s transport agnostic, for one.

I think a big part of it is defining a stateful connection and codifying the concepts of prompts and tools.

Another issue with OpenAPI / swagger is that you still need to create a client per API, but with MCP it’s all uniform.

The US is a smaller EV market than Europe.