dartos

I know because that’s how these systems are built.

I’ve built systems like it.

If it was something brand new, Anthropic would be bragging hard about it.

> You could 100% create the tool to search and chose results, go through links, read more pages, etc.

That’s exactly what I’m saying. _YOU_ could build a tool that does that. The LLM essentially acts as an intent detector, not a web crawler.

If you train on any test material, the tests are worthless.

A (afaics non peer-reviewed) paper published by a random organization endorsed by Anthropic does not proof make.

In fact, it still very much seems like marketing. Especially since the paper was made in association with Anthropic.

Again. Learn some media literacy.

> dialog is a intuitive interface that users enjoy

Do people enjoy chat interfaces in their workflows?

I always thought that cursor/copilot/copy.ai/v0.dev were so popular because they break away from the chat UI.

Dialog is cool when exploring but, imo, really painful when trying to accomplish a task. LLMs are far too slow to make a real fluid conversation.

I may be missing something, but if you train a model to pass a specific test… isn’t it obvious that it would do better on that test?

I thought we called models with test data in their training set “poisoned”

You should look up the definition of “universal”

Splitting hairs, but LLMs themselves don’t search.

LLMs themselves don’t choose the top X.

That’s all regular flows written by humans run via tool calls after the intent of your message has been funneled into one of a few pre-defined intents.

If we’re being technically correct, yes, but I was just trying to give an analogy to someone who I figured was pretty new to web tech.

I can read pretty fast, but prefer videos for introductions to any new tech.

Then if I decide I like it I read the manual.

I think using a card is friendlier than using an app.

It takes fewer “clicks” so to speak and is more accessible to the elderly.

But anything government which doesn’t require to keep track of a purchaser’s identity should accept cash. Otherwise what’s the point of cash if your own state doesn’t accept it.

Metal tokens that don’t change in functionality at the whims of a random VP at some private company?

Yes. High tech != smart.

Why is it beyond stupid to allow payment using a currency without needing several layers of non-government middlemen (Apple and ParkMobile, for example)

Seems beyond stupid to require me to have an Apple program or a Google program and specific hardware to pay for parking when I can just have the coins in my pocket.

Especially since ParkMobile could just decide not to support the version of android I’m running, so that I’m forced to purchase a new device.

Coins aren’t subject to code rot.

A bonus is that meter toll collectors would have a job.

There’s a bit of truth to all of what you said.

These tools aren’t useless, obviously.

But people do really learn hard into confirmation bias and/or personification when it comes to LLMs.

I believe it’s entirely because of the term “artificial intelligence” that there is such a divide.

If we called them “large statistical language models” instead, nobody would be having this discussion.

That’s a marketing article, bub.

You should learn a bit about media literacy.

MCP is analogous to REST.

It doesn’t define what behavior should be where in a given app, just how to communicate what that behavior is and how to invoke it.

Yeah, quants are another breed.

The real questions wrt the title is:

Will specialized models also hit a usefulness wall like general models do? (I believe so)

And

Will the model’s blindspots hurt more than the value a model creates? (Much more fuzzy and important)

If so, then even many specialized models will be a commodity and the application on top will still be the product end users will care about.

If not, then we’ll finally see the return on all this AI spending. Tho I think first movers would be at a disadvantage since they need much higher ROI to overcome the insane cost spent on training general models.

I don’t think those 2 problems are mutually exclusive.

I think it far more likely that we have an intelligence collapse.

If AI becomes good enough to replace intelligence workers (pretty sure it won't anytime soon, but let’s pretend) what’s the incentive to be intelligent?

Once being smart isn’t marketable, who’s going to study?

Once it doesn’t pay, who’s going to go to college?

Any gain humanity gets from offloading thinking to machines is offset by us doing less thinking.

It’s why “computer” is no longer a profession all by itself.

Do you remember when people also thought rabbit would be a revolutionary AI device?

You have it all wrong.

People are not equations.

Let’s define risk in this discussion.

Risk is the relative impact your decisions have on the org at large. If you make the wrong choice as the CEO, you can tank the company.

If someone’s job involves risk, and you pay them peanuts, they’ll either go to a job that doesn’t involve that much risk or make decisions that are always maximally safe, regardless of potential upside, usually at the cost of company growth.

You want your staff to feel comfortable taking risks so that the company can grow, because a stagnant company will die. So you pay your positions more when they have more risk.

> The CEO asked in reply what happens if we don't train the people and they decide to stay?

Then they both chuckle and say they’ll hire some other schmuck, it doesn’t matter bc they already got series A funding.

> instinct to take the path of least resistance.

It’s least resistance initially, but much harder in the long run.

It’s like the children with marshmallows experiment.

Give a kid a marshmallow and tell them if they don’t eat it for 20min, they’ll get 2 more.

Some kids will eat it right away and some will wait.

We’re just seeing that play out for adults.

That’s why we use math to describe nature in a way that fits in our brains.

That’s the whole point of math.

But why stop at “some” pattern when you can find the most general pattern.

The AI photos aren’t from that era either… they don’t show accurate depictions of life from that time.

They literally can’t. There’s no photographs…

What’s the point? You’re just guessing at the prompt used to generate the photo.

That might be more interesting, actually. AI photo game where you need to guess the prompt.

Well there be any consequences for the company?

Weird argument, but I’m down to bikeshed a bit.

most of the time there’s an old ticket in the work tracker.

I’d bet those tickets contain at least as much info as any prompt.

I was excited by the title until I saw it was all AI :(

It would’ve been cool to collect actual images from history. I’m sure there are 1000s of public domain images that could be used.

Me too. That resonates really strongly with my feelings.

I’m hoping that most people aren’t full steam in AI.

I haven’t had any coworkers who just rely on AI… I have had some bosses who do tho…