Hacker Newsnew | past | comments | ask | show | jobs | submit | codemax98's commentslogin

Testing with a different criteria because we heard models are being trained on the pelican

I enjoyed reading this, good insights, doesn't feel like it was written by a machine


uh oh this doesn't seem good. but I feel like we've been inundated with articles like this for years so it's hard to tell when it's actually an emergency


It's been an emergency for years.


Emergencies are things that must be dealt with immediately or there will be immediate consequences.


So a broken ankle in the wilderness isn't an emergency?

Having only 10 days of food when you expect to be 30 days away from resupply isn't an emergency?

Would hate to be stuck in a tricky situation with you (which... it turns out I am)

Emergencies are problems where the consequences emerge close to or earlier than the time required to mitigate it.


Notice how neither of your examples are measured in years?


Notice how neither example takes years to reverse course or mitigate?

Let's just rattle a few more off, across the successfully-mitigated and not-yet-mitigated or failed-to-mitigate categories:

- Hormuz oil shortage (likely to materialize within months if unaddressed now)

- Related LNG → fertilizer shortage → food price shocks (significant shocks over the next few years are already baked in)

- Ozone depletion (successfully mitigated in time)

- Dust bowl (materialized after more than a decade of warnings)

- Military buildup and political entanglements across Europe materializing into WW1 and WW2 after years of warnings

- Bengal famine (materialized after more than a decade of warnings)

- Flint water crisis (materialized after several decades of warnings)

Which of these were not emergencies well before they materialized into direct negative outcomes?

The idea that there's any absolute definition of "problem happening soon" is flatly ridiculous. Obviously the definition of soon is relative to how long does it take to avoid the problem.

A car stopped 100 feet ahead of another car is not an emergency.

A car stopped 1,000 feet from a train is an emergency.

A car stopped 5 miles from a landing 737 is definitely an emergency.

Even if each of these takes the same 2-3 minutes to actually result in a collision.


Very much appreciate you driving this home. I had penned a lengthy response about relative timescales, but you summed it up nicely.


If I see a glass off water on my coffee table placed precariously close to the edge it's concerning.

When my dog starts running around the living room swinging her rope around, it becomes more urgent.

When she bumps the table and the glass starts falling, it's an emergency.

When the glass hits the floor and breaks, it's a crisis.


> When my dog starts running around the living room swinging her rope around, it becomes more urgent.

Because you can intervene before the consequence gets baked in

> When she bumps the table and the glass starts falling, it's an emergency.

Because it will take longer to intervene than it would for the consequence to arrive

> When the glass hits the floor and breaks, it's a crisis.

Because the consequence has arrived

Thank you for demonstrating my point.


There have been consequences every single year. You're just not paying attention.


Absci is bullshit


AI coding agents made the terminal the primary surface again, and the tooling to build good TUIs finally got good at the same time


But why exactly are AI agents using the terminal? Can't they use accessibility tools to read GUIs?


Sometimes. Sometimes they can read screeshots, too.

But why waste tokens when bash (etc.) exists and many of the command line toools are configurable to what they emit when used?


I said TUI in the title, not CLI.


And

> and the tooling to build good TUIs finally got good at the same time

Was in the thing you replied to to ask

> But why exactly are AI agents using the terminal? Can't they use accessibility tools to read GUIs?

Which is what I replied to.


Eww paywall


Yes, we've buried the submission, because it seems there's no paywall workaround.


I love exciting scientific news like this


insane. it literally makes no sense that we're letting him operate this way.


Models follow examples far better than instructions. "Write like this:" with examples


Even with the examples, I've found that explicitly pointing out what not to do is moderately helpful if the model is given some time to self-evaluate. I wish this was something that came out of the box though.


Can you do one for verifying Nigerian princes?

Reminds me of the calls I get from my 89 year old grandma every once in a while, you might be on to something with this.


this is funny since there are thousand tribes with their own princes and a relarive happened to be married to one of them lol.


That's on the roadmap! lol So happy you appreciate it!


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: