fallingmeat

in other news, local man rounds up on grocery bill for charity

"make more better tests to decrease yolo" haha

Wow, good sleuthing!! dang do you have a tool to do that or are you just encyclopedic?

Is this the promise of Amazon Sagemaker?

still not sure of the value of the langchain abstraction. does anyone use it in prod?

Fun fact: the author of these rules (Gerard Holzmann) is also the inventor of the Spin model checker (spinroot.com)!

Thanks! fixed it now that I'm back from vacay...

neato! anyone interested in a collab on a more generic natural language graph making tool (think GoJS powered by LLM to make SysML diagrams etc). Demo of api here, I cant do front end, it is a weakness…

https://grapher.baselines.cloud/

hit me up on linkedin: /in/thenatefisher

10% of "revenue"? For the amount of energy put into making this language model, the business model effort seems to be disproportionate. I'm glad they made it commercially available, though! I hope they see success here and that success drives further development.

Great question! I don’t know the end game there. Maybe if they suspected their model was used they would sue, and in discovery find you used their model for training?

That is against their ToS though if you use your new LLM commercially.

similar to the DARPA active flow control demonstrator that just received is designation (X-65):

https://www.darpa.mil/program/control-of-revolutionary-aircr...

I recently made an LLM-based tool to identify inconsistencies with the INCOSE guide to writing requirements (identifying vague language, for example). Also have had some success translating those outputs to more structured language like LTL. I think it is doable

Most high integrity software development includes roles for systems engineers that do this (avionics, medical, nuclear, rail, automotive). See standards like ARP4754A, INCOSE working group on requirements engineering for leads, or if you are really into it, applications of formal methods like DO-333

google “overjustification”

1 point by fallingmeat 1 minute ago | next | edit | delete [–]

Why do similar articles or headlines seem to come up at the same time on front page? e.g. there is another article on Rails that happens to be front page right now that is seemingly otherwise unrelated (the other article is called The Rails Doctrine).

Why do similar articles or headlines seem to come up at the same time on front page? e.g. there is another article on Rails that happens to be front page right now that is seemingly otherwise unrelated (the other article is called The Rails Cheat Sheet).

haha thats good

Meh. IMO it’s not particularly great at impersonating things convincingly without fine tuning or thorough prompting.

Example, ask it to be an air traffic controller. You will need to spend a lot of time making it sound reasonably accurate.

yep. good question! no idea

procedure design. if you want to design your own approach procedure, for example. or if you are designing a takeoff/landing facility.

That’s brilliant! Very cool way to inspire.

Next they can move on to TERPS procedure design! haha

It's not about how hard it is to make, it's about how willing someone is to exchange dollars for it. I think Stew and co. are just trying to rapidly find paying users that can help them steer to a better product vision, which they can then build out.

Don't make a startup to do hard things. That's what a hobby is for.

Just tried this with ChaGPT for giggles. It spit out words, but I wouldn't call it sound reasoning.

__user__:

- A system in composed of two elevators. - Each have their own unique velocity parameter. - At each time step in the system, the elevators can move up or down a number of floor equal to their velocity, or change their velocity. - Assume that both elevators move at the same time step. - The elevators may never occupy the same floor.

- Elevator A starts on floor 11, and Elevator B at floor 3. - Both elevator velocities start at 1.

Is it possible for A to ever reach floor 2?

__assistant__:

No, it is not possible for Elevator A to ever reach floor 2 while maintaining the rule that the two elevators may never occupy the same floor.

To see why this is the case, note that Elevator A starts at floor 11 and Elevator B starts at floor 3. Since both elevators move at the same time step, Elevator A can never move to a floor lower than Elevator B without violating the rule that they may never occupy the same floor.

Furthermore, since Elevator A and Elevator B have the same velocity at the start and Elevator B is already on a lower floor, Elevator B will always reach any floor below Elevator A before Elevator A can get there. This means that Elevator A can never occupy any floor that is below the current position of Elevator B, including floor 2.

Therefore, it is not possible for Elevator A to ever reach floor 2 while maintaining the given constraints.

__user__:

what if at the first time step, both velocities became equal to 5?

__assistant__:

If both elevators have their velocities set to 5 at the first time step, then it is possible for Elevator A to reach floor 2.

With both elevators moving at velocity 5, Elevator A will reach floor 6 after the first time step, while Elevator B will reach floor 8. Since Elevator A is now on a lower floor than Elevator B, it is possible for Elevator A to continue moving down to floor 2 without violating the constraint that the elevators may never occupy the same floor.

Specifically, Elevator A can continue moving down at velocity 5, reaching floors 1 and 2 after two more time steps. Elevator B, with velocity 5, will reach floor 13 after the same two time steps.

So, if both elevators have their velocities set to 5 at the first time step, then it is possible for Elevator A to reach floor 2 while maintaining the given constraints.

Yes, if they were to move any number of floors at the same time step. But the language model isn't necessarily going to automatically discover those rules for you. Even if it did, the model won't necessarily be able to frame the problem as a BDD and solve it.

At best (and the is a valuable trait) the LLM would be able to reframe your problem in such a way that you could discover the necessary precision in the problem statement. Perhaps it could even encode it in some syntax for another tool.

sorry, that wasn't my point. I just meant, there are tools for that job. LLMs don't seem to be the most appropriate tool. Thanks for helping me to clarify

No it's not, it's a language model. It manipulates words. If you want reasoning, you should find the right tool. For example, encode your problem for an SMT solver (like Z3, NuSMV, etc).

Let's say you have an elevator start on floor 11, and another at floor 3. The rules are, they can never occupy the same floor. Is it possible for the first elevator to ever reach floor 2? No, they would have to pass each other.

GPT-4 may be able to come to the conclusion by accident. But not reliably.

do you foresee sysml like diagrams in your roadmap? if you could do activity diagrams with swimlanes…dude. that'd be huge.

I like where your head is at! make that the vision.

/fire suit on

“… TTP/Global Entry sort of fixes this…”

wish that were true, but the variance of wait time, even with tsa precheck, etc defeats the purpose. needs to be reliably consistent (and short) for me to take advantage of a 30min show time.