Even within the same project, for a given PR, there are some parts of the codebase I want to modify freely and some that I want fixed to reduce the diff and testing scope.

I try to explain up-front to the agent how aggressively they can modify the existing code and which parts, but I've had mixed success; usually they bias towards a minimal diff even if that means duplication or abusing some abstractions. If anyone has had better success, I'd love to hear your approach.

Comments URL: https://news.ycombinator.com/item?id=43709287

Points: 4

# Comments: 1

He invests most of his time visualizing the inputs and outputs of his systems very carefully, instead of focusing super heavily on metrics; this is a lot more effective than I thought it would be, particularly in the early stages of a project.

[1] https://github.com/meta-llama/llama3/blob/14aab0428d3ec3a959...

[2] https://github.com/meta-llama/llama3/blob/14aab0428d3ec3a959...

One thing that helps is hooking metabase up to its own database and building queries on your queries, e.g.:

    select *
    from report_card 
    where dataset_query ilike '%' || {{query}} || '%'

We also try really hard to keep the Collection directory structure clean and consistent. But it's still really hard.

> In computational complexity theory, a numeric algorithm runs in pseudo-polynomial time if its running time is a polynomial in the numeric value of the input (the largest integer present in the input)—but not necessarily in the length of the input (the number of bits required to represent it), which is the case for polynomial time algorithms.

https://en.m.wikipedia.org/wiki/Pseudo-polynomial_time

- You can suggest some other contribution. "Would you mind bringing snacks? / Would you mind handling music on the drive? / Would you mind giving X a ride?"

- You can allow them to reciprocate in less expensive situations, like taking the check when you are at a cheaper place.

Is it inverse frequency, so common bigrams get split last? And the goal is to be able to search on a larger gram that covers the more common trigrams as often as possible?

> At Bagradas (255 B.C. – a rare Carthaginian victory on land in the First Punic War) [...] the elephants disorder the Roman line. In the spaces between the elephants, the Romans slipped through, but encountered a Carthaginian phalanx still in good order advancing a safe distance behind the elephants and were cut down by the infantry, while those caught in front of the elephants were encircled and routed by the Carthaginian cavalry. What the elephants accomplished was throwing out the Roman fighting formation, leaving the Roman infantry confused and vulnerable to the other arms of the Carthaginian army.

[...]

> Elephants could also be used as area denial weapons. One reading of the (admittedly somewhat poor) evidence suggests that this is how Pyrrhus of Epirus used his elephants – to great effect – against the Romans. It is sometimes argued that Pyrrhus essentially created an ‘articulated phalanx’ using lighter infantry and elephants to cover gaps – effectively joints – in his main heavy pike phalanx line. This allowed his phalanx – normally a relatively inflexible formation – to pivot.

[...]

> Thus the war elephant wasn’t a ‘battle winner’ so much as a dangerous complication thrown in the way of the enemy’s plan of attack. And at that, they were awesome.

Props to this professor; struggling in confusion is a lot more fun if you know that's the game you're supposed to be playing.

Physical mastery often looks relaxed, natural, and simple, because all extraneous effort has been removed. When you're training hard to reach that state, the "just" can really sting. It feels like: "not only are you bad at this, but it's simple to not be bad at this".

If you don't mind some off-topic feedback on the Dominion website - have you considered making the Tables screen default to "New" only? At busy times, that screen consistently lags for several seconds while displaying hundreds (thousands?) of running and finished games. It's always a pain to wait for it before de-selecting "Running" and "Post-Game".

Also, am I crazy or do the celery docs not even clarify their delivery semantics? Isn't that table stakes for a queueing system? As best as I could tell you can get close to "at least once" with

  acks_late=True, task_reject_on_worker_lost=True