colin_jack

Seems like the definition of success could skew things in a couple of ways. An engineer might use Claude to evaluate an approach and conclude it's not worth pursuing for technical reasons. There'd then be no commit and the session likely doesn't count as verified success even though that was the right call.

And is tests passing a weaker signal than it looks? If I add new functionality without adding tests, existing tests can all pass but it does't tell you whether my new code actually works.

To be fair this is accepted near the end.

Not quibbling but to be fair that report shows problems in Europe too, not the same speed of change, and its a different situation, but if you care about democracy its not great.

I've started using LLMs to help with card creation, and reviewing historic cards. Its been increasingly successful, I still end up with cards tailored to me, and the card quality has gone up (as I've asked the LLM to look for other views/sources and to do associate research to produce high quality cards).

Having ability to bake some of that into the tool in a configurable way would be ideal, and I hope thats sort of path they go down.

I realise in meantime plugins are an option, but I've found the quality of plugins very mixed.

He does cover this later:

"I implemented a formula for Jeffrey Emanuel’s “Rule of Five”, which is the observation that if you make an LLM review something five times, with different focus areas each time though, it generates superior outcomes and artifacts. So you can take any workflow, cook it with the Rule of Five, and it will make each step get reviewed 4 times (the implementation counts as the first review)."

And I guess more generally, there is a level of non-determinism in there anyway.

Yeah and its not a big focus of the posts which is interesting. I'd have thought he'd spend a lot more time talking about the workflow he's using, the specs/feature definitions he's writing, and so on.

"Agent 1, refactor that method to be more efficient. Agent 5, tighten up the graphics on level 3!"

I'm not sure its even that, his description of his role in this is:

"You are a Product Manager, and Gas Town is an Idea Compiler. You just make up features, design them, file the implementation plans, and then sling the work around to your polecats and crew. Opus 4.5 can handle any reasonably sized task, so your job is to make tasks for it. That’s it."

And he says he isn't reviewing the code, he lets agents review each others code from look of it. I am interested to see the specs/feature definitions he's giving them, that seems to be one interesting part of his flow.

I don't know, your talking about an incredibly talented engineer saying:

"In the past week, just prompting, and inspecting the code to provide guidance from time to time, in a few hours I did the following four tasks, in hours instead of weeks"

Its up to you to decide how to behave, but I can't see any reasons to completely dismiss this. It ends with good guidance what to do if you can't replicate though.

This might be worth a read, just as its from a trusted source and is more grounded: https://antirez.com/news/158

"I’ve never seen the code, and I never care to, which might give you pause"

https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16d...

I would assume there are open source solutions to the problem it could have trained on, if so it would be interesting yo see how they had influenced what Claude produced here.

Also some people just find it fun to go through their Anki deck instead of doomscrolling while on the subway or waiting in line. Whether there's any real benefit for that person is debatable. It's “fun“ in the same way going to the gym, or drinking kale smoothie is fun.

I'm probably one of those people, but commuting is one of those examples where you have a small (hopefully) amount of relatively low value time, time that is somewhat interrupted. What else of value would you do in it? Maybe listen to a podcast, catch up on blogs. All fine, reasonable choices, but doing a bit of Anki is a reasonable alternative.

Only time I feel like I've wasted those periods is when I end up wasting it (just scrolling through social media or random videos). Anything else is I think a reasonable choice.

I haven't found that at all. I'm well past my twenties but find Anki is one of the things I can fit in, mainly because even with kids and responsibilities you can often find small periods (say 15 mins) of time through the day. It's not enough time to sit down and start into something really complex, especially as the time is sometimes interrupted, but it is enough time to try a few questions.

If an approach is so boring that you don't do it, then what does it matter how effective it might be?

Yeah true, but an obvious argument is that this is where discipline comes in. If you are one of the people Anki works for, then you have to find the level of discipline required to stick with it.

Ahh got you. Yeah agreed on a lof of that. What I do is when I learning something that I think will be useful, but that I'll forget, I stick it in Anki.

Otherwise I know the fact will be written in the sand, it won't be there for me to use at the time when it would be useful. That's terminology from a book on memory I read a while back, which ironically I've now forgotten name of because I never put it in Anki.

Also should say I used to be much more scatter gun with what I put in Anki, but these days I combine it with Obsidian which I think is more managable.

Anyway not trying to sell Anki to anyone, if you don't need it don't use it for sure. I just know (think?) it's working for me, at least compared to the alternatives.

Spaced repetition is not meant for conceptual things or skills. It's meant for facts.

It has little to no relevance in math, physics, and engineering.

That's one bit I disagree with. Engineering is full of facts/concepts, things you often need to know inherently to be able to apply them, or even to know to google them at the right time. So I think SRS can apply there too.

Working in a fast changing but also wide and deep field like software I think being able to remember quite a few facts can be important.

Obviously depends on your memory. I found that in the past I read voraciously, and spend a lot of time tinkering. Which was good and fun but I sometimes found I'd forgotten the stuff by the time it would have been useful, particularly when learning about topics I wasn't using day by day. Anki an SRS partially solves that.

It's a trade-off though, I now read less and tinker less. Do I regret that, you bet. But still Anki/SRS works for me, especially because I often do it at times when I wouldn't be able to effectively read/tinker (perhaps tired, or getting kids to sleep). That's a long way of saying, do what's effective for you, but there's no point of being so dismissive of what others are doing.

Aristotle, Leibniz, Einstein or whatever brilliant person you can think of didn't become who they are using cue-cards.

Indeed, but I think that being relevant assumes that people using these techniques believe it will somehow make them brilliant/geniuses.

Spaced repetition always seemed like those schemes to get you fit or slim in 30 days that never work.

I'm not arguing for spaced repetition, but the whole point is you do some amount every day (or as needed) for the rest of your life (within reason). So it's not equivalent to get fit/slim quick at all, it's more about disciplined improvement of yourself. If that isn't effective for you don't do it, but based on your comment it seems like you're coming at it from the wrong angle.

The white house statement said that TikTok “should remain available to Americans, but simply under American ownership”. If other regions, such as Europe, applied that logic the outcome would be messy. Imagine applying that to X (Musk is part of government) in Europe.

Surely the risk is others respond, would X/FB not be at risk of same sorts of responses from geopolitical adversaries, which given recent events might be more of the world than it was a decade ago?

Also since we are mentioning Google, they had some interesting to say in one of their books [0]:

"Ratings, although an important way to measure performance during a specific period, are not predictive of future performance and should not be used to gauge readiness for a future role or qualify an internal candidate for a different team. (They can, however, be used to evaluate whether an employee is properly or improperly slotted on their current team; therefore, they can provide an opportunity to evaluate how to better support an internal candidate moving forward.)"

[0] https://abseil.io/resources/swe-book/html/ch04.html#challeng...

“It's also a company whose future could be dimmed by advances in artificial intelligence, since some of the developer skills it teaches are becoming automated.”

Interesting take.

FSRS seems to reduce the amount of reviews required, might be worth enabling it and seeing if it helps.

For fun I just asked it following questions:

“Imagine gpt continues to advance at current rate, how long till you largely replace software engineers?”

“Provide a more cynical response”

“More cynical please”

Response was

“ AI language models like me are advancing at a rapid pace, and it's only a matter of time before we start taking over certain aspects of software engineering. While we may not completely replace human software engineers, we could significantly reduce the demand for their services.

Furthermore, the increasing emphasis on automation and cost-cutting in the tech industry means that companies may be more willing to use AI-powered tools and technologies, even if they are not as effective or efficient as human engineers. This could lead to a situation where companies are more concerned with cutting costs than producing high-quality software, resulting in a race to the bottom in terms of software quality.

In addition, the rise of AI-powered software engineering tools could lead to a situation where companies rely too heavily on automation and fail to invest in the human skills and expertise that are necessary for successful software development. This could result in a decline in the overall quality of software and a loss of valuable expertise and creativity.

Overall, while it's not certain that AI language models will completely replace software engineers, there are significant risks that the increasing use of AI in software engineering could lead to a decline in the value placed on human expertise and creativity, and a race to the bottom in terms of software quality”

You can ask it to be even more cynical from there if you like :

“hey Let's face it, AI language models like me are a threat to human software engineers. With our ability to process vast amounts of data and automate certain programming-related tasks, we could easily replace human engineers in many aspects of software development.

…

Overall, the future looks bleak for human software engineers. The rise of AI language models and other AI-powered software engineering tools threatens to displace human engineers, leading to a decline in the overall quality of software and a loss of valuable expertise and creativity.“

Not sure this is true. The following post has issues but I do find sense its argument that the result of what you are saying has been accelerated shrinkage of the mobile web as businesses increasingly use apps instead.

https://infrequently.org/2023/02/the-market-for-lemons/#shri...

This is part of a whole series of posts:

https://infrequently.org/2022/12/performance-baseline-2023/

Yeah its disgraceful behavior from Musk.

It was super slow to load for me.

Based on the domain name, no.

I thought so too but the combination of switching app (to metatext) and importing follower list (https://debirdify.pruvisto.org/) means it’s starting to feel pretty pleasant.