Yes - but also there are numerous examples of people generating highly functional code. And that fact that this code is at a none working state does not imply that the entire 5k LOC are be distracted, there is probably a lot of business logic coded in there.
The issue here is that the author is now forced to use AI to review the 5k lines and he seems to opposed to the idea hence my point that these developers will eventually be forced to adapt.
It doesn't but it also will not create a different sentence to you to practice a word.
I tried this (use sentences instead of words), it sucks: you spend more time per card, but you learn less efficient, as you learn not just a word but all the hints coming from the same sentence.
I wrote my own program 2 years ago, that generates new sentence on each review using AI (at that time I used Anki as scheduler). That was fun and quite effective as far as I can tell.
I would recommend others learning language past A1, maybe A2 level to do the same. This day you don't need Anki for this, just use one of FSRS libraries for scheduling.
> It doesn't but it also will not create a different sentence to you to practice a word.
Anki doesn't create either words or sentences. Drilling words doesn't work. You've drilled words and found it doesn't work. Try using clozes. It is not "less efficient" than drilling words, because drilling words does not work.
No word in French means anything in English. The meanings of words are just surveys of the contexts in which they are used, and French words are used in French contexts. Learning that a French word means an English word is just bad information that is cluttering up your brain.
If you're learning Spanish, French, German or Italian, try these.
20,000 clozes, words are repeated all the time, treat them as disposable. Read the sentences out loud and say the answers out loud to keep yourself honest.
edit: you don't need AI to generate sentences. There are already sentences available in every language.
I ran into that issue too w/ sentence-based flashcards, where I almost immediately memorize the sentence itself and the whole thing becomes self-defeating. Similarly, I thought to use LLMs to generate fresh sentences on-the-fly, but the output was never reliable enough for my use-case... e.g. I came across some grammatical construction that the LLMs refused to use correctly. But, this probably depends on the language in question, since it sounds like you had success with it! Maybe I should try again at some point, now that the consumer models are a lot beefier than they were even six months ago.
I used it for practice Spanish. LLMs have no issues with generating decent sentences in it. Also as I practiced vocabulary building, I didn't need specific grammar constructs or even to be very sensitive to the occasional mistakes in this area (which I don't remember to notice).
It took me quite a bit of iterations to get AI to generate output I found good enough and with acceptable errors rate for my practice but that was the part of fun.
If anybody wants to try the same route, the biggest mistake not to repeat is to trust LLMs to generate words definitions by them itself. It does it quite well, but still gives wrong definitions from time to time (~1% in my case for the OpenAI models 2 years ago). This wasn't acceptable to me and took a lot of effort to review and correct cards I already had. Feed vocabulary article with word definition to LLM instead if you want it to generate word definitions suitable to your level instead.
I'll throw a shout out to the new Google Translate practice feature - it generates sentences around a theme you specify and speaks them to you at varying speeds in your native and learning languages.
Personally, I feel this isn't really an issue, as someone who has learnt multiple languages to proficiency (some with anki involved, some without). I mean I'm at a level where I speak mostly fine, read historical literature, novels, academic/political works, and engage with significantly divergent dialects without any trouble (I'm interested in such things, so.)
You find a word you don't know, you add a sentence. Find a different use in the wild or in the dictionary, add a sentence. Want to harvest many examples at once? If your dictionary lacks example sentences, you should really find a better one (of course, dictionaries with historically oriented/early-attestation examples are not neceessarily bad - many of the best dictionaries are such - but they're not great for mining examples of usage). (Granted, some languages simply don't have dictionaries that meet this criteria, in that case you must work with what you have.) Corpora of speech or literature can also be used if they exist. Most of this process can be automated to 1-2 keypresses.
Yes you get hints from the sentences, but you do too when reading/listening. Seeing a single word in the wild is also likely to lead to some free association shenanigans - ask any native speaker of whatever to define a polysemous word. Multiple sentences can just be added for one word, as needed, though the 'bulk' of sentences for 'review' is going to come from exposure in the wild.
In my own experience, word cards are less efficient because they accrue many more reviews - even for languages closely related to my L1 - having higher average difficulty and a higher fail rate. I would hazard a guess that I could review ~20 sentnece cards across my lifespan in the time it takes to learn ~1 non-trivial vocabulary card (ie. not words with zero polysemy, technical terms, things that don't occur in idiosyncratic phrasal constructions with specific converbs/clitics/prepositions/case usage abnormalities or syntactic abnormalities). And you get reinforcement of ~15 other words for free (depending on how long the average sentence is in your TL). I never worry if I end up adding a couple of sentences 'for' a single word.
It's basically targeted re-reading/listening, those being things that one does anyway and which accounts for the bulk of learning vocabulary.
AI is a non-starter for me (personally) in this for two regards:
1) they're awful at most of the languages I've studied at a trivial level.
2) those that they're 'good at' (in terms of grammaticality) they're often just horrible stylistically. I would like to be able to speak/write in a way that is pleasing in nuance/idiom, not just comprehensible.
But if these don't really apply to you or don't match your experience with your TL (they're better with some languages than others), more power to you!
Very sad. I never took his courses, for various reasons, but having something like this as an option available was quite inspiring in my never ending quest to become a better software engineer.
I wish David the best with his new teaching direction.
And I hope David will find a practical way to make these courses available for self-study.
I have very similar experience. I vibecoded a foreign language practice app for myself. It works decent from functional perspective and I don’t see too many bugs. But the biggest productivity constraint I see is the time I need to spend using it in order to understand what is working and where the issues are.
„I was able to vibecode those 5 apps I always wanted but never had time to code them myself … it is so different now because — I don’t have time to use them”.
That is wrong - the limit isn't time. The limit is you need to create a good ui so you don't need time. time is the limit for a fun game (see other comnents on how hard this is).
However an app needs to be a good enough ui that it does something for you and often they don't. Paper todo lists still beat all the todo apps people try to make because the ui works for paper. Paper has large limits and so it seems there must be an app out there that is better - but so far everyone has failed to create it.
Idea behind the joke is that those apps were not needed at all in the first place, it is only that person from the joke had grand idea lurking in their head that those will be super useful. If those apps would be needed a person would take time to build them. Now building useless apps costs much less.
Better UI doesn't matter if something is purely a "wannabe application" in someones head.
Ergo, if you don't have time to build out some idea - that is a clear indication such idea is not good or not so useful - even if you still have an urge to build it, subconsciously you already know.
You absolutely can. This is one of recommended directions with agentic coding. But you can go farther and ask llm to write tests too. The review/approve them.
There is quite a bit of difference between not making a profit and consistently losing around $100m a year with apparently no path to at least revenue neutrality.
So it loses pocket change for a multi billionaire?
Edit: The consideration being that perhaps billionaire toys need not be profitable per se, but are purchased for different reasons. Twitter is another example here.
That's assuming the pro-billionaire propaganda it produces doesn't make him many hundreds of millions more.
In that light an arbitrary but vaguely plausible reason to fire anyone who insists on doing actual journalism and not billionaire propaganda is a useful tool.
This is a distinctly Zed solution - trying to move the agent experience into the editor, rather than just giving the agent an interface with which to control and read from the editor.
Not only do the most popular editors have little-to-no incentive to implement it (they’re more interested in pushing their own first-class implementations, rather than integrating those of others), it’s much more work to integrate the evolving agent experience into the IDE than it would be to provide IDE integration points for the agents themselves.
So, I think this project would have been much more successful if it had been more focussed on keeping the agent and IDE experiences separated but united by the protocol, instead of trying to deeply marry them. But that’s not in line with Zed’s vision and monetization strategy.
It won’t be long before the big players start to release their own cloud-based editors. They’ll be cloud-based because the moat is wider, and they’ll try to move coding to the cloud in the way that Google Workspaces moved docs to the cloud. Probably with huge token discounts to capture people. If you squint, you can already see this starting to happen with Claude Desktop, which runs its agent loop on the cloud (you can tell because skills appear to need to be uploaded).
Notably, Microsoft, with VSCode and GitHub have a web-based editor advantage in this space, but no models.
It's not just Zed, Emacs has has a thriving ACP implementation in agent-shell[0], and allows for some very cool integrations[1]. There are a fair number of other clients[2] as well.
The second half of this is spot on. The now is making IDEs that can integrate with agents, not the other way around. Soon the Claude and Codex will do that for us on their hosts and the argument is it will save sending the context up.
Generally it is a good advise, I found similar things very useful for me.
I think emphasis on paper and one notebook is wrong here and likely will fail quite a bit of people who will try it. Also prescriptions like "They're very detailed" (i.e. notes) are IMO too rigid.
Start from wherever suit you, play, experiment and pay attention what works for you, adjust and iterate. Don't fixate on shiny concepts, i.e. "engineering notebook", and the "need" its records to be dated, etc.
Try something, let it lapse. See if you are worse without it, then adopt it back. If you don't see the difference, so be it.
Huge amount of *non-working* code, as illustrated by this article.
Also author is not super clear on this point, but according to article the partner has programming background.
reply