Weekend long read on how Hard Fork’s Kevin Roose has oversold new AI tools coding without knowing how to code, as a case study in a larger phenomenon of leading journalists hyping AI: open.substack.com/pub/garymarc... - ThreadSky

garymarcus.bsky.social • 6 days ago

Weekend long read on how Hard Fork’s Kevin Roose has oversold new AI tools coding without knowing how to code, as a case study in a larger phenomenon of leading journalists hyping AI:

https://open.substack.com/pub/garymarcus/p/decoding-and-debunking-hard-forks

Comments

mikefulk.bsky.social•6 days ago

You’re right it’s primarily marketing for big tech. No real objectivity there.

scardeleta.bsky.social•6 days ago

If people stopped learning how to code then so would LLMs because they wouldn't have new training data of people's code to tackle new problems.

lipstickscribbles.bsky.social•6 days ago

Ok, let me explain why tech writers not actually playing the tech game any more than football fans play football, are fucking idiots.

Let's list the problems w/ Roose's mental model for his app & the actual use cases that make his regurgitated code a useless kit bash and not AI.

garymarcus.bsky.social•6 days ago

I regret not having added a riff like this!

lipstickscribbles.bsky.social•5 days ago

The only group that gets me hotter than tech bullshit artists are the human centipedes like Roose, Kara, and others who eat that shit by the foot like candy and think they haven't caught the same mental diabetes.

lipstickscribbles.bsky.social•6 days ago

Refrigerators are morning more than containers. An object that contains other objects. Take a photo of the interior contents, and you get a mix of other objects, which might be an entity (say an apple) but also another container (carton of milk). A carton of milk isn't an entity tho.

lipstickscribbles.bsky.social•6 days ago

It's another container, arguably with a constraint that allows it to only hold the milk entity, but also a volume. But how much is in it? You could perhaps perceive maximum volume from external perception, but is it transparent? Well there's your first big unknown. Everything liquid is a guess.

lipstickscribbles.bsky.social•6 days ago

1. Go outside.
2. Take a picture of your car.
3. Tell me exactly how far you can drive it today until you run out of gas/battery.

See the problem? Roose doesn't.

Sure it can suggest you recipes (not AI or deep btw), but recipes consume the entities in your fridge in varying amounts.

lipstickscribbles.bsky.social•6 days ago

Now add in expiration dates, because if your fridge is like mine techbro, that half full container of chicken stock is probably bad. Well what are the expired entities? I know that because I remember, mostly, when I bought it opened them, but Roose showing up to take one picture doesn't.

lipstickscribbles.bsky.social•6 days ago

Now maybe I take photos of the inside of the fridge after every meal, integrate known constraints gleaned from internet for expiration, still didn't solve opaque container problem though. No idea how much I used, even if I take photos of the finished products (jfc, that's a lot of photos).

cyberphilosohack.bsky.social•6 days ago

These articles are so boring now, make him explain how that code is safe from the OWASP top 10, how it is maintained, the patching process, the feature requests, accessibility, platform portability, integrations, etc.

I will wait

maijav.bsky.social•6 days ago

Genuine question, because it baffles me: how many experienced / senior software architects / developers consider genAI to be in any way useful? I have done some enterprise architecture and I can’t grasp what genAI has to offer to any serious software project whatsoever.

maijav.bsky.social•6 days ago

OTOH, that NYT is again without the slightest application of any journalism propagating misconceptions like this is .. well, very much in character for what NYT has become :<

garymarcus.bsky.social•6 days ago

the AI reporting per se is still good (eg great scoop on Cruise’s huge remote operation crew), but the editorial side has lost its way. and letting Roose write as if he was a reporter is confusing to readers.

opinions should not be presented as breaking news.

maijav.bsky.social•6 days ago

Any time I see someone unironically suggesting genAI for writing software it’s someone who has never written anything beyond trivial spaghetti scripts and definitely never architected or maintained software at scale for any length of time.

lipstickscribbles.bsky.social•6 days ago

As someone whose designed/architected/written/years/deployed frameworks/systems/applications in critical industries, I find AI coding a joke.

A NYT writer doing anything more than kit bash code is hilarious.

lipstickscribbles.bsky.social•6 days ago

Those of us who've played the game at the high end have zero use nor tolerance for engaging a random answer tool when we've already adjusted to modeling systems in our heads before we even touch a keyboard.

AI for coding is like what tee ball is to baseball.

lipstickscribbles.bsky.social•6 days ago

Mastering coding is ultimately mastering language, an abstraction form for expression.

People who are great at it can write code like poetry.

pulsarcube.bsky.social•6 days ago

Architecture work? No (I don't think anyone has claimed that). But I find it saves a lot of time with coding when used as a fancy auto-complete tool. Let's face it, the vast majority of code is not particularly interesting or novel, and I'm happy to be able to offload some of that stuff.

tehomcd.bsky.social•6 days ago

It speeds up some of the tedious bits with sensible autocompletion of docstrings, framework boilerplate, and the like. Is it going to help with some novel optimization? Obviously not. But I've found when grinding out boring-as-hell bits it makes me want to kill myself marginally less

iamjonjackson.bsky.social•6 days ago

Thank you for this teardown Gary Marcus. Insightful connection between Hinton’s radiologist remark.

I’m also reminded of Jensen Huang’s “no need to learn coding” nonsense from a year ago.

(Cross posting from LinkedIn)

dziezycmaciej.bsky.social•6 days ago

@edzitron.com 🤨

emmasculating.bsky.social•6 days ago

Hard Fork’s AI coverage is always a mess. They love to critique the disastrous rollout of a specific product, then follow it up with “BUT, all my Silicon Valley friends have convinced me that AI is going to change the world!!”

samuelosinaike.bsky.social•6 days ago

I dunno, as a data analyst I feel like it can already handle 90%+ junior analyst tasks

abolibibelot.bsky.social•6 days ago

Kevin Roose is notorious for writing a gushing guide to crypto in the NYT as every sign were there that the ponzi scheme was about to crash (before being propped up again). Stellar track record.

abolibibelot.bsky.social•6 days ago

The title is hilarious in itself https://www.nytimes.com/interactive/2022/03/18/technology/cryptocurrency-crypto-guide.html
There are grifters, and there are just dumb marks.

I think Roose is the latter.

ieleino.bsky.social•6 days ago

Not just crypto, Roose was quite optimistic about NFTs too. 😬

garymarcus.bsky.social•6 days ago

Link?

kristinmbranson.bsky.social•5 days ago

It bothers me so much that the @nytimes.com publishes Roose's AI hype. I hope your post will help convince them to stop (I wrote my one and only angry letter to the editor about his previous articles).

kristinmbranson.bsky.social•5 days ago

I think AI will eventually change how we code, and we should think about how to make it a useful tool for making coding more efficient and easier to learn, as high-level languages have. I'm interested in how much knowledge about algorithms/coding are needed to use AI effectively for coding.

kristinmbranson.bsky.social•5 days ago

Given its error level, using AI-generated code reminds me of the difficulty of using somebody else's research-grade code, and what fraction of time I can get away without having to run the debugger through it to figure out what is going on

Comments

Posting Rules

Reply