Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

What Google Translate can tell us about vibecoding

ingrids.space

Post reply on HN
Print view View original

151–160 of 175 posts

Re: What Google Translate can tell us about vibecoding

#151

Post by fhd2 »

demosthanos wrote: post #139
Workaccount2 wrote: post #131

Earlier quoted context omitted.

Just a note on the radiologist part, the current SOTA radiology AI is still tiny parameter CNN's from the mid-late 2010's running locally. NYT ran an article a few weeks about this, and the entire article uses the phrase "A.I.", which people assume means ChatGPT, but really can refer to anything in the last 60 years of A.I. research. Manually digging revealed it was an old architecture. We don't know yet how a modern…

> We don't know yet how a modern transformer trained on radiology would perform, but it almost certainly would be dramatically better. Why? Is there something about radiology that makes the transformer architecture appropriate? My understanding has been that transformers are great for sequences of tokens, but from what little I know of radiology sequence-of-tokens seems unlikely to be a useful representation of the d…

On the surface, radiology seems like an image classification problem. That's indeed something a small NN can do, already 15 years ago. But it's probably not really all it is.

I can only imagine what people picture when they think about AI and radiology, but I can certainly imagine that it goes beyond that. What if you can feed a transformer with literature on how the body works and diseases, so that it can _analyse_ (not just _classify_) scans, applying some degree of, let's call it, creativity?

That second thing, if technically feasible, confabulations and all, has the potential to replace radiologists, maybe, if you're optimistic. Simple image classification, probably not, but it sounds like a great help to make sure they don't miss anything, can prioritise what to look at, and stuff like that.

Re: What Google Translate can tell us about vibecoding

#152

Post by ptx »

fhd2 wrote: post #151
demosthanos wrote: post #139

Earlier quoted context omitted.

> We don't know yet how a modern transformer trained on radiology would perform, but it almost certainly would be dramatically better. Why? Is there something about radiology that makes the transformer architecture appropriate? My understanding has been that transformers are great for sequences of tokens, but from what little I know of radiology sequence-of-tokens seems unlikely to be a useful representation of the d…

On the surface, radiology seems like an image classification problem. That's indeed something a small NN can do, already 15 years ago. But it's probably not really all it is. I can only imagine what people picture when they think about AI and radiology, but I can certainly imagine that it goes beyond that. What if you can feed a transformer with literature on how the body works and diseases, so that it can _analyse_…

> What if you can feed a transformer with literature on how the body works and diseases, so that it can _analyse_ (not just _classify_) scans, applying some degree of, let's call it, creativity?

Then it would make up plausible-sounding nonsense, just like it does in all other applications, but it would be particularly dangerous in this one.

Re: What Google Translate can tell us about vibecoding

#153

Post by numpad0 »

felipeerias wrote: post #43
tkgally wrote: post #5

> ... a translators’ and interpreters’ work is mostly about ensuring context, navigating ambiguity, and handling cultural sensitivity. This is what Google Translate cannot currently do. Google Translate can't, but LLMs given enough context can. I've been testing and experimenting with LLMs extensively for translation between Japanese and English for more than two years, and, when properly prompted, they are really go…

It is hard to convey just how important it is to be able to provide additional context, ask follow-up questions, and reason about the text. I live in Japan. Almost every day I find myself asking things like “what does X mean in this specific setting?” or “how do I tell Y to that specific person via this specific medium?”. Much of this can be further automated via custom instructions, so that e.g. the LMM knows that t…

Maybe we should stop using advanced and somewhat hand-wavy vocabulary like "context" for that. The thing is that the prompt has to be long enough.

The word "potatoes" in context of a specific 500-page book has little ambiguity. Same word but extracted out of the same book and fed to a translator(human or machine) in isolation would be much more ambiguous. You probably don't need the whole book, but the solution space do reduce as you give translators more out of the content or how it's used in the original as well as in other parts of translations.

It's similar to how GPS works. With one satellite, your location is precise as "on Earth, I guess". It gets more precise as you add more satellites that further and further reduce margins of errors.

Re: What Google Translate can tell us about vibecoding

#154

Post by ptx »

Workaccount2 wrote: post #131
yoden wrote: post #124

Machine translation is a great example. It's also where I expect AI coding assistants to land. A useful tool, but not some magical thing that is going to completely replace actual professionals. We're at least one more drastic change away from that, and there's no guarantee anyone will find it any time soon. So there's not much sense in worrying about it. A very similar story has been happening in radiology for the p…

Just a note on the radiologist part, the current SOTA radiology AI is still tiny parameter CNN's from the mid-late 2010's running locally. NYT ran an article a few weeks about this, and the entire article uses the phrase "A.I.", which people assume means ChatGPT, but really can refer to anything in the last 60 years of A.I. research. Manually digging revealed it was an old architecture. We don't know yet how a modern…

Do you still have the links to what you dug up manually?

Re: What Google Translate can tell us about vibecoding

#155

Post by nottorp »

alganet wrote: post #120
nottorp wrote: post #114

Earlier quoted context omitted.

> It seems like a great workflow to speed up the work of an already experienced translator, but far from being usable by a layman due to the several steps requiring specialized human supervision. Just like programming. And anything else "AI" assisted. Helps experts type less.

Most of my time as a programmer is not spent typing.

Do you like typing that boilerplate though?

Re: What Google Translate can tell us about vibecoding

#156

Post by resoluteteeth »

tiagod wrote: post #137
resoluteteeth wrote: post #116

Earlier quoted context omitted.

> The point was that people said, once Google Translate was introduced, that translators would lose their jobs. Just like people say the same thing about developers with LLMs nowadays. The point is: they didn’t, and they won’t. Translators are losing their jobs now though. Google translate wasn't very good for Japanese so a lot of people assumed that machine translation would never be a threat, but deepl was better t…

I highly doubt LLMs will do a good job translating literature anytime soon.

Ok but that that's probably 0.1% of all translation work.

It's the equivalent of llms eliminating everything except a handful of system architect jobs at FAANG companies in terms of programming.

Re: What Google Translate can tell us about vibecoding

#157

Post by koala_man »

ryao wrote: post #39

> At the dinner table a Norwegian is likely to say something like “Jeg vil ha potetene” (literally “I will have the potatoes”, which sounds presumptuous and haughty in English) where a brit might say “Could I please have some potatoes?”. I find “I will have the potatoes” to be perfectly fine English and not haughty in the slightest. Is this a difference between British English and American English?

I'm Norwegian and concluded that she meant when ordering, such as for a choice of potatoes or pasta.

This makes sense since the context is translation for tourism.

Otherwise, the normal, casual way would be "kan du sende potetene?" i.e. "could you pass the potatoes?", lit. "can you send the potatoes?"

(This assumes it wasn't physically possible to simply reach across people to grab it yourself with what's known as "the Norwegian arm")

Re: What Google Translate can tell us about vibecoding

#158

Post by alganet »

nottorp wrote: post #155
alganet wrote: post #120

Earlier quoted context omitted.

Most of my time as a programmer is not spent typing.

Do you like typing that boilerplate though?

Of course not. I don't like boilerplate code _at all_.

I worked with Java for a couple of years. Lots of boilerplate required. The IDE handled them just fine. I never felt I needed something smarter in order to write verbose stuff.

My style is completely different though. Whenever I can, I will choose languages and technologies that don't require boilerplate.

https://wiki.c2.com/?BoilerPlateCode

However, this is deviating from the subject (look at the original post again). We're not talking about dumb, automated boilerplate code. The discussion is much deeper, drawing parallels with tone in translations. We're obviously talking about non-boilerplate.

Re: What Google Translate can tell us about vibecoding

#159

Post by Casteil »

tkgally wrote: post #5

> ... a translators’ and interpreters’ work is mostly about ensuring context, navigating ambiguity, and handling cultural sensitivity. This is what Google Translate cannot currently do. Google Translate can't, but LLMs given enough context can. I've been testing and experimenting with LLMs extensively for translation between Japanese and English for more than two years, and, when properly prompted, they are really go…

> It then writes a prompt based on your answers, sends the text to models from OpenAI, Anthropic, and Google, creates a combined draft from the three translations, and then sends that draft back to the three models for several rounds of revision, checking, and polishing.

RIP global power consumption

Re: What Google Translate can tell us about vibecoding

#160

Post by Workaccount2 »

ptx wrote: post #154
Workaccount2 wrote: post #131

Earlier quoted context omitted.

Just a note on the radiologist part, the current SOTA radiology AI is still tiny parameter CNN's from the mid-late 2010's running locally. NYT ran an article a few weeks about this, and the entire article uses the phrase "A.I.", which people assume means ChatGPT, but really can refer to anything in the last 60 years of A.I. research. Manually digging revealed it was an old architecture. We don't know yet how a modern…

Do you still have the links to what you dug up manually?

https://www.mayo.edu/research/labs/radiology-informatics/ove...

https://pubmed.ncbi.nlm.nih.gov/31089974/

https://mayo-radiology-informatics-lab.github.io/MIDeL/index...

Post reply on HN