What it feels like to work in AI right now • Hacker News Board

Re: What it feels like to work in AI right now

#331

Post by otabdeveloper4 » Fri, Apr 07, 2023, 6:22 AM UTC

wuiheerfoj wrote: post #215

cgearhart wrote: post #152

From my perspective it’s just _confusing_ to work in AI right now. We have some massive models that are doing some really neat stuff, and apparently hundreds of millions of people are using them—but I keep wondering: to do _what_, exactly? I’m not asking what the models can do, I’m asking what people want the models to do every day, all the time. I’ve been shown some neat pictures people made that they thought were c…

I‘ve been using ChatGPT to: - write and explain me more optimised algorithms for certain cryptographic operations - explain funny mathy bits of papers that I don’t understand - plan me a few days of activities for a city holiday So far it’s been great on all accounts!! I was able to get a faster turnaround time in understanding the papers than I would if I were probing a colleague

This post is a clever joke, right?

Right...?

Re: What it feels like to work in AI right now

#332

Post by espe » Fri, Apr 07, 2023, 6:23 AM UTC

chopete3 wrote: post #167

At least, people working on ML models that handle these tasks must be feeling terrible. They know their models will be abandoned sooner or later and composed on top an LLM. 1. Classification 2. Named Entity Recognition (NER) 3. Dialog Engine 4. Sentiment Analysis 5. Tone Analysis 6. Language Translation 7. Summarization 8. Tokenization 9. Simple NLP Tasks (part-of-speech tagging, dependency parsing, lemmatization, mo…

actually that might not be the case. don't underestimate the value of older, better understood and much smaller models. also, why not call bert-style (encoder) models LLMs as well. i would expect last-gen models to give us an edge in controlling the effects of the latest ones (cf. the alignment discussion).

Re: What it feels like to work in AI right now

#333

Post by SanderNL » Fri, Apr 07, 2023, 6:27 AM UTC

jrumbut wrote: post #248

nwienert wrote: post #204

Earlier quoted context omitted.

I mean Google revolutionized search. Apple revolutionized personal computing. OpenAI revolutionized… rewriting things with slightly different wording? I’ve seen so many breathless people posting “this would have taken me so long to search” and then I type 3 keywords from their massive prompt they crafted and find it instantly on Google. We’re talking 1000x or more faster. I feel like the same is happening in your com…

I think this is a classic case of us overestimating the immediate impact and underestimating the long term impact. Right now, they are definitely useful time savers, but they need a lot of handholding. Eventually, someone will figure out how to get hundreds of LLMs supervising teams of millions of LLMs to do some really wild stuff that is currently completely impossible. You could spin up a giant staff the way we do…

We call that “companies”. We just need to apply what we learned in business school to a different set of workers, slightly deficient workers.

Re: What it feels like to work in AI right now

#334

Post by avereveard » Fri, Apr 07, 2023, 6:27 AM UTC

riffraff wrote: post #298

jack_pp wrote: post #243

Earlier quoted context omitted.

I thought about learning Go and Fiber to build the backend of a side project of mine and I did but as it goes for any new language / stack I wasn't feeling confident in it. Then ChatGPT came out and I thought what the hell, let's see what all the fuss is about. So I asked it to write me a struct for a table with the "id, name, longitude, latitude, news" columns. That worked well, I was surprised it automatically infe…

> Again, no bugs, perfect code How do you know? Either you know enough Go to be able to tell, and thus you'd be able to write this yourself, or you don't and thus you can't really judge. I mean, it probably is right, but this "I don't know something, but I trust what the chatbot told me" is what worries me about the rise of the LLMs.

On one hand one may not know the exact syntax for a function call but still understand looking at code where a function is defined and where it's called, those are two different knowledge set, the first doesn't transfer from other languages but the second does.

On the other hand you can ask gpt to write test and validate at the outside layer that data is transformed the way you need to.

Re: What it feels like to work in AI right now

#335

Post by eitland » Fri, Apr 07, 2023, 6:28 AM UTC

nwienert wrote: post #204

brookst wrote: post #180

Earlier quoted context omitted.

You’re asking “what’s so big about GUIs? Literally nobody has asked to move a pointer around a screen”. It’s the use cases these thing enable that are important. Today, I wrote a draft product announcement. Only after I was done did I realize I had written it in a really impersonal third person (“users will be able to”). No big deal, but maybe 10-20 minutes of work to make it energetic and second person (“now you can…

I mean Google revolutionized search. Apple revolutionized personal computing. OpenAI revolutionized… rewriting things with slightly different wording? I’ve seen so many breathless people posting “this would have taken me so long to search” and then I type 3 keywords from their massive prompt they crafted and find it instantly on Google. We’re talking 1000x or more faster. I feel like the same is happening in your com…

> and then I type 3 keywords from their massive prompt they crafted and find it instantly on Google.

Seems I and you have different Googles and you still have the one I had pre 2010.

For over a decade now, Google has been including things I never asked about to the point where it would sometimes be easier to find it using Marginalia.

Some say it is just because internet has changed and there is less ham and more spam, but the last few months I have been using Kagi and it proves it is possible to create a better search experience.

And, if Google works for you, fine. Maybe you search other topics, use other keywords or are in another bucket wrt experiments, but from my perspective Google is now the same as its predecessors.

Re: What it feels like to work in AI right now

#336

Post by otabdeveloper4 » Fri, Apr 07, 2023, 6:28 AM UTC

weatherlite wrote: post #272

bmitc wrote: post #207

Earlier quoted context omitted.

Fully functional in what way? As far as I can tell, ChatGPT is a box that I put sentences into, and I get grammatically correct sentences that contains topics or words loosely statistically correlated to what were in my sentences, that may or may not be correct and often are not. The box has little to no memory. I honestly don't see what's so useful about this box.

> and I get grammatically correct sentences that contains topics or words loosely statistically correlated to what were in my sentences, that may or may not be correct and often are not. It's correct enough to pass the bar exam, medical exams, it scores 90-93 percentile on the SAT. This is way more complex and efficient than what you make it to be imo.

> It's correct enough to pass the bar exam, medical exams, it scores 90-93 percentile on the SAT.

So is Google Search, and we had that for a long time now. Is a slightly different and more verbose UI really a game changer?

(Let's ignore the fact that Google Search is broken from all the SEO spam and monetization. Especially when we have no evidence that ChatGPT is any more resitant to this than Google was.)

Re: What it feels like to work in AI right now

#337

Post by SanderNL » Fri, Apr 07, 2023, 6:30 AM UTC

onion2k wrote: post #324

brookst wrote: post #180

Earlier quoted context omitted.

You’re asking “what’s so big about GUIs? Literally nobody has asked to move a pointer around a screen”. It’s the use cases these thing enable that are important. Today, I wrote a draft product announcement. Only after I was done did I realize I had written it in a really impersonal third person (“users will be able to”). No big deal, but maybe 10-20 minutes of work to make it energetic and second person (“now you can…

That's very cool, and right now it's a good idea, but I strongly suspect GPT only looks clever and does a good job in isolation . If everyone starts using it product announcements will start looking very similar, and they'll lose a lot of their impact. This is definitely the case with cover letters for jib applications. The ones written by GPT appear to be pretty obvious - my guesses could be wrong, but after seeing…

I find it interesting it’s very hard to make it do typos. The type of errors humans make. It can do typos but even those seem weird. If I push it to make errors it either over- or undershoots.

The “correctness” of it is a definite give away IMO.

Re: What it feels like to work in AI right now

#338

Post by otabdeveloper4 » Fri, Apr 07, 2023, 6:31 AM UTC

achenatx wrote: post #242

cgearhart wrote: post #152

From my perspective it’s just _confusing_ to work in AI right now. We have some massive models that are doing some really neat stuff, and apparently hundreds of millions of people are using them—but I keep wondering: to do _what_, exactly? I’m not asking what the models can do, I’m asking what people want the models to do every day, all the time. I’ve been shown some neat pictures people made that they thought were c…

its good for so many things: Need to write a speech as a best man Write a eulogy for a deceased family member Recently used it to speed up the time to fix an oauth problem in code I didnt write, knowing nothing about oauth. Replaces stackexchange with much more tailored answers. can take hundreds of pages of text and distill it down to an executive summary of any length Daughter uses it to explain how to solve algebr…

> Need to write a speech as a best man

> Write a eulogy for a deceased family member

Yeah, if you're okay with sounding like marketing copy written by a robot.

Remember last month when Stable Diffusion was supposed to put Disney out of business? Not gonna happen.

Re: What it feels like to work in AI right now

#339

Post by eitland » Fri, Apr 07, 2023, 6:33 AM UTC

spondylosaurus wrote: post #229

rvz wrote: post #216

Earlier quoted context omitted.

> Taking into account the downsides it looks like a hype bubble right now to me, and a draw in the long run. There’s just a whole lot of tech people trying to cash in on the hype. Techies will realize that they are just giving ideas to O̶p̶e̶n̶AI.com, Microsoft Word, Google Docs and Notion. It is just the same AI bros re-selling their hallucinating snake oil chatbot that are under a new narrative for AI. There is a r…

Notion going all-in on the "AI" stuff is annoying/concerning to me. Mostly just that I live and die by a personal Notion wiki to keep my life organized, and if they eventually tank their service by investing too many resources into features that don't take off and I have to find a new tool to offload my brain into, I'm gonna be pissed...

I went with Logseq and for the first time in a number of years (actually since OneNote 2016, the last self hosted version) I am actually happy with my tooling again.

It doesn't cover everything OneNote 2016 did, but it does a lot more in other areas and it is progressing nicely.

Re: What it feels like to work in AI right now

#340

Post by donkeyd » Fri, Apr 07, 2023, 6:36 AM UTC

cgearhart wrote: post #152

From my perspective it’s just _confusing_ to work in AI right now. We have some massive models that are doing some really neat stuff, and apparently hundreds of millions of people are using them—but I keep wondering: to do _what_, exactly? I’m not asking what the models can do, I’m asking what people want the models to do every day, all the time. I’ve been shown some neat pictures people made that they thought were c…

> What am I missing?

I'm going to guess (assume) you probably haven't worked in a 'real' business. A place where elbow grease still does the majority of the work and where Windows 8 was only just phased out.

The killer app (to me) in case of GPT is GPT itself, not ChatGPT. ChatGPT just allows me to easily test use cases for GPT. There are many interesting use cases for those elbow grease businesses for GPT. For example:

Data entry. There's still a lot of data entry being done from unstructured text. Where specifics like names and addresses need to be extracted from letters and emails and contracts and other stuff. I've worked on these challenges before using different strategies and GPT blows my mind with what it can do just by asking to grab this data and format it as JSON. Is it 100% correct? Nope. You still need people to review the data (depending on the use case), but that already saves tons of work.

Categorization. Some companies still get tons of emails that need to be forwarded to specific departments. This is another thing that GPT does surprisingly well out of the box.

And that's just GPT. There are many other legacy business processes that can be automated (partially) by other models that are coming out right now. Even just 'segment anything' that Meta just released is incredibly useful for many use cases I've seen in my daily work.

Killer apps are always a combination of a technology to solve a real world problem. If you don't venture into the real world and only stay part of the tech world, seeing the killer app is very difficult and ends up leading to Juicero-like products.