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

Ask HN: GPT4 Broke Me

news.ycombinator.com

Post reply on HN
Print view View original

171–180 of 183 posts

Re: Ask HN: GPT4 Broke Me

#171

Post by remich »

leke wrote: post #149

I was speaking to a lead dev today. He said he thought the biggest risk AI poised was to junior devs. He currently gives specs to junior devs who do the work and then he reviews it. He thinks it's nearly possible to give the specs to AI, it does the work, he reviews it. So maybe no more need for junior devs. He thinks this is especially possible when AI has access to your whole framework and product.

Except, as we saw with the Google article today, companies will soon get wise to the fact that feeding their product source code into a cloud-based GPT model is a great way for that source code to leak out.

Sure, maybe OpenAI or Google will sell a "for-business" model for thousands upon thousands of dollars a year, but will people really trust that it's not phoning home? Will there be rate limits on requests that make it not make sense to use it for everyday tasks?

Re: Ask HN: GPT4 Broke Me

#172

Post by remich »

hnlmorg wrote: post #35

The value developers add isn't writing code, it's solving the problems. More so for the more experienced end of the roles, such as you said you are. If it were that easy for AI to step in and replace developers, then we'd have seen the same thing with "codeless" frameworks. Maybe one day in the distant future it will be a different story. But I'd wager you'd be dead before that happens.

Yep. No-code/low-code tooling never really took off beyond some small use cases for a pretty simple reason that will also apply to non-engineers using GPTs:

There is always a tension between having a simple-to-use tool and getting a custom result. The simpler the tool is to use, the less likely you'll be to get a custom result. Tools that are great at generating custom results are rarely simple to use.

GPT models demonstrate this very well. If you attempt to prompt it without any knowledge of a field you can get something back... but it won't be unique. To get a workable result to a more complicated problem you need to know how to prompt it, how to review what it spits out, how to modify your prompt in response to correct it, how to stitch the eventually working piece with other pieces. And so on.

Re: Ask HN: GPT4 Broke Me

#173

Post by remich »

psytrx wrote: post #34

I'm in a similar position. Top x% in my field, more than fair compensation, somewhat financial security. I worked my butt off towards burnout to get there (currently slowly recovering). I recently listened to a conversation between Lex Fridman and Stephen Wolfram [1]. Stephen said something that stuck with me for a while. Something along the lines: GPTs are statistical word guessing machines. They regurgitate the ave…

The coming LLM content ouroboros is a real problem. Garbage in, garbage out.

Re: Ask HN: GPT4 Broke Me

#174

Post by remich »

naveen99 wrote: post #148
BizarreByte wrote: post #120

Earlier quoted context omitted.

You don't think the law could basically make them infeasible? Right now they rely on leeching off content generated by real humans. Cut that off and they're going to have issues even training them. Get something by Disney included in the training data by mistake and just watch what happens.

I don’t think god can stop it now, not disney, not the lawyers, not even the communists. Your best hope is LLM’s in the current state aren’t enough, and we need to wait for another algorithmic leap or a lot more computational power improvements before we get to human level.

Litigation stops a whole lot of things in its tracks. Copyright stuff barely scratches the surface of the legal issues these things raise.

Other things that are soon to come - trade secrets no longer being trade secrets because company staff are feeding internal documents and source code into a cloud model (a practice that doesn't really align with "reasonable efforts to keep something secret"). Or even just the basic question of, who gets the blame when something generated by an LLM causes a significant problem? It doesn't take much in the way of judgments or settlements before an org realizes the risk of un- or under- reviewed output simply isn't worth it.

Re: Ask HN: GPT4 Broke Me

#175

Post by thaway_thaway34 »

Parka3458 wrote: post #96

In 2019, I read "AI-Superpowers" by Kai-Fu Lee, where he went through most of the worries we have today. In the final chapters however, he started talking more about how we, as humans, need to adapt, and he also mentioned UBI as a (possible) way forward. At some point, AI indeed will "replace" what people like you (and me, who has pretty much the same background as you) do. We are just tools that knows how to code. Q…

Scifi just became real, and I don’t like it.

In many ways GPT-4 does a better impression of a human than any we have seen in Star Trek. Data couldn’t understand humor.

We were so sure of our own complexity, we projected onto the AI characters weaknesses. Then an AI comes up doing jokes in any topic you throw at it in any way you like.

Yet still some suggest I must be trolling with my post. I know my take is quite pessimistic but far from exaggeration, I didn’t even get as far as suggesting any SciFi level dangers in longer term.

Re: Ask HN: GPT4 Broke Me

#176

Post by DougN7 »

ryandrake wrote: post #152
thaway_thaway34 wrote: post #127

Earlier quoted context omitted.

>> I've always felt like with new technology, I could grok at a high level how things worked. But LLMs like GPT seem like magic This is exactly how I feel. I felt so out of my depth looking at the ML architectures and I could not make any sense of it. I thought perhaps, they get inspired by neuroscience for the layers etc. But a friend who works on LLMs mentioned, the architecture of large ML models, are mostly exper…

Checking in with the same feeling. If I had to do an interview and they asked me to sketch out on a whiteboard a high level diagram of how anything from the last 20 years of computing worked, I could probably muddle my way through it. A 3D engine, a database, a word processor, a web site with a REST API, you name it. It might not be 100% right in the details, but I could at least describe it in the general sense and…

It’s OK. From what I’ve read, nobody actually knows how it works. I mean we know we have layers and weights, but how those emit what looks like intelligence is not understood by anyone.

Re: Ask HN: GPT4 Broke Me

#177

Post by sitkack »

30 year mark. These LLMs will change what is attainable. If we continue to just build the same, those less skilled will be able to achieve. You can now do things that took a team. Embrace that.

Re: Ask HN: GPT4 Broke Me

#178

Post by sitkack »

cj wrote: post #97
me_me_me wrote: post #84

Earlier quoted context omitted.

> Technology has been displacing jobs for decades. Life is about learning and adapting. As long as you commit to adapting quickly, learning new skills as necessary and being open to different types of jobs, you should be fine. That seems to be a contradiction of your story. Your dad was lucky to be able to retire. He did not adapt. He had to quit and he got low end job. Its easy to say adapt re-skill, innovate but to…

Yea my dad isn't the best example. He had a business partner that was refusing to buy the expensive scanners and 3d printers because the business partner didn't want to make the investment in new technology. They stayed old school and went out of business. It's probably more of an example of how not to handle this sort of situation. Instead of being afraid of or avoiding the technology coming for your job, embrace it…

My dentist does this in-house now using machines they own.

Re: Ask HN: GPT4 Broke Me

#179

Post by pillefitz »

constantcrying wrote: post #147
mclightning wrote: post #123

Earlier quoted context omitted.

why are people in such a denial on this point, makes me mad... the same field who has the most imposter syndrome rants, also has most people who think they can't be replaced by someone with less experience equipped with AI. Everyone is talking about the junior, mid-tier programmer who will replaced, but not themselves. It just sounds like willful ignorance.

>who think they can't be replaced by someone with less experience equipped with AI. Because in a shrinking market experiece + AI is vastly superior to little experience + AI. Experience is becoming more important, since experience is required for the tasks AI is very bad at. AI is drastically increasing the barrier to entry, since it can do many of the things entry level people are/were useful for, but it can't do an…

The points described in your last paragraph seem to be well in reach of a GPT model.

Re: Ask HN: GPT4 Broke Me

#180

Post by random_cynic »

constantcrying wrote: post #27

Nothing I have seen from ChatGPT has made me worried for my job. If your job is producing 20 line snippets solving variations of common coding tasks, ChatGPT is the least threat to your job security. To be honest the far greater threat to your job is an increase in the ammount of programmers plus an increase in individual productivity, with an industry declining in growth. There is your threat.

You're in denial and delusional. Most people in this thread are.
Post reply on HN