Ask HN: Are tech layoffs related to developments in AI?

5 points by lukeplato ↗ HN
I have seen a lot of big companies announcing large lay offs in the tech industry and most discussion seems to related to over-hiring and the recession. Is it possible that the layoffs are because of insider knowledge involving a reduced need for software engineers as LLM code generation improves in the upcoming months?

21 comments

[ 4.2 ms ] story [ 51.9 ms ] thread
IMO it is quite possible, as anticipation of productivity to come. But, more likely caused by over-hiring in enterprise, by empire-building management.
If that was happening (and I don't think we're there yet, or will be there in the next six months), I don't think they would lay people off pre-emptively, and it takes a while for any company to shift their processes over to a new way of doing things significantly anyway. It would likely happen after it starts to actually make people redundant, not before.
Are tech layoffs related to developments in AI?

I do not believe so, at least not to any significant degree. Rather I believe businesses are adjusting to global inflation and the global economic current and soon to be changes.

This will be a wildly unpopular opinion right now but I believe most of the talk around AI is just that. Machine Learning and Big Data have been rebranded with new window dressing. I suspect people will tire of it quickly once they realize it has no ability to show it's work and is just the next algorithmic evolution beyond the tools used in social media using tailored data-sets and easy to manipulate algorithms. What people are calling AI is impressive however until such a time this AI has truly open source code with completely open obfuscated data-sets and can be self hosted with easily reproducible outcomes then history has taught me to distrust the great and powerful Oz.

It's related to developments in capitalistic greed more than anything.
Was it ever non greedy? All investors care about is earnings . Employees are a drag on earnings, so getting rid of them is a built in feature of the system. Its called "efficiency"...
No.

What is called "AI" today is just brute force machine learning. There is no intelligence, as ChatGPT has amply demonstrated, just blind regurgitation of a black-box statistical analysis of data.

True AI, i.e., the kind of AI in movies, is still several decades away. It's actually further away now than it was 10 years ago, as the brute-force method has pushed AI research into a local maxima that will be difficult to get away from without starting over.

What makes you think we are in a local maxima? Looking at GPT1 - GPT2 - GPT3 - ChatGPT progression there's plenty of room for improvement left, I don't see any signs indicating that GPT4 won't be as big of an improvement over GPT3 as GPT3 was over GPT2.
Okay, it's worse than I thought. We haven't even reached the local maxima yet: we're still on the way up to a local maxima.

The problem is that the hill "AI" is currently climbing is just the hill of blind statistics; this method of AI will never be capable of developing understanding.

GPT puts words together based on structure and statistical analysis of what were seems like it is most likely to come next. But GPT doesn't understand the prompt, so the result is frequently nonsensical: it looks structurally sound (and is, because grammatical structure is easy), but most of the time the output doesn't mean anything.

True AI, i.e., "intelligence" will understand contexts and semantics. But it's not something you can get to with brute force.

Have you actually used ChatGPT? The output is way beyond being grammatically correct. It does make sense and is correct on all levels - vast majority of the time. I'm talking about multi-paragraph long semantic coherence, to the extent where it's impossible to tell it was not human produced. Who cares whether it "means" anything in a philosophical sense?

What makes you think our brains do anything other than "blind statistics"? What is a human brain if not a multi-trillion parameter statistical machine, trained on a vast amount of data, implicitly over the billions of years of evolution, and explicitly over decades of individual lifetime? It's actually amazing that a completely randomly initialized model like GPT can be so rapidly trained to communicate effectively - using just text as training material!

GPT still fails on many occasions, such as ignoring context or simply not knowing enough about the world, but these are not fundamental limitations. It can and will be trained to pay attention to context, to query external data sources, and to construct a better world map. It will be trained using multiple data types, like audio and video inputs in addition to text, that way it will learn a lot more about how the world works, physics, etc.

People are constantly redefining "true intelligence" as whatever these ML models can't do yet. That definition has changed a lot in the last ~20 years.

I have used ChatGPT. To say that is is correct on all levels, or even most of the time is like saying that the sky is neon green.

ChatGPT does not communicate effectively; it communicates quite poorly. It simply does a good job of hiding the lack of substance in its output. ChatGPT is syntactically correct, but it's rare to get something that is semantically correct, or meaningful, for anything but the simplest prompt.

To put it bluntly: AI isn't intelligent until it can reason, and that's not something that can be learned from statistics. ChatGPT can say the sky is blue, and regurgitate a Wikipedia article about it, but it will never understand why the sky is blue, or be able to use that knowledge to answer a question that for which the answer isn't already in its dataset. That's not intelligence. That's just a very good search program paired to a wiki.

I don’t know why the sky is blue. Someone once tried to explain it to me and I didn’t quite understand it. What does it make me?

Many of my answers - including most of this one - are regurgitation of information I learned previously. Many, if not most people just follow established patterns when communicating, with no deep understanding involved. Even that deep understanding is also pattern matching, where patterns are simply more complex.

You can easily find examples of an ML model performing multi step reasoning to answer a complex question, just like a human would. I can google those examples for you if you like.

You're changing the goal posts from what I said. I merely said understanding; you've decided this means "deep understanding," whatever that is.

I don't appreciate gaslighting, so I'll be withdrawing from this conversation and you're free to continue your fantasy about ML taking over the world, like VR, blockchain, and big data before it.

This would be true if there were indeed tools developed , tested and deployed in production to do your job. But it’s a few years away so relax. Plus there are so many other things to take your job like H1Bs , pandemic , recessions and investors. BTW where is the labor condition approval notices for AI tools that will prove that there is no qualified American worker available to do its job?
The layoffs are due to higher interest rates. Higher interest rates are due to higher inflation. Higher inflation is due to money printing. Money printing is due to covid and other central bank activities.
> Money printing is due to covid and other central bank activities.

A very small nuanced clarification that you might or might not agree with, but I'd argue that it was the reaction to Covid, and not Covid itself, that led up to the issues we face now and will face for years.

Money printing due to super low interest rates for 20 years now.
The industry has been blowing a bubble for a long time. The bubble is now popping. This is a normal (for this industry) correction that happens every generation or so.
I mean you can read and listen in to most investor calls where majority of the knowledge is shared with the public shareholders.

For example take Microsoft last week:

https://www.fool.com/earnings/call-transcripts/2023/01/24/mi...

Read through the comments on how they see the current AI shift.

> So, that's sort of fundamentally how we view it. And then the other aspect I'd also say is simultaneously investing in this new AI trend because I don't think any application start that happens next is going to look like the application starts of 2019 or 2020. They're all going to have considerations around how is my AI inference performance, cost model is going to look like. And that's where we are well positioned again.

> So, that's how I view it. The market, you all are better readers of, quite frankly, what's happening out there. We can tell you what we see. What we see is optimization and some cautious approach to new workloads and that will cycle through, but we do fundamentally believe on a long-term basis, as a percentage of GDP, tech spend is going to go up.

No, correlation is not causation. It's due to heightened investor scrutiny so they want to be seen as frugal.
Can the AI today really replace labor work of developers?
I wonder how LLMs deal with large codebases. They're great in doing ad hoc things like writing a function or an SQL query, but how are they doing with analyzing your entire codebase and building features on it? I don't know yet. If they can do that, though, I don't know how we can all keep our current salary levels or even jobs.