It's deeply saddening to see how fixated people are on the here-and-now, while ignoring the terrifying rate of progress, and its wide-ranging implications.

We've gone from people screeching "deep learning has hit its limits" in 2021 to models today that are able to reason within limited, but economically relevant contexts. And yet despite this, the same type of screeching continues.

It's the same kind of people who claimed human flight would not be possible for 10,000 years in 1902. I just can't understand how narrow your mind has to be in order to be this skeptic.

Or the same kind of people who claimed Theranos was a scam, or that AI in the 70s wasn't about to produce Terminator within a few years, or that the .com bubble was in fact a bubble...

Maybe some of us aren’t actually impressed with the “progress” since 2022? Doing well at random benchmarks hasn’t noticeably improved capability in use for work.

Does that mean it will never improve? Of course not. But don’t act like everyone else is some kind of moron.

> Maybe some of us aren’t actually impressed with the “progress” since 2022? Doing well at random benchmarks hasn’t noticeably improved capability in use for work.

Then perhaps you should strive to explore outside of the realm of benchmarks. I've been lucky in that I've seen legitimate value-adding uses in my workplace that simply were not possible pre-2022.

> But don’t act like everyone else is some kind of moron.

Not everyone - just the ones that ignore a trend of continued progress towards a goal, claim we can't achieve the goal, yet offer no meaningful explanation as to why we can't get there.

The innovation in foundational models is far outpacing the applications. Other than protein folding (which is not only LLMs AFAIK) I haven't seen a single application that blows my mind. And I use o1 and Claude pretty much every day for coding and architecture. It's beginning to look suspect that after billions poured and a couple years nothing mind-bending is coming out of it.

Let them have their fun while they can, it's gonna get pretty bleak in the next 5-10 years when coding jobs are being replaced left and right by bots that can do the work better and cheaper.

Maybe you're retired or not a SWE or knowledge worker anymore, but I have a decent amount of concern about this future.

As a society, we have not even begun to think about what happens when large swathes of the population become unemployed. Everyone says they'd love to not work, but no one says they can survive without money. Our society trades labor for money. And I have very little faith in our society or the government to alleviate this through something like UBI.

Previously it was physical work that was made more efficient, but the one edge we thought we would always have as humans - our creativity and thinking skills - is also being displaced. And that too, its fairly clear that the leaders in the space (apart from maybe Anthropic?) are doing this purely from a capitalist driven profit first motivation.

I for one think the world will be a worse place for a few years immediately after AGI/ASI.

Why not just hire out of bangladesh though?

Bangladeshis don't have perfect English communication skills or work 24/7, and the most skilled engineers there have generally already moved overseas, so the only remaining ones aren't top-tier (or aren't cheap).

On the other hand you can go from concept to finished product with a bangladeshi team today. And you can’t do that with copilot. So why the risk for western devs when you can already get close enough to their quality of work for a fraction of the price, yet they still exist? Clearly there are other factors involved beyond raw cost to explain this.

People said that 5 years ago.

Personally I welcome companies to try. I'll laugh all the way to the bank when they need to rehire all the talent they've laid off at a premium to fix the security holes and damage caused by LLMs. Assuming that they survive having their entire company ransacked in the first place.

5 years ago, GPT-2 had just been released. People were definitely not saying this 5 years ago lol.

people have been saying that since before i even started my degree over a decade ago.

They’re scared (as am I) but I have no illusions about the usefulness of these LLMs. Everyone on my team uses them to get their tickets done in a fraction of the time and then just sit around till the sprint ends.

Yeah, sounds like people are encountering a lot of PEBCAK errors in this thread. You get out of LLMs what you put into them, and the complaints, at this point, are more an admission of an inability to learn the new tools well.

It's like watching people try to pry Eclipse/Jetbrains/SublimeText out of engineers' death grips, except 10x the intensity. (I still use Jetbrains fyi :p)

Well thats the argument most people here are making - that current LLMs are not good enough to be fully autonomous precisely because a human operator has to "put the right thing into them to get the right thing out."

If I'm spending effort specifying a problem N times in very specific LLM-instruction-language to get the correct output for some code, I'd rather just write the code myself. After all, thats what code is for. English is lossy, code isn't. I can see codegen getting even better in larger organizations if context windows are large enough to have a significant portion of the codebase in it.

There are areas where this is immediately better in though (customer feedback, subjective advice, small sections of sandboxed/basic code, etc). Basically, areas where the effects of information compression/decompression can be tolerated or passed onto the user to verify.

I can see all of these getting better in a couple of months/few years.