Let's be real if OAI is losing money on a 200$ subscription with hyper advanced effeciency methods are you really going to save money?
You should also enjoy the free VC money while it lasts. Just like cheap uber rides were great until the vc money dried up.
I've hosted and run a lot of large LLM experiments myself. You are in no way "saving" money doing so. It's also a giant pain to be avoided if possible.
Best thing to do right now is enjoy the cheap AI and when the free money stops use the winning mature open source platform.
Prefect is amazing. Built out an ETL pipeline system with it at last job and would love to get it incorporated in the current one, but unfortunately have a lot of legacy stuff in Airflow. Being able to debug stuff locally was amazing and super clean integration with K8S.
Other guy said it right. These work and are fine but you lose the legacy stuff. If you know your limits and where the eventual system will end up it's great and probably better.
If you are building a expandable long term system and you want all the goodies baked in choose airflow.
Pretty much the same as any architecture choice. Ugly/hard often means control and features, pretty/easy means less of both.
On the surface the differences are not very noticable other than the learning curve of getting started.
+1 to this. other solutions over-promise, under-deliver, poor developer relations and communication, "open-source, but pay us" style open-source, and is indeed a mess
Very typical SV argument that R&D is "complex" and everything else is "simple".
Would it blow your mind if I told you 10yrs ago that we'd have AI that can do math/code better than 99% of humans but ordering a hotdog on doordash would be cutting edge and barely doable?
I don't disagree that "common" tasks are more valuable. I only argue that the argument these are easily automatable is a viewpoint based on ignorance. RPA has been around for over a decade and is not used in many tasks. AI is largely the same, until we get massive unrestriced access to the data for it we will not automate it.
> if I told you 10yrs ago that we'd have AI that can do math/code better than 99% of humans
This not even remotely close to true. Like not even a little bit. I use Cursor and Gemini for work daily and I'd be hard pressed to think AI is a "better" programmer than any professional software engineer. Sure it makes writing code faster and more efficient, because you just click tab and three lines are written for you. It absolutely isn't better than me at coding though.
The claim about math is even more unbelievable than the claim about coding. We still don't have a single theorem proved and published by a LLM without human aid. LLMs barely follow a discussion in basic topology. It's incredibly ridiculous to state they're better than 99% of people. More like 0% of mathematicians and maybe 50% of college freshman.
> We still don't have a single theorem proved and published by a LLM without human aid.
I'm pretty sure that by "do math" the parent was referring to applying math, as one would do in the course of other tasks, and not mathematical research, just as by "code" they likely referred to writing code to solve a problem and not to algorithmic research.
And from my experience teaching & tutoring both math and programming at various levels, I would absolutely agree with the claim that AIs like Claude 3.7 Sonnet surpass over 99% of humans at typical short tasks.
It'll probably take some more time until context, memory and tool-use are improved sufficiently to allow AIs to tackle longer-term tasks effectively, but I'm sure it'll get there. And just as an example of progress, there was recently a post about the first "fully AI-generated paper to pass peer review without human edits or interventions" [0].
I realized today while coding with cursor that AI seems to operate exactly the way I intuit it does, which is it acts like a junior engineer who works by copying existing code but doesn’t understand why. For a lot of tasks that works great, I do this a lot as a senior engineer, but I know when not to. you can’t let it run wild, because it doesn’t know when not too.
Doordash is not the same as traditional online ordering. DD and all delivery apps are 3rd party middlemen that set their own menu prices and operate separately from the restaurant.
Through this kind of obfuscation, they incentivize the growth of things like ghost kitchens, which are basically faceless factories. Nobody would order from them if they drove by one. but on the apps, they are displayed as standalone restaurants.
While I know there have been some issues with working conditions at ghost kitchens, I've also heard tons of horrible stories from regular kitchens, so it's not clearly to me that there's a significant difference on that front.
As for referring to them as "faceless factories", I can't even start to imagine what sentiment that should evoke in me. I don't have an issue buying food products made at actual factories, and have visited quite a few. As such I don't have any issue ordering from a ghost kitchen located in an industrial area that may look like a factory on the outside.
The code that AI is good at currently is exactly common tasks.
Good luck getting it to write a competitive video card driver for Nvidia hardware or anything else that requires actual creative problem solving that isn’t github boilerplate.
It's only common among trained experts. For most people, even simple code is astonishingly difficult.
I personally get more value from AI when coding more complex and novel things. Not fully automated, but English has become the most valuable language for me when coding.
He's right but at the same time wrong. Current AI methods are essentially scaled up methods that we learned decades ago.
These long horizon (agi) problems have been there since the very beginning. We have never had a solution to them. RL assumes we know the future which is a poor proxy. These energy based methods fundamentally do very little that an RNN didn't do long ago.
I worked on higher dimensionality methods which is a very different angle. My take is that it's about the way we scale dependencies between connections. The human brain makes and breaks a massive amount of nueron connections daily. Scaling the dimensionality would imply that a single connection could be scalled to encompass significantly more "thoughts" over time.
Additionally the true to solution to these problems are likely to be solved by a kid with a laptop as much as an top researcher. You find the solution to CL on a small AI model (mnist) you solve it at all scales.
Not exactly related, but I wonder sometimes if the fact that the weights in current models are very expansive to change is a feature and not a "bug".
Somehow, it feels harder to trust a model that could evolve over time. It's performance might even degrade. That's a steep price to pay for having memory built in and a (possibly) self-evolving model.
For a kid with a laptop to solve it would require the problem to be solvable with current standard hardware. There's no evidence for that. We might need a completely different hardware paradigm.
That implies learning. Solve continual learning and you have agi.
Wouldn't it amaze you if you learned 10 years ago that we would have AI that could do math and code better than 99% of all humans. And at the same time they could barely order you a hotdog on doordash.
Fundamental ability is lacking. AGI is just as likely to be solved by Openai as it is by a college student with a laptop. Could be 1yr or 50yrs we cannot predict when.
Strictly speaking I'm not sure if it does require learning if information representing the updated context is presented. Though it depends what you define as learning. ("You have tried this twice, and it's not working.") is often enough to get even current LLM's to try something else.
That said, your second paragraph is one of the best and most succinct ways of pointing out why current LLM's aren't yet close to AGI if though they sometimes feel like it's got the right idea.
In context learning, learning via training. Both are things we barely understand the mechanism of.
RAG is a basically a perfect example to understand the limits of in context learning and AI in general. It's faults are easier to understand but the same as any AI vs AGI problem.
I could go on but CL is a massive gap of our knowledge and likely the only thing missing to AGI.
Long explanation. Simple terms, you can't use a fixed box to solve an unbounded problem space. If your problem fits within the box it works, if it doesn't, you need CL.
I tried to solve this via expanding the embedding/retrieval space but realized it's the same as CL and in my definition of it I was trying to solve AGI. I did a lot of unique algorithms and architectures but Unsuprisingly, I never solved this.
I am thankful I finally understood this quote.
"The first gulp from the glass of natural sciences will turn you into an atheist, but at the bottom of the glass God is waiting for you."
Clear your history often. My youtube is actually incredible, massive variety and useful topics.
I clear it about once every 2 weeks or month depending on how many of the same topics I see.
It works really well in that if you ignore the content you saw before it forces the algorithm to find unique content because it thinks you don't like the stuff you've seen.
That and cleaning your subscription list. Easily the best platform I have as of now because of that.
Long horizon problems are a completely unsolved problem in AI.
See the GAIA benchmark. While this surely will be beat soon enough, the point is that we do exponentially longer horizon tasks than that benchmark every single day.
It's very possible we will move away from raw code implementation, but the core concepts of solving long horizon problems via multiple interconnected steps are exponentially far away. If AI can achieve that, then we are all out of a job, not just some of us.
Take 2 competing companies that have a duopoly on a market.
Company 1 uses AI and fires 80% their workforce.
Company 2 uses ai and keeps their workforce.
AI in its current form is a multiplier, we will see company two massively outcompete the first as each employee now performs 3-10 people's tasks. Therefore, Company two's output is exponentially increased per person. As a result, it significantly weakens the first company. Standard market forces haven't changed.
The reality, as I see it, is that interns will now be performing at Senior SWE, senior SWE engineers will now be performing at VP of engineering levels, and VP's of engineering will now be performing at nation state levels of output.
We will enter an age where goliath companies will be common place. Hundreds or even thousands of mega trillion dollar companies. Billion dollar startups will be expected almost at launch.
Again, unless we magically find a solution to long horizon problems (which we haven't even slightly found). That technology could be 1 year or 100 years away. We're waiting on our generation's Einstein to discover it.
Companies that have no interest in growth and are already heavily entrenched have no purpose for increasing output though. They will fire everyone and behave the same?
On the other hand that means they are weaker if competition comes along as it's expected that consumers and business would demand significantly more due to comparisons.
AI doesn't have institutional knowledge and culture. Company 2 would be leveraging the use of LLMs while retaining it's culture and knowledge. I imagine the lack of culture is appealing to some managers but that is also one of it's biggest weaknesses.
Let's be real if OAI is losing money on a 200$ subscription with hyper advanced effeciency methods are you really going to save money?
You should also enjoy the free VC money while it lasts. Just like cheap uber rides were great until the vc money dried up.
I've hosted and run a lot of large LLM experiments myself. You are in no way "saving" money doing so. It's also a giant pain to be avoided if possible.
Best thing to do right now is enjoy the cheap AI and when the free money stops use the winning mature open source platform.