Fair point then about the arguement around safety. For me the bigger issue is control. Cars with kill switches and conditions to use is a slippery slope. Just look at what’s happened with software and media. Don’t want to have to pirate my car or load custom firmware so I can use it as I want.
I don’t think there is a car where the seat belt is tied to anything besides a little notification beep. Seems like a different situation if the “safety” feature dictates how the car is used.
Do you still use WASM? I’ve been exploring the space and wasn’t sure what the best tools are for developing in that space.
I appreciate the response!
I’ve definitely used tools like LocalStack before and when it works it’s great, but sadly doesn’t usually provide a 1-to-1 replacement.
Seeing your different approaches is helpful and I will have to see what elements I can pull into my current projects!
Hey OP, it looks like you’re the author of the post? If so I’m curious how you handle cloud services like AWS or Azure when taking this approach? One of the major issues I’ve run into when working with teams is how to test or evaluate against cloud services without creating an entire infrastructure in the cloud for testing.
Definitely sounds like it could be real. If I had to guess their mounting a drive (or another partition) and it’s defaulting to read only. When restarting it resets the original permissions as they only updated the file permissions, but not the mount configuration.
Also reads like some of my frustrations when first getting into Linux (and the issues I occasionally run into still).
This is just the estimates to train the model, so it’s not accounting for the cost to develop the system for training, collecting the data, etc. This is just pure processing cost, which is staggeringly large numbers.
LLMs do suck at math, if you look into it, the o1 models actually escape the LLM output and write a python function to calculate the output, I’ve been able to break their math functions by asking for functions that use math not in the standard Python library.
I know someone also wrote a wolfram integration to help solve LLMs math problems.
Not sure if you’re serious, but they were making a joke because Intel, who makes chips, is a competitor to TMSC the chip manufacturer from the article.
So they played on that relationship by treating the word Intel in your “thanks for the Intel” comment as meaning the company.
All the evolution in AI right now is just trying different model designs and/or data. It’s not one model that is being continuous refined or modified. Each iteration is just a new set of static weights/numbers that defines it’s calculations.
If the models were changing/updating through experience maybe what you’re writing would make sense, but that’s not the state of AI/ML development.
This approach has been around for a while and there are a number of applications/systems that were using the approach. The thing is that it’s not a different model, it’s just a different use case.
Its the same way OpenAI handle math, they recognize it’s asking for a math solution and actually have it produce a python solution and run it. You can’t integrate it into the model because they’re engineering solutions to make up for the models limitations.
Yeah I was really disappointed when I heard it was a cloud solution. I think it’s due to complexity of python runtime environments, but I doubt msoft minds the opportunity to take more control.
My biggest issue with it is that it runs everything in the cloud. So you’re shipping your data to Microsoft and have latency to run anything. Seems insecure and added complexity to get a bit more out of excel.
I mean I can list a lot of things AI (and I’ll limit it to Transformers, the advancement that drives LLMs) has enabled:
AI isn’t a scam, but it’s being oversold and it’s limitations are being purposefully hidden. That being said, it is changing how things are done and that’s not going to stop. We’re still seeing impacts from CNNs, one of the major AI/ML breakthroughs from over a decade ago, make impacts.
What was so obvious in that instance was the board members trying to push him out were calling out the lack of openness OpenAI was trending towards. They were literally calling him out for not upholding the vision of why the company was founded.
All the engineers clearly saw their payday slipping away and revolted for that reason. Can’t say I blame them, but it was a scenario where the board was actually doing the right thing and everyone turned on them for profit.
Originally all their work was supposed to be published and shared with the world, hence the “open” in OpenAI. However somewhere along the way they made a for-profit break off of the original company and started pulling everything in that direction.
This is what I was going to say.
Microsoft actually ported their keyboard to android, called “Microsoft SwiftKey” or similar. It’s a great keyboard, but apparently now has copilot ಠ_ಠ
Yeah I read another article on this and it’s very unclear what was illegal. If I had to guess they’re getting him on the technicalities of the process rather than on the actual streaming.
Edit: so I looked it up and realized wire fraud is “electronic” fraud, not bank wiring - Online definition
Which given the way the guy did it definitely seems to meet that definition.
Threads all run on the same core, processes can run on different cores.
Because threads run on the same core, the only time they can improve performance is if there are non-cpu tasks in your code - usually I/O operations. Otherwise the only thing multi threading can provide is the appearance of parallelism (as the cpu jumps back and forth between threads progressing each in small steps).
On the other hand, multiprocessing allows you to run code on different cores, meaning you can take full advantage of all your processing power. However, if youre program has a lot of I/O tasks, you might end up bottlenecked by the I/O and never see any improvements.
For the example you mentioned, it’s likely threading would be the best as it’s got a little less overhead, easier to program, and you’re task is mostly I/O bound. However, if the calculations are relatively quick, it’s possible you wouldn’t see any improvement as the cpu would still end up waiting for the I/O.