Demand for AI "Surging"

[jamesbrown]

Thats because Fortran was released around 1958 and Basic was released a few years later in 1963.

The main difference being that Fortran is still widely used today in various numerically-oriented disciplines.

[dsc]

The main difference being that Fortran is still widely used today in various numerically-oriented disciplines.

Any ideas why instead of using something newer?

[jamesbrown]

Any ideas why instead of using something newer?

It's computationally very efficient, so it's often used in demanding numerical simulations, and it's very well suited to certain mathematical operations, such as matrix algebra. There's probably a legacy aspect too as it has been used for decades in aeronautical engineering, space science etc.

[ladymuck]

It's computationally very efficient, so it's often used in demanding numerical simulations, and it's very well suited to certain mathematical operations, such as matrix algebra. There's probably a legacy aspect too as it has been used for decades in aeronautical engineering, space science etc.

Sounds like a sensible case of "it ain't broke so don't fix it".

[rocktronAMP]

Sounds like a sensible case of "it ain't broke so don't fix it".

You can also trust the library algorithms written and tested by computer scientists.

Numerical Algorithm NAG was and still is very popular. In fact, C programmers did inoke these Fortran 77 implementation for accurate computation rather rely on some newer source. Perhaps, by now, the Python libraries have caught up and over taken the Fortran 77.

In what seems like a lifetime ago, when everything in front was very golden, I remember invoking a NAG function LU123X in Fortran 77 or something way back in poly/uni for some Linear algebra matrix decomposition homework. I have forgotten decades long ago that olde cowtulipe about Eigenvectors amd Eigen values from the K.A. Stroud, THE BIG Engineering Mathematics book. I am pretty sure it comes up again in the data science stuff as a platform.

LUD stuff see here https://en.wikipedia.org/wiki/LU_decomposition

[willendure]

Python is pretty easy to learn. One thing about python code is that it is comparatively easy to read, so its also not hard to pick up from other peoples code or online python notebooks like Google Colab and so on. I don't really like python, but respect its position as a getting-things-done kinda language. I used github copilot and found it was really helpful when learning python.

There is also Mojo, which is a new language that extends python and makes it more efficient. There seems to be a lot of excitement around that at the moment, as something that could really improve the python eco-system substantially. python is dog slow, so all the math libs for it are actually written in other languages, mainly C/C++. But Mojo can compile down efficiently, so will help to bring all that stuff back into the language itself where a better system of libraries can be developed.

The tech underpinning Mojo is MLIR, which Chris Lattner of LLVM fame has been leading. Its an IR (intermediate representation) that allows more high-level concepts to be represented, and is ideal for compiling numerical/AI stuff down onto the growing range of AI chips that we are starting to see. The AI surge of today is largely driven by advances in chips, which are getting crazy big. And we will see advances there reflected in compilers and software eventually too.

python -> Mojo -> they will be around for a long time and play an important role in where HPC is headed, so great skills to acquire. I am also interested that MLIR could open up new compiler front-ends for languages I like better than python to be able to play a role in all of this too.

[Fraidycat]

Elon Musk is looking for software engineers for his xAI startup in London.

Perm roles, onesite mostly and looks like a horrible interview process, but stock options/equity are on offer. The company could easily be worth 10s of billions, or even $100 billion+ one day, and your options could end up being worth millions..


Backend role: https://boards.greenhouse.io/xai/jobs/4276949007

Front end role: https://boards.greenhouse.io/xai/jobs/4276959007

[willcodeforbread]

So, I am actually currently working in a contract outside ir-35 doing AI stuff with language models and python and lang-chain and so on. I started out doing some work for the same company as a UI developer. Then segwayed into taking on a new AI project for them. I was the obvious person to pick that up due to my background in AI all the way back in 1998, I was able to pick up how these things work quite quickly after doing a bit of research and reading. I am currently building an AI, but not starting from machine learning point, using LLMs as foundational models and consuming the tool sets that are available in Python land to work with them. I should also get the opportunity to iterate on this work, as its going to take a lot to refine it to the point of being good enough to show to the end client. Its interesting, it is also difficult.

So I am expecting to get some valuable learning and hands on experience in the generative-AI era. I keep an eye on the job market though, because its always good to think about what skills are in demand when picking a direction to go in.

For example, I see quite a few well paying DevOps jobs, it particularly seems to pay quite well. I recently went for an architect role doing a cloud re-platforming since I have the AWS solutions architect certification.

AI is definitely interesting, but I have reached a stage in life where well paying but not particularly mind-bending roles in infrastructure have their appeal too. I know its not sexy, but I do still find the sheer grandady-of-the-cloud scale of AWS an inspiring thing. I even like putting money into my pension these days...

So really, this is me wondering what skill sets to focus on from an outside ir-35 contractors perspective with an eye on the money, or trying to divine where the demand is shifting to.

> Its interesting, it is also difficult.

I'm traditionally a backend engineer, and did an AI diploma with my uni's applied maths dept back in 2001. Started doing more "document data extraction" work from about 2013 onwards, which pivoted to machine learning, which then more recently pivoted to NLP, which got me good gigs in the mean time (receipt scanner for DunnHumby, and last year building a "RiskGPT" for an insuretech (had very good success extracting larger structured outputs from VertexAI Bison, and answering specific questions with smaller 7b models).

But, yes - it's very very difficult work. My nose is constantly in academic papers, reading newsletters/tweets by PhDs, trying to stay abreast of the latest developments (last year is was LoRA, QLoRA, multi-modal, mixtures of experts, more advances in prompt engineering, etc) to a point where I took a 2 month holiday to completely unplug.

However, I'm back looking for my next role, and the market is a bit tough for contractors, especially "backend with ML leanings". I'm by no means a mathematician or trad ML person, so it's hard to find those cross-over roles.

[edison]

Elon Musk is looking for software engineers for his xAI startup in London.

Perm roles, onesite mostly and looks like a horrible interview process, but stock options/equity are on offer. The company could easily be worth 10s of billions, or even $100 billion+ one day, and your options could end up being worth millions..


Backend role: https://boards.greenhouse.io/xai/jobs/4276949007

Front end role: https://boards.greenhouse.io/xai/jobs/4276959007

I passed one of xAI's US based researcher roles to a friend who's out there - £350k basic salary.

As Willcodeforbread mentions, this is difficult work. My friend has been a pioneer in this field for years and I'm staggered at how many research papers he reads. The people who will go and work for xAI (at least in the early days) and similar companies are in the top 1%.

Still, it's encouraging that such a high profile company is starting to recruit in the UK.

[willendure]

https://www.youtube.com/watch?v=NQmN...InternetofBugs

The presenter is saying that AI tools are auto customizing CVs and then spamming all new openings?

I think there may be some truth. One of my cusomters posted a job opening on LinkedIn and got 350 CVs. I found 4 or 5 through my personal contacts/groups that were people much more focussed on the particular skill set sought out. The CEO said he was glad I did, becuase they were much more obviously of the right quality than the 350 obtained through LinkedIn.

The question is how to avoid becoming collateral in this situation? I do feel more and more that:

1. My CV just dissappears into black hole now.
2. All job postings are so specific in the long list of things they want (that are just details you could easily learn on the job), that you are almost forced into this game of CV customization from the get go, rather than leading with a strong general CV, then tweaking for submission to a client to highlight relevant experience and so on.