hereonout2

Thanks, sounds different and would certainly instill a certain work ethic.

I'm similar age and have been in paid employment since the age of 13, a lot of different things through farming, manufacturing, retail and catering before I started in tech.

I was always grateful for the work ethic but also important was the experience of being the most lowly employee being made to do the grunt work. I think that gave me the confidence to get on with the job without complaint or too much assistance, I feel that especially helped me progress early on in my tech career.

Still, I'm not sure how I'd feel about my own kids working so young. And in all honesty, don't really think those opportunities would be there for them now anyway.

It sounds like your saying from the age of 5-11 years old you worked 7 days a week as a farm hand.

Could you say more about what era this was and how many how's a day?

Siri came out 12 years ago, wake words are probably a bad example.

Better examples are the magic eraser on my 2 generation old pixel phone or the fact llama2 runs genuinely fast on a Mac mini

Yeah, that's happening right now really? There has been loads of developments in the mobile space already, in many ways lower powered arm devices are way more optimised for AI applications than the current crop of intel machines.

This example, whilst impressive, feels way more in the "Doom running on a calculator" vein of progress though.

Check out my reply to the sibling comment

Things like this give a good overview of the problems being face in productionising ML:

https://research.google/pubs/whats-your-ml-test-score-a-rubr...

Note they start to discuss things like unit testing, integration testing, processing pipelines, canary tests, rollbacks, etc. Sound familiar yet?

The same author has also written this book:

https://www.oreilly.com/library/view/reliable-machine-learni...

I don't see a software engineer's skills becoming redundant in this field, especially if you have a good level of experience in cloud infra and tooling. It seems more valuable that ever to me (e.g. I have worked with ML Researchers who don't grasp HTTP let alone could set up a fleet of severs to run their model developed entirely in Jupyter Notebook).

I have found it helpful to equate myself with the correct tools and terminology in order to speak the right language - there's specific tools lots of people use such as Weights & Biases for "Experiment Tracking", terms like "Model Repository" which is just what it sounds like. "Vector Databases" (Elastic Search had this feature for years), "Feature Stores" - feel familiar to big table type databases.

Reading up on a typical use case like "RAG - Retrieval Augmented Generation" is a good idea - alongside starting to think about how you'd actually build and deploy one.

Above all having a decent background in cloud infra, engineering and how to optimise systems and code for production deployment at scale is a very in demand at the moment.

Being the person helping these teams of PHDs (many of whom have little industry experience) to productionise and deploy is where I am at right now - it feels like a fruitful place to be :)

Presuming you want to work in the field and already have software development experience why not look at the confluence between ML and engineering?

Things like ML ops, application of DevOps, testing and ci/cd in the ml space, how to train across multiple gpus, how to actually host an LLM especially at scale and affordably.

In my experience there are hundreds of candidates coming from academia with strong academic backgrounds in ML. There are very few experienced engineers available to help them realise their ambitions!

Not open source but chess.com app is brilliant, it's a freemium model - you can play for free - but they just keep trying to sell you tip and tutorials.

Also seems to have all the grand masters on there as players, Magnus ignored my request for a game though.

Which comments are these? I seem to be reading a wall of praise here!

I think the drivers are really getting shafted, not sure about customers.

I have not taken a taxi or Uber for years but had to take 3 in the last few months. First a ten minute ride for £6 (UK) felt like what I'd have expected to pay a taxi 20 years ago.

Then a 35 minute ride in traffic for £11 feels like it should break some minimum wages laws after overheads are considered.

Finally a 70 minute ride for 6 through city traffic costing £45. At which point it's on par with public transport costs (trains).

Most meetings are not coding but they are all work. No chance I would be attending without getting paid and by the same token I'd likely get sacked if I failed to attend.

I can't imagine a solo indie dev has those meetings though. So their definition of work is somewhat different.

Also I guess the three hours is a consequence of them having a successful product, which is somewhat akin to winning the lottery. I can't imagine if they were struggling to make rent they'd down tools at 11am for an afternoon of reading and walking.

Indeed! As a more senior programmer at a large corporate, I'm lucky if I only have three hours of meetings per day.

I might be wrong, but sleeping in till 1pm with an 8 month old baby sounds like maybe something else may be going on.

My partner had a very difficult birth that has led to a series of mental health issues. I didn't spot these for months initially, and it took a lot for it all to come to a head and out in the open.

It was easy to miss the issues with a child to care for.

Possibly / quite probably whoever wrote this knows very little about postgres.

I think this is common in companies with a lower salary range, no stock options, bonuses etc.

This is particularly apparent in large cities, where salaries can be almost double at each end of the spectrum. From what I have seen promotion to senior or hiring inexperienced staff at senior level is the only way these companies can retain or recruit staff.

I feel they could have used a more convincing example to be honest. Yeah it's cool it recognises so much but how useful is the demo in reality?

You have someone with a tool box and a manual (seriously who has a manual for their bike), asking the most basic question on how to lower a seatpost. My 5 year old kid knows how to do that.

Surely there's a better way to demonstrate the ground breaking impacts of ai on humanity than this. I dunno, something like how do I tie my shoelace.

> once you start needing to horizontally scale it'll get messy fast.

It gets expensive fast, but not messy, these things scale horizontally really well. All the state is encapsulated in the request, no replication, synchronisation, user data to worry about. I'd rather have the job of horizontally scaling llama2 than a relational database.

It runs fantastically well on M2 Mac + llama.cpp, such a variety of factors in the Apple hardware making it possible. The ARM fp16 vector intrinsics, the Macbook's AMX co-processor, the unified memory architecture, etc.

It's more than fast enough for my experiments and the laptop doesn't seem to break a sweat.

Doesn't this depend a lot on your application though? Not every workload needs low latency and massive horizontal scalability.

Take their example of running the llm over the 2 million recipes and saving $23k over GPT 4. That could easily be 2 million documents in some back end system running in a batch. Many people would wait a few days or weeks for a job like that to finish if it offered significant savings.

Hey if this makes it congratulations! I've never contributed to something so ubiquitous before but often thought how satisfying it'd be.

Like in years down the line, maybe as you retire and hang up your keyboard, you could sit back and smile as you realise your code is still deployed on millions, possibly billions of devices? That the code could far outlast 99.9% of code code anyone has ever written?

I'm interested in this aspect of llms too, are you simply just passing it some input (email, customer message) and asking chat gpt to decide if it's spam? Do you provide any prior examples of spam / not spam or just rely on the knowledge already embedded within the model?

Spam detection is a classic example for classification problems. I guess I'm trying to gauge whether there's an entire suite of traditional problems that llms solve well enough by simply asking a question of the base model. I've found a few areas in my own work where this is the case.

It seems a little superfluous to me, but the paper itself does a little too? Isn't this akin to publishing a paper on say mongodb vs postgres?

Some of the models in TheBloke's repo are optimised for things like ggml, which runs well on Mac chips. Not sure if that's what you mean?

Have been experimenting with deploying to Mac's for inference as they are cheaper and use less power. But also still deploy to a substantial amount of gpus too.

Either way I try to take connections to huggingface hub out of the equation at deployment time.

Perl came out in the late 80s and by the mid 2000's was really on its way out? I hired for my last perl position in around 2006.

Just saying, your definition of "kids today" could well include a decent portion of developers under 45 years old. Referring to this cohort repeatedly as "kids" is also a little cringe.

The obligatory perl replaced awk decades ago comment.

Soon to be followed by the ones saying nobody should be writing shell scripts at all anymore.

The 6502 bit is fun, but the NES PPU not so much! I never got past that part, maybe one day ...

Dockerhub is great for sharing images with the world at large, much like HF is for models. But once you're at the point where you want to host your images / models for business use do you really still want them?

All cloud providers seem to have an image hosting feature, why go through hassle of managing and paying for a docker hub account when you're already on AWS and can integrate better with ECR? The same problem exists for ML model hosting IMO.

Github is totally different, it's a more complex and human proposition. For many github has replaced an entire raft of tools, not just code hosting.

I've been asking myself that for a while. Would they be so popular if they didn't host & serve petabytes of model files for free? How will they monetize that aspect to match this valuation?

Whilst the scale of their model hosting is impressive, the functionality seems pretty basic. The models are just BLOBs in git LFS repos, you're usually relying on knowing which users to follow, then learning the way they name their models and how that naming convention applies to the particular framework and hardware you're using.

As an example the user "TheBloke" is prolific at publishing LLM models for various hardware / framework combos, but look how little the HF interface actually helps navigate or find what you're after: https://huggingface.co/TheBloke

Also, is anybody using HF Hub in "production"? We've deployed a few LLMs now and once we've decided on a model the first thing we do is get it off HF and into our own storage ready for deployment. There seems no reason to tightly integrate HF Hub with production systems given it's a just a bunch of files you can copy and keep.

I remember this being discussed on HN before. It felt a bit like "till receipts are killing us" which in the context of someone flying a 40 year old plane with noted defects might not be a primary concern.

I wanted to buy a thermal printer myself but couldn't find a definitive answer or evidence of the BPA part being a risk. I got one that uses BPS instead - other papers also raised concerns about this, but again I couldn't find direct evidence of health issues associated with the handling of till receipts.

I decided as the receipt printer will be used infrequently and I have been handling till receipts for the last 4 decades I wouldn't worry about it. BPA is also present in food cans and plastic bottles - the link you sent states the concentrations are higher in receipts - but at the same time I don't intend on eating the receipts or rubbing them into my skin.

Generally I'm quite cautious about chemicals in everyday life, I try and reduce the amount in our house using natural products for cleaning for example.

At the same time we're constantly surrounded by toxins so I often find myself having to do this sort of rationalisation to avoid worrying about seemingly everything all the time.

The keyboard is also extra, which bumps the price up to more than I originally thought.

Biggest downside is a 3-4 month lead time though, I'm too impatient for that unfortunately, especially given this would be a bit of a punt for me.