mistymountains

Again, the problem is custom kernels in CUDA. It’s not straightforward for many applications (LLMs are probably the most straightforward).

It doesn’t sound like having a house will magically make you feel better. Plenty of people are just as exposed to markets as you yet respond differently. I suggest exercise, nutrition, nature, and extended travel if those are not already a part of your life.

These kinds of comments make me think few people have actually tried. My experience has been 1 work day of getting things set up to work the same as before for training and testing (PyTorch).

Unless you develop in CUDA, you can easily train code (e.g. PyTorch) written for training on Nvidia hardware on AMD hardware. You can even keep the .cuda() calls.

I’m a AI Scientist and train a lot of models. Personally I think AMD is undervalued relative to Nvidia. No, chips aren’t as fast as Nvidia’s latest and yes, there are some hoops to get things working. But for most workloads in most industries (ignoring for the moment that AI is likely a poor use of capital), it will be much more cost effective and achieve about the same results.

Why would you want (versus need) a Tesla? It’s no longer an aspirational product, it’s an appliance.

For all you know he had a hard adjustment to the college workload. Maybe his high school was not serious and nobody really challenged him. You all could have interrogated why he may have struggled and shared your strategies for success, lifting him up rather than beating him down. That would make you true “brothers”.

My (biotech, mostly remote) company does this. They may not love it but they realize they have to hire from SF, Boston, NYC etc to get the best ML talent and people expect market salaries / don’t want to up and move if they don’t have to.

The issue is that actually, outside of technical paths, businesses do like to hire business grads as it allows them to do even less training and they usually don’t care if someone is well read or opinionated (possibly prefer the opposite).

Or, just maybe, AGI is a mirage with the bulk of its current utility as a marketing tool for much more realistic, if ultimately mundane, applications. OpenAI, of course, knows this.

You won’t teach yourself by running papers through Claude, and you won’t need to if you went from first principles rather than rushing.

You gave your 9 year old a smartphone?

It’s annoying how much things have shifted now that you can’t really own a performance car without worrying if someone will mess with it.

Yeah we all need to do better and trust China more! Lol.

Is this your diagnosis? That things are more corporate in terms of outcome? Genuinely curious, as a Bay Area native, why UT/Texas can’t begin to compete given how many issues face SF currently.

Cool it with the italics.

Comment datasets are valuable for conversational AI, it’s the same reason Reddit locked down the API I imagine.

It’s probably the best from their training standpoint. The city is quite small relative to others and their only hope is to essentially memorize San Francisco in the neural networks. IMO you could not take a cruise and drop it into any other city and have it work.

They’re too deep, their networks are overfit to San Francisco at this point. Making it work in other cities would require the insane training hours to basically memorize that city.

I’ve seen this stated a ton and it’s not really true. Once trained, the model (except for decoding) is deterministic, and you can enforce determinism fairly easily. ChatGPT is not deterministic at the chat window but that’s not inherent to the model.

Welcome to the real world. Move somewhere more affordable.

Grow up.

I’d be very careful about this.

May I ask, are you a VC? This is a complete non sequitur.

They don’t. They simply assume the model’s most likely output is meaningfully correlated with true rankings even though it was never trained on this task and certainly has not been trained to output the most likely prompt given a prompt in some meaningful order. It’s hogwash.

Agreed. This is a pretty terrible idea.

It’s no longer useful / never was that useful. It’s main function was a shiny toy for investors.

No, you need to show the model is better on this narrow task, not just assert it is because it’s a great general LLM. It’s quite possible you’re correct but just saying GPT is the best, prove me wrong, reeks of a VC or AI bandwagoner.

The training classes in this case are words in the vocabulary in the context of a sentence.

It’s not similar other than that attention relates tokens.