4 ms·

The main cause here is that training AI models is extremely expensive so companies are cutting costs where they can so they can make these big investments witho

by kypro 3y ago

The main cause here is that training AI models is extremely expensive so companies are cutting costs where they can so they can make these big investments without upsetting investors. It's also partly related to the higher interest rate environment where investors have made it clear that growth without profit isn't something that will be rewarded anymore. So anything that isn't adding to the bottom line, and isn't likely to add to the bottom line near-term, is now being cut. Unfortunately, it's hard to see when this trend will end. Tech hiring trends tend to be driven by the latest technology trends and AI products unlike past trends do not require large teams of software engineers. Instead AI companies need their capital for compute and only require a handful of researchers who can make improvements to their models. Additionally, we should remember that software engineering isn't like traditional engineering in the sense that software doesn't require maintenance like a physical building or bridge would. Once a software product has been built that's most of the work done. Amazon, Google and Facebook don't need to be constantly rewriting their core products – they simply need to a small team to make minor improvements and fix bugs. Obviously, if a fancy new language or framework comes out they could choose to rebuild parts of the product, but as software ecosystems mature this is likely to happen less and less. Similarly, SaaS companies that built out software products over the last decade are now enabling companies to build complex software products with much fewer engineers. A decade ago it was quite common for companies to have their own eCommerce products, for example. Today most will just use something like Shopify and hire a small team of engineers for the integrations and customisations. I think over the next few years more companies are going to realise a lot of the building has been done and hiring large teams of expensive software engineers is no longer needed. Some industries like gaming where new products are constantly being developed will continue to need software engineers, but companies like Google with established products will likely require fewer and fewer. Those here who like to talk about reskilling in my opinion are not understanding the shift that's happening here. This is less like past market shifts we've seen where a C developer might need to learn C++, or a PHP developer might need to learn React. It's simply that less software engineers are needed today and regardless of what skills you acquire jobs are far more limited. That said, there will obviously remain some demand for software engineers and I'm not even saying that the total number of software engineers will decrease. What I suspect will happen is that demand for software engineers in the coming years will increase much slower than the past, while supply will continue to increase rapidly. Those with good experience and niche expertise will likely continue to find work, but even for these individuals it will be much harder going forward. This is all just my opinion anyway. I'd be interested to hear counter arguments. I hope I'm wrong, but I really struggle to see how demand for software engineers is going to continue to double every few years as it has been for the last several decades. And this isn't even factoring in other demand issue like outsourcing, increasing automation, and the massive influx of new tech talent we saw during the pandemic. I know it's an unpopular opinion here, but I suspect many of us will need to find new careers this decade. Perhaps not because we we can't find work, but because the work we will find isn't going to be as lucrative as in the past. Our wage growth will stall out. Our work conditions will worsen. And many of us might find there are just better more rewarding industries to reskill into.

__loam 3y ago
> The main cause here is that training AI models is extremely expensive so companies are cutting costs where they can so they can make these big investments without upsetting investors. I highly highly doubt this. Shareholders put pressure on big tech companies to do layoffs because they perceived that these companies had more people than they needed and interest rates rose. LLMs aren't that expensive to train for big organizations like Google and Meta that have the capital to train foundation models. Gpt-4 cost $100 million and Google makes 18b in profit a quarter and already has massive pools of computational capital. I doubt inference is a big expense for these orgs too but who knows. > Additionally, we should remember that software engineering isn't like traditional engineering in the sense that software doesn't require maintenance like a physical building or bridge would. This is false. > Similarly, SaaS companies that built out software products over the last decade are now enabling companies to build complex software products with much fewer engineers. A decade ago it was quite common for companies to have their own eCommerce products, for example. Today most will just use something like Shopify and hire a small team of engineers for the integrations and customisations. This is a good point I'm worried you're right about most of the building being done but I can't help but be hugely bitter about it. The software industry has been pretty good about expanding the franchise and giving employees some ownership over the systems they built. That these still hugely profitable systems will continue to exist but that the only people allowed to realize the value of them might be shareholders deeply disgusts me and makes me think if this gets worse, we should nationalize a lot of these firms.
- kypro 3y ago
  > LLMs aren't that expensive to train for big organizations like Google and Meta that have the capital to train foundation models. Gpt-4 cost $100 million and Google makes 18b in profit a quarter and already has massive pools of computational capital. The cost of training of the last generation of LLM models was ~$100 million, but companies like Amazon, Microsoft, Google and Meta still need to buy the GPUs to build out their data centres. Unlike big tech OpenAI just had to pay for the compute, not for the physical GPUs which have rocketed in price. Plus, next gen models are likely to cost even more to train and big tech companies are not just working on one or two of next gen models, but many of them. The cost is easily in the billions. https://www.cnbc.com/2024/01/18/mark-zuckerberg-indicates-meta-is-spending-billions-on-nvidia-ai-chips.html https://www.cnbc.com/2024/01/18/mark-zuckerberg-indicates-me... > That these still hugely profitable systems will continue to exist but that the only people allowed to realize the value of them might be shareholders deeply disgusts me and makes me think if this gets worse, we should nationalize a lot of these firms. If I had to guess we'll see regulators get increasingly aggressive with big tech companies. My guess is that they will continue to grow more profitable in the coming years, but also increasingly likely to be broken up. This isn't tech specific, but the future in general worries me in regards to equality. Many people naively think that the reason workers grew richer in the 20th century was because workers realised for the first time in history they could protest for worker's rights. Similarly people naively think the reason women entered the work force was because of similar protests. In reality technology made reduced the physical requirements of work, and each worker become far more productive through technology so companies could compensate them increasingly more. At the end of the day it only ever makes sense for a company to compensate you equal or less than your productivity as a worker, and prior to the 20th century worker productivity had little changed since the development of agriculture. Today technological advancements aren't making us that much more productive as workers, but it is making companies far more productivity through automation. I don't see this trend changing and it's only going to accelerate with AI. There really is no way to stop the rich growing richer, and trying to stop this by giving that power to the government is risky for other reasons.

by kypro 3y ago

__loam 3y ago

> The main cause here is that training AI models is extremely expensive so companies are cutting costs where they can so they can make these big investments without upsetting investors. I highly highly doubt this. Shareholders put pressure on big tech companies to do layoffs because they perceived that these companies had more people than they needed and interest rates rose. LLMs aren't that expensive to train for big organizations like Google and Meta that have the capital to train foundation models. Gpt-4 cost $100 million and Google makes 18b in profit a quarter and already has massive pools of computational capital. I doubt inference is a big expense for these orgs too but who knows. > Additionally, we should remember that software engineering isn't like traditional engineering in the sense that software doesn't require maintenance like a physical building or bridge would. This is false. > Similarly, SaaS companies that built out software products over the last decade are now enabling companies to build complex software products with much fewer engineers. A decade ago it was quite common for companies to have their own eCommerce products, for example. Today most will just use something like Shopify and hire a small team of engineers for the integrations and customisations. This is a good point I'm worried you're right about most of the building being done but I can't help but be hugely bitter about it. The software industry has been pretty good about expanding the franchise and giving employees some ownership over the systems they built. That these still hugely profitable systems will continue to exist but that the only people allowed to realize the value of them might be shareholders deeply disgusts me and makes me think if this gets worse, we should nationalize a lot of these firms.

kypro 3y ago
> LLMs aren't that expensive to train for big organizations like Google and Meta that have the capital to train foundation models. Gpt-4 cost $100 million and Google makes 18b in profit a quarter and already has massive pools of computational capital. The cost of training of the last generation of LLM models was ~$100 million, but companies like Amazon, Microsoft, Google and Meta still need to buy the GPUs to build out their data centres. Unlike big tech OpenAI just had to pay for the compute, not for the physical GPUs which have rocketed in price. Plus, next gen models are likely to cost even more to train and big tech companies are not just working on one or two of next gen models, but many of them. The cost is easily in the billions. https://www.cnbc.com/2024/01/18/mark-zuckerberg-indicates-meta-is-spending-billions-on-nvidia-ai-chips.html https://www.cnbc.com/2024/01/18/mark-zuckerberg-indicates-me... > That these still hugely profitable systems will continue to exist but that the only people allowed to realize the value of them might be shareholders deeply disgusts me and makes me think if this gets worse, we should nationalize a lot of these firms. If I had to guess we'll see regulators get increasingly aggressive with big tech companies. My guess is that they will continue to grow more profitable in the coming years, but also increasingly likely to be broken up. This isn't tech specific, but the future in general worries me in regards to equality. Many people naively think that the reason workers grew richer in the 20th century was because workers realised for the first time in history they could protest for worker's rights. Similarly people naively think the reason women entered the work force was because of similar protests. In reality technology made reduced the physical requirements of work, and each worker become far more productive through technology so companies could compensate them increasingly more. At the end of the day it only ever makes sense for a company to compensate you equal or less than your productivity as a worker, and prior to the 20th century worker productivity had little changed since the development of agriculture. Today technological advancements aren't making us that much more productive as workers, but it is making companies far more productivity through automation. I don't see this trend changing and it's only going to accelerate with AI. There really is no way to stop the rich growing richer, and trying to stop this by giving that power to the government is risky for other reasons.