Leaderboard

If token prices are unsustainable and it causes the bubble to pop and big companies to collapse, there's still going to be a number of users who want to vibe code, run agents, and do other things. They should be doing it locally with hardware that would hopefully become cheaper in the aftermath. Such as Strix Halo, DGX Spark, RTX 6000 Pro Blackwell Workstation, M3 Ultra Mac Studio or later (256-512 GB models were discontinued), etc. It becomes a matter of electricity and heat rather than tokens.

@Azaran -

Based on my observations, at the same price point, the equivalent build is a better value, especially when paired with SteamOS—in fact, SteamOS is the one getting more attention. My concern is what will happen if the PS6 or Xbox follow the same pricing standard, which is just a hunch of mine.

anyway i have full access on Gemini pro and gemini apps and other AI model at my workplace but rarely use it lmao Yeah, I also hope those cards wont be ultra-luxurious anymore

I somewhat agree with the part about “making us dumber.” I try not to rely too heavily on AI or show cognitive surrender when using it. Especially when it comes to completely eliminating fact-checking. In my opinion, AI can certainly be used as a second opinion, but the decision ultimately comes down to you. Lately, many people have been touting AI’s ability to boost productivity by 10x—including some of my acquaintances who’ve said the same thing. The “10x developer” buzzword, which has been around for a while, is starting to make more sense. It’s undeniable that there are many types of AI, but I’m particularly concerned about these large language models (LLMs and gen AI). Back when buzzword like “deep learning,” , "reinforcement learning", “expert systems,” and “neural networks” were all the rage, it was clear that many people wanted to dive into the technical aspects especially the calculations behind them. AI development actually teaches you not to hate math like advanced statistics and when it comes to LLMs, there’s SFT, RM, and RLHF behind the scenes. But what’s worrying now is that many people I’ve met (including the developers mentioned earlier) have been more focused solely on the prompting aspect for generaive AI or LLM. Of course, people do tend to focus on specific areas of knowledge based on their needs so i cant fully blame them Whenever people mention “Gen AI,” I always think of generative adversarial networks which were all the rage back in the 2015 and that just shows how out of touch and outdated I am with the world of AI (and this isn’t even my field). 2022-2026 AI progress is basically too much for me and hard to keep up i am concerned with reassurance seeking and validation epidemic, Here's an article that's pretty close to that : When AI Becomes a Compulsion: Reassurance Chatting. Sometimes AI pleases you too much and i think they're optimized to please, not for the truth Basically, I treat AI as a tool for executing mundane tasks and am aware of its limitations.

If we see cuts in DDR5 pricing, SSD pricing, and GPU pricing, then prices can come back down. 1000 USD doesn't have to be the new normal, but you obviously need the AI bubble to pop. Rough but faster pre-built equivalents of the Steam Machine, also using 8 GB GPUs, were available around 550-600 USD on sale last year (ebay ones are refurbished). Steam Machine may not be as DOA as expected depending on where you live, and if people see it as a desirable SFF gaming PC, but I heard it's shipping in low volume anyway. And there are going to be many examples of prebuilts outperforming the Steam Machine around 1050 USD, but especially at around 1400 USD which is outrageous. You can install SteamOS, Bazzite, etc., especially if you get an AMD GPU, and buy a Steam controller separately if you wish. It appears that even an RTX 5050 will outperform the power-constrained 7600M-equivalent in the Steam Machine, so it's very easy to undercut it. RX 9060 XT 8GB is much faster, and it may be possible to build with relatively expensive RX 9060 XT 16GB while matching it on price. I would be surprised if the CPU is bottlenecking much despite being pretty weak, but Valve is not promising more than a single stick of 16 GB, in other words, single-channel memory. They claimed this doesn't affect its gaming peformance but I don't see how that could be the case. The silver lining is you can drop in one more stick for a relatively cheap upgrade.

I would say i tip my hat to those who use there own noggin to get things done. Ai is a wonderfull tool but it should be restricted to problems that we cannot handle due to either disability or as an extension to our own capability. Not as a replacement. Let it handle the tasks that are to great for our own minds or risk the next generations getting dumber. The signs that we went to far are allready there and i see them every day. People since my time have forgot spelling in there own mothertongue, they lost simple math skills and even reasoning to. Simply believing that since it is a machine it is infallible which it is not since it can only be as infallible as the people who wrote it and they have there own agendas which might not be for your benefit. In fact some people who are generally non trustworthy have insane sums tied up in having ai everywhere (i doubt very much these persons just started to feel generous). Like a calculator or a forklift an ai has its uses, but right now it is showed down our throats in every app. I would be very wary and you should be to. So i use it very sparsely and only where i would have difficulty getting done without it.

I just want to share this: https://www.businessinsider.com/uber-coo-andrew-macdonald-ai-token-spending-harder-justify-2026-5 https://lantian.pub/en/article/fun/ai-agent-bankrupted-their-operator-scan-dn42lantian.lantian/ https://www.verysane.ai/p/did-anthropic-ask-for-this By the way, today I was surprised by a website demo created with Vibe Coding by a colleague of mine. One developer was bragging about how cool the new Claude Fable is. Others were complaining about frequently running out of tokens and encountering bugs. I’m often looked at strangely by both my colleagues and acquaintances for rarely using AI for work to create X, Y, Z, and so on. Behind all that, personally, I feel that the excitement and passion were still there before the current LLM craze, back when everyone was still focused on ML buzzwords—how cool Word2Vec, GRU/LSTM, RBMs/DBMs, MoE, Google BERT,probabilistic graphical models, geometric deep learning going beyond euclidean data, and various neural networks were. I followed these discussions on HN throughout 2016–2019/2020. since ChatGPT became a sensation and AI started training on people’s data indiscriminately, it all stopped being fun. From there, I also came across discussions about books authored by Peter Norvig and Marvin Minsky, as well as articles about them. I also learned about Geoffrey Hinton from there. I’m glad I found that forum. The overuse of AI-related jargon has also contributed to that dwindling excitement. What I’m still curious about for the game developers here is: how does AI help with your workflow? *edited : An interesting topic from Ilya Sutskever’s podcast,i remember this came up there : Ilya Sutskever — We're moving from the age of scaling to the age of research

Well now we have all those other places around Lampfirehills and Jewall lake to build.. hmm the distillery sounds like a nice place to build..