Industry reality is that for bespoke software solutions we have been running for decades on non-technical people straight out of a 5 day "boot camp" copy/pasting together "solutions" from SO, or "Sharepoint Configurators" cobbeling together a LoB process where is takes 3 minutes to get to the next screen with a 10% error rate etc.
There are plenty SXM to PCIe adapters and I do expect these to keep up with newer generations as demand surfaces. But apart from homelabbers, will there be enough demand to have a reasonable secondary market outlet? The scenario after all is either a 'bubble burst' meaning thete will be oversupply of new gen, or tech advances rendering previous gen commercially obsolete in a commodity market. The latter you can compare to 2 gen old server equipment on eBay. That is going for 10% of original rather than 50-70%, so basically a full writeoff.
Learning from crypto, a GPU's life foes not end because it fails, althougg ofc that happens, but because an alternative comes along that makes running the old equipment no longer power efficient under the new market conditions.
Even though I do not host on Codeberg, but I do run my own Forgejo, and have no particular interest in Cryptocurrency projects, I consider this a serious red flag as to me it indicates an activist tendency, for lack of a better word, that I do not want from an IT infrastructure provider.
Any benchmark gaining some, even a little, traction before a model release date should at this point be considered tainted. Create your own, never publish it or write about it in any detail.
OpenAI ran a specific red team break out exercise in an environment that was not even air-gaped but connected to the open internet? It breached Hugging Face, and then Hugging Face is 'grateful for the collaboration'? wtf?
All depends on context coherence. E.g. in Claude Code with the "1M context" models I am reluctant to push past 30%. So if that Codex 272K is 100% coherent than it is less of a difference. Still, even then the "hard fail" boundary has moved. Longer running agentic processes will surely run into this more often.
Different era. At the time choosing the JVM and Java language for a multiplatform project that targetted a portable runtime was not controversial at all.
Regulatory capture. You get them to outlaw the part of the competion (safety!) that is unwilling to pricefix and participate in your margin and market division agreements.
"We could have designed our protocols to be minimally compatible with “a nation of laws,” but the tech bros insisted that compromise was treason, and, as a result, we will lose more privacy than necessary"
Unfortunately, no, you can't have a prophilactic that just makes you a little bit pregnant. We used to know this.
Humanity has a proven infinite capacity for 'make work'. What that will mean for you depends on whether you will be lounging, poisoning and backstabbing in Capital City, or breaking large rocks into little rocks in the Districts for your next cup of porridge.
My first thought would be their server side extentions, code excecutoon sandboxes and document RAG search, being on by default? Probably should be an opt-in instead of an opt-out.
Next step? Maps Pro subscription 'blue' plan will guide you through the fastest routes, Maps Free users will be routed on the fast 'green' plan, but might ocasionally be diverted to other routes to 'balance' (free up blue routes') traffic under extreme congestion conditions.
"Ask the model" is usually the point where I realize that either what you need is not a specific answer to their specific question, but a tutor to guide you on a personal journey through the underlying fundamentals of more than one prerequisite areas of understanding, or, the question is too niche or domain specific that I would have to 'ask the model' as wel as I do not know. Now I did observe, not just with LLM but also with Google search, that many people struggle with this and never seem to get good results. They need meta-learning on how to ask the right questions, which I can offer, but often I notice at that point their eyes starting to glaze over as they want a 'direct' solution, me searching or llm guiding for them.
I'm running PBS on a separate mini pc, it is the perfect forever near instant incremental backups that sold me. I did perfectly recover in 20 minutes after a full hardware failiure. Best recovery experience ever.
Let's hope it did not special case (overfit) for the specific tests. One of the failiure modes that in my experience takes the most effort to mitigate for.
"My main project right now is to establish a framework for large-scale, unsupervised code generation in our codebase...
sifting through the unsupervised agent’s (Qwen’s) output"
Ouch. I love my local AI setup with Qwen, but that is a mismatch right there. That model is not the right match for that project. It's like trying to develop a major software solution by just throwing in hundreds of fresh junior programmers and have them spew out random code bits, while what you needed is a good PM, a great architect and a handfull of senior engineers. Might as well pack it in for a year until your model has grown into the ability for those rolls. There is a reason why Opus 4.6 and now Fable dramatically changed the SWE capabilities, and IMHO Qwen is not there yet.