Frankly, most essays in MBA or almost any social science school can be outsourced to AI with a small context window so teachers need to get more creative; either force people to read more papers (larger context window) or make certain checkpoints throughout the assignment that need to continue context from previous portions in some specific ways.
I think the art of creating own UIs was lost once Windows started dominating; prior to that every single game or app had to create their own GUI from scratch. It only somewhat survived in gaming.
We need something like GNU for AI, "UNAI is not AI" to take on all these business folks working against our interests by making their business models unprofitable.
Here we go, people here were ridiculing right-to-work-on-AI licenses not that long ago and now we have it coming right from the main AI boss, throwing the interest of most of us (democratized AI) down the toilet.
How about this modus ponens chain: "eating less meat" -> "lowering meat production" -> "hiking up meat prices" -> "autoimmune sufferers unable to buy meat" -> "autoimmune sacrificed"
They can be doing limited testing for now. I worked for a company that tracked everything they could get away with and managers routinely enforced minimal threshold of activity so I won't be surprised if some larger bank jumped on the same bandwagon.
I purposefully chose an extreme example to put it onto the radar of everyone pushing veganism everywhere. Also if you really study sustainability, you see that reducing meat production doesn't lead to better sustainability anyway.
If you reduce meat production, you'll hike up prices, likely pricing out those with autoimmune diseases that barely survive already. So technically you are true like in "you can still eat meat in theory", but in practice those folks won't be able. Now where is the good faith in that argument?
Regular publishers have a bit different approach: they contact like 5 potential authors at the same time, ask each to write a book, pay small advance payment (~5k) then wait 1 year and pick 1 book to publish. 1 author maybe gets paid something, 4 authors just wasted 1 year of their time.
They need to find a niche that won't be threatening to any of the big honchos. Possibly marketing themselves as "lawyer AI", "MBA AI", "MD AI" etc. Ideally if there was some common opensource base like Linux distros but for GPT models, constantly getting updated and fine-tuned with anyone simply pulling them and using them when needed.
Not always, some can hit the sweet spot of being simple in all phases of development and quick to learn. They are super rare though and their lessons are forgotten; industry often standardizes on inferior things with better marketing.
One can keep all tensors in the RAM, just push whatever needed to GPU VRAM, basically limited by PCIe speed. Or some intelligent strategy with read-ahead from SSD if one's RAM is limited. There are even GPUs with their own SSDs.
How about reloading parts of the model as the inference progresses instead of splitting it into GPU/CPU parts? Reloading would be memory-limited to the largest intermediate tensor cut.
I guess once we get to 1M characters in the context window, 90% of it would be preloaded by all existing laws in order to avoid any hint of breaking the laws. In areas where two or more laws contradict each other, LLMs will behave like insane persons.
Just watched the first two seasons of Friends and the first season of Seinfeld (never seen them before so a trip to the past) and Seinfeld is so much more intelligent and Friends so shallow and boring (outside some Joey gimmicks). Is there any such show with playful intelligent humor as Seinfeld these days?
Can't this be solved by adversarial training? Just add as many confusing negotiations as possible with the correct outcome sentence/label during fine-tuning and see an LLM to finally get it working?
That's fine, but the point is that EU will be super vulnerable to any AI because they didn't invest in it nor created an environment hospitable for it.
France's industrial might is going down quickly and Germany without cheap resources is losing against China that copied most of its tech. Moreover, China is already outcompeting US in the AI albeit focusing on population control models. EU will become an open air museum for rich travellers.
Deluge is pretty cool, it's like Novation Circuit on steroids. Circuit is a bit better for travel and for people who like solid defaults, Deluge is for tinkerers who grew past Circuit limitations.
That assumes human brains will be able to cope with intelligence tuned beyond their reach. We are already at the stage where many PhDs aren't capable of producing outputs coming from LLMs in some areas. The only approach regular humans used when meeting "superintelligence" were belief systems.
Potato batteries are back? Quercetin would be onion/apple flavor, riboflavin will be liver/egg flavor. Activated charcoal will prevent anything from being absorbed in the gut.