Oder “Fadenmäher” oder “Kantenschneider”

I guess it comes down to whether it’s legal to train image generation models on copyrighted material. Midjourney etc can’t produce a very accurate image of copyrighted characters if those characters aren’t in the training set.

With discord already being pretty shitty, I am interested in what ideas they are coming up with.

Ublock and Sponsorblock make YouTube bearable.

If you’re on mobile, use tubular - it has AdBlock as well as Sponsorblock integrated.

I work in this field. In my company, we use smaller, specialized models all the time. Ignore the VC hype bubble.

Funnily enough, this is also my field, though I am not at uni anymore since I now work in this area. I agree that current literature rightfully makes no claims of AGI.

Calling transformer models (also definitely not the only type of LLM that is feasible - mamba, Llada, … exist!) “fancy autocomplete” is very disingenuous in my view. Also, the current boom of AI includes way more than the flashy language models that the general population directly interacts with, as you surely know. And whether a model is able to “generalize” depends on whether you mean within its objective boundaries or outside of them, I would say.

I agree that a training objective of predicting the next token in a sequence probably won’t be enough to achieve generalized intelligence. However, modelling language is the first and most important step on that path since us humans use language to abstract and represent problems.

Looking at the current pace of development, I wouldn’t be so pessimistic, though I won’t make claims as to when we will reach AGI. While there may not be a complete theoretical framework for AGI, I believe it will be achieved in a similar way as current systems are, being developed first and explained after.

In the case of reasoning models, definitely. Reasoning datasets weren’t even a thing a year ago and from what we know about how the larger models are trained, most task-specific training data is artificial (oftentimes a small amount is human-generated and then synthetically augmented).

However, I think it’s safe to assume that this has been the case for regular chat models as well - the self-instruct and ORCA papers are quite old already.

The goalpost has shifted a lot in the past few years, but in the broader and even narrower definition, current language models are precisely what was meant by AI and generally fall into that category of computer program. They aren’t broad / general AI, but definitely narrow / weak AI systems.

I get that it’s trendy to shit on LLMs, often for good reason, but that should not mean we just redefine terms because some system doesn’t fit our idealized under-informed definition of a technical term.

Ah yes Mr. Professor, mind telling us how you came to this conclusion?

To me you come off like an early 1900s fear monger a la “There will never be a flying machine, humans aren’t meant to be in the sky and it’s physically impossible”.

If you literally meant that there is no such thing yet, then sure, we haven’t reached AGI yet. But the rest of your sentence is very disingenuous toward the thousands of scientists and developers working on precisely these issues and also extremely ignorant of current developments.

No, at least not in the sense that “hallucination” is used in the context of LLMs. It is specifically used to differentiate between the two cases you jumbled together: outputting correct information (as is represented in the training data) vs outputting “made-up” information.

A language model doesn’t “try” anything, it does what it is trained to do - predict the next token, yes, but that is not hallucination, that is the training objective.

Also, though not widely used, there are other types of LLMs, e.g. diffusion-based ones, which actually do not use a next token prediction objective and rather iteratively predict parts of the text in multiple places at once (Llada is one such example). And, of course, these models also hallucinate a bunch if you let them.

Redefining a term to suit some straw man AI boogeyman hate only makes it harder to properly discuss these issues.

You’re approaching this from a point where it’s already too late.

If you’re not capable of taking proper care of your pet, don’t get a pet in the first place. Picking up the shit your dog left in a public place is part of owning a dog.

If your kid has a baseball game the next day, don’t go drinking today. That’s the selfish part. Although I would argue if you do get drunk, you kind of just have to deal with it and go to your kids game regardless.

I think I have some bad news about your ex gf 💀

They probably confused the R1 Qwen distill with something else. Afaik there is no 32b model from DeepSeek directly.

Are you really this braindead or are you just on Xi’s or Putin’s payroll?

Laundry detergent is bad for the environment and, according to some studies, also for you. Scented versions even more so.

Raid and off-site backup.

Or trust in a hosting provider that has backups and redundancy.

I’m not sure. I would assume that clipboard / file sharing etc would still work, but I have never tried something like that.

+1 for KDE Connect.

Especially in OPs use case of transmitting small snippets such as urls, the automatic clipboard synchronization should be very useful.

Don’t need a Linux PC to use KDE Connect, it works perfectly fine under windows too

You seem to misunderstand your own sources. What you cited only proves how utterly insane Russia’s conditions were / are. Of course NATO won’t let Pootin blackmail them into giving up their stations etc.

Russia and brainwashed tankies like yourself always seem to reject the notion that former Soviet nations are actually sovereign and might have an interest in increasing their defensive strength in light of, wait for it, HISTORY.