torben-friis
Could a model then transmit concepts to another?
Say that you have a model that has been biased in some way. could another model "read" the bias straight in the weights? could it edit such a bias on the spot?
What about at-scale generation of custom models? If I want to study chemistry offline, could astra just hand me a local-runnable model tailored to have chemistry knowledge?
Most importantly, this feels very singularity coded. the ability to read and reshape weights by an llm.
In the past you might have been asked to add to the CV that you're into devops when you're just expected to trigger deploy jobs, or to explain that you take data-driven decisions when you're just adding some metrics to a powerpoint to justify a move that would have been done anyway, or to define yourself as a product engineer when they just want you to give a crap about users and be able to interface normally with managers.
Now there's a lot of "forward deployed engineers" that are actually just consultants or people interfacing with third parties, or AI-native devs that are just a guy using claude and maybe touching a couple mds.
This is not to say that significant changes aren't coming to the field, but the job application buzzword game is business as usual.
They now have different classmates they don't have a connection with, they enter the new class having as their identity "the person who failed the previous year", and they'll probably have to sit bored in the classes of other subjects they did pass.
It's almost guaranteed to result in lower engagement.
I'm not denying the opposite is also bad btw. Skipping works sometimes for subjects like history where you'll encounter different topics the next year, but math is impossible to advance without a foundation, and sciences depend on it.
The only good answer is probably breaking the factory-system for schools, where students of the same age are batches going through a pipeline. having students of different levels mixed together like in victorian schools, being able to advance per-subject, something like that.
Growing up in Europe, while preparing for college access exams, my teacher recommended American textbooks for self study.
His reasoning was: "they have the smartest writers in the planet creating textbooks for the world's dumbest students".
It was a stereotypical joke, of course, but I think there was something to the thought. There's a great combination of STEM talent and a cultural expectation for not letting people behind. The latter probably sucks in classroom settings, when you're stuck at the lowest student's pace, but it works for self study material when you can freely skip and advance.
Would Anthropic themselves be paying this much if it was actually billed? It is not obvious.
Is Anthropic a good sample for AI use? Surely not, since they have special needs (dogfooding) and a vested interest in proving that AI usage is the future.
Can we compare their results pre and post AI usage? Close to definitionally impossible.
Are their employees good examples of AI-assisted engineers? Not really, as (one would hope) they're more well versed on the topic than an average eng.
Is it bot-raised or are people voting based on just the title?
I'm saying this as a completely orthogonal conversation to regularisation, ethics, etc. It's just amazing to see how little value these companies place in the general public's perception of them.
Can you think of any other companies selling to the mass marker that care so little about it?
I haven't ever seen a developer, engineer or scientist voice that need. Not sure about media creators since there aren't many in my bubble, but it's kind of a weird claim. Why would people interacting mostly with text need touch? Half of them barely even use the mouse/touchpad.
My Instagram app (besides showing no ads) shows just the stories and feeds of people I follow, chronologically. Just a couple days this way will make you realize how little content there is, my friend's Korea trip has been the top one for like a couple weeks already.
Then you can get the DMs to avoid FOMO and realize how antisocial the whole thing is.
No idea if it works for x, I left that long ago and everytime I've checked out of curiosity I've been shocked at the absolute cesspit it is.
At least if I had to write one, I saw cd is available and seemingly working as expected, and wrote a cd to an arbitrary target I would not expect it to behave that way at all - my expectation would be that either the next command always runs in the cd'd path, or at least that cd errors on a path in a different drive.
What a footgun.
Hard to do otherwise when everything from reviewers to reddit comments are astroturfed.
I'd disagree. The shooting itself is as old as space invaders, but there is a neat mechanic there worth exploring regarding the combinations of powerups and 'character powers' IMO. You have to choose what to fuse/powerup to keep the damage scaling well enough (though the amount of combinations isn't large enough to last more than a couple weekends, I'll give you that).
What's actually happening is people buying knockoffs not knowing they're knockoffs, or no one buying those but the original being buried in the volume anyway.
As for the 'think about the non privileged', AI is a pretty poor answer for them. Once we're at a point where the person isn't able to pay for genuine stuff, they might as well go for piracy and emulators rather than a mindless knockoff. A world where the privileged are given intellectually stimulating material and the rest are fed empty slop is the worst possible scenario.
Clones always existed, kids chose pokemon rather than a dollar store dragomons because pokemon became a household name by the time bootlegs appeared, legitimate shops only sold pokemon and the cartridge itself showed a quality difference.
Nowadays the clone can reach people before the original, the store shows everything equally, and the sprites, videos and description can be good enough to fool people. Steam works as a platform for the purchase itself but not as a discovery mechanism, because there's no "seal of quality" you can consider (other than skipping indies entirely).
No one can review at the scale needed, and it's entirely possible for a great game to have devs that are awful at marketing and get buried. I'm not sure what the solution is, but I think some kind of threshold is necessary for the shops, or at least for an optional filter.
Denshattack!, ballxpit, blue prince, balatro, outer wilds, pipistrello... There's plenty of quality there too!
It's true that the average quality we're exposed to is lower, because there's so many random pet projects and big players take fewer risks. Relying on word of mouth or gaming reviewers you trust is important nowadays. But I'm convinced it's a matter of discovery rather than existence.
I wouldn't dismiss the help of lower friction of entry. Lots of recent Linux and neovim users due to how easy it is to debug an issue or edit a config now.
The larger the regular userbase, the more chances people gradually go deeper into the system.
Perhaps the synchronization between renfe and others is a weak point in the system. Decades of assuming that everything in the network is theirs, increased wear and tear, etc.
- Understanding how effort and class interact (Is a high effort terra above a low effort Sol? Are larger models mostly always better regardless of effort?)
- understanding when (if ever) a new model tends to be better than an older but higher class one. Is an old astra better than a new terra?
If increasing effort and increasing model class produce improved results in a different way, I miss making it explicit in which way. In other words: What kind of issue warrants going up in effort? what kind of issue warrants going up in model class? What kind of issue is unlikely to be solved by either?
For a junior coder, there's usually a path to recommend, joining some open source project or doing some home scripting, etc. Something where you can be useful even if it's not your full time job.
I'm wondering whether there's some similar hobbyist path available here, or whether there's too much of a jump between the educational/pop sci stuff and the professional realm. If it's the second, that's also fine, I just want to know if I'm missing something.
Whenever I've tried to learn about LLMs, machine learning, etc. I always come out with a bad feeling, compared to learning about software.
Theoretical courses make me feel like it's all baby steps, unless you assume a level of commitment equivalent to a math degree. Practical/applied stuff seems to require levels of compute, training data, etc. that make it unapproachable for a home hacker for everything beyond basic exercises like a handwritten digit classifier.
Now this feeling of "impracticality" is perfectly fine. It would be the expected outcome if I was taking a course on classic literature or modern art. But machine learning stuff feels like it could have room for learning-by-doing applications that aren't entirely pointless, so I wonder if there's some way to approach the topic I'm simply not finding.
Did anyone have the same experience? any tips if so?
Not necessarily the reasoning. It might also be that they think their slice of the pie is about to become much smaller, for example.
Would love to have some insider knowledge on this decision.