I was shopping for a new microphone this morning and had a moment: I didn't even look at Amazon reviews. I went straight to Wirecutter's recommendation.
That's wild when I think about it. The entire 2010s was built on this promise of democratized information. Crowdsourced reviews. Wisdom of the crowds. Data-driven everything. We were supposed to route around traditional gatekeepers and let the people decide.
Turns out all of it is gamed now. Amazon reviews are mostly bot farms and incentivized 5-stars. The FTC just sent warning letters to 10 companies in December about fake reviews. Bot traffic crossed 51% of all web traffic in 2024 and hasn't looked back. It's only getting worse with AI-generated content and AI agents everywhere.
So I'm back to expert curation. Wirecutter. Consumer Reports. Specialist forums where real humans who actually know stuff talk to each other. The exact model we thought we were disrupting.
The irony: the democratized system became easier to manipulate than the old gatekeepers ever were.
@VicVijayakumar@dhh From what I've seen on the AMD AI Max 395 it smokes earlier devices for AI. So yes, SER8 good enough for some local AI but much weaker than SER9 w AI Max395. Unsure if 3x better performance exactly though
@denaegteemil Spændende med et dansk indslag her! Teknologi mæssigt slår amerikanske @ImpulseLabs_ dem da de ikke kræver specielle pander/gryder som @larsbalker påpeger for Ztove, men Impulse Labs ser dog dyrere ud for kogepladen, så måske same-same 😅
I don't have too too much to add on top of this earlier post on V3 and I think it applies to R1 too (which is the more recent, thinking equivalent).
I will say that Deep Learning has a legendary ravenous appetite for compute, like no other algorithm that has ever been developed in AI. You may not always be utilizing it fully but I would never bet against compute as the upper bound for achievable intelligence in the long run. Not just for an individual final training run, but also for the entire innovation / experimentation engine that silently underlies all the algorithmic innovations.
Data has historically been seen as a separate category from compute, but even data is downstream of compute to a large extent - you can spend compute to create data. Tons of it. You've heard this called synthetic data generation, but less obviously, there is a very deep connection (equivalence even) between "synthetic data generation" and "reinforcement learning". In the trial-and-error learning process in RL, the "trial" is model generating (synthetic) data, which it then learns from based on the "error" (/reward). Conversely, when you generate synthetic data and then rank or filter it in any way, your filter is straight up equivalent to a 0-1 advantage function - congrats you're doing crappy RL.
Last thought. Not sure if this is obvious. There are two major types of learning, in both children and in deep learning. There is 1) imitation learning (watch and repeat, i.e. pretraining, supervised finetuning), and 2) trial-and-error learning (reinforcement learning). My favorite simple example is AlphaGo - 1) is learning by imitating expert players, 2) is reinforcement learning to win the game. Almost every single shocking result of deep learning, and the source of all *magic* is always 2. 2 is significantly significantly more powerful. 2 is what surprises you. 2 is when the paddle learns to hit the ball behind the blocks in Breakout. 2 is when AlphaGo beats even Lee Sedol. And 2 is the "aha moment" when the DeepSeek (or o1 etc.) discovers that it works well to re-evaluate your assumptions, backtrack, try something else, etc. It's the solving strategies you see this model use in its chain of thought. It's how it goes back and forth thinking to itself. These thoughts are *emergent* (!!!) and this is actually seriously incredible, impressive and new (as in publicly available and documented etc.). The model could never learn this with 1 (by imitation), because the cognition of the model and the cognition of the human labeler is different. The human would never know to correctly annotate these kinds of solving strategies and what they should even look like. They have to be discovered during reinforcement learning as empirically and statistically useful towards a final outcome.
(Last last thought/reference this time for real is that RL is powerful but RLHF is not. RLHF is not RL. I have a separate rant on that in an earlier tweet
https://t.co/RMIpFPVpuM)
@LBacaj@pkumarn19 Valid reasons for choosing Unity, just have a look into the Unity pricing controversy, they have backtracked to a better place but who knows the future? Just if you want to turn it into a revenue stream someday :) Starting article: https://t.co/vlwmZuunVF
@KatrineVillar Jeg kan ikke sende DM's, men SparNord skiftede fra Visa/DK til Visa-debit for mig sidste år og har nu 500k per døgn - Måske AL kan samme? https://t.co/s4fXnsfgG4
@VicVijayakumar@samijaber_@ImpulseLabs_ Ahh I didn't load the AGA Range before - very British high end on that. I see how that doesn't really vibe with your current aesthetic. I am a sucker for Impulse Labs design, but being from Denmark I of course also like the clean lines of the New Nordic style 😅
@VicVijayakumar@samijaber_ I would point to @ImpulseLabs_ for high performance and what looks like nice knobs👌
Though you mention other places of not being fancy, but a Europoor like me would think a techy from the US could swing it 😆
@DiogoExMarques@nathudgens What?? The app takes a huge load of CPU power if on the homescreen to display cover art in video format. It does not remember anything if rebooting Windows and it does not sync with your iPhone to control music...
Spotify does both giving the Apple magic feel which Apple fails😵💫
@sdamico@whoismazu Here in Denmark we now use 13A fuses on normal wires (1.5mm^2) and many stoves/ovens will be 16A on 2.5mm^2 (rated for 20A).
I would definitely recommend this stove to people if you figure out decent European shipping 👌
@KatrineVillar Måske også være at nævne at der var deepfake porno af Taylor Swift den anden dag. Swifties var vidst hurtige til at overrende hashtags med andre billeder. Ikke for at tage noget fra præsidentvinklen som helt sikkert også er meget aktiv resten af året.
time estimates from software devs actually make sense if you think of them as "this is how long it would take me the second time if i did this twice in a row"
for 50 years, all world governments bought their encryption machines from the leading encryption firm in Switzerland
turns out the company was owned by the CIA, who was listening in all along 🤣
no civilian has a chance against a tier-1 intelligence agency. most nations don't