storus

I doubt they did any distillation as Hinton defined it (requiring logit access). They most likely ran a bunch of prompts/conversations and captured the results. Those conversations already missed thinking tokens, replaced by some confusing quasi-summaries. Then they took those and ran basic SFT or maybe DPO if they had competing responses. As there is no copyright on the output of AI, I am not sure where is the "covert industrial distillation" part of the problem.

Old Reddit is the only way to read text there for me so if that's login-walled or gone I am done with them as well.

Isn't it right next to the largest agglomeration in Poland? Also close to two (maybe three) other countries connected via highways?

DeepSeek V4 hallucinates like crazy and often forgets explicitly mentioned parts of the context. I guess compressing tokens and cherry-picking attention comes at a cost.

I would rather see them releasing 3.7-27B, 3.7-122B or their 3.8 versions. Qwen/QwQ were always about the best available local inference at home.

Africa is going to be full of old Versaces, Balenciagas, Guccis and Valentinos.

Windows Phone aesthetics was repulsive to most people at that time; we finally got TrueColor 4k screens and all MS could do was to use 10 colors everywhere and start the flat fad that destroyed UX on most systems. What a waste.

Capacitive screens were out of possibility for them as Apple bought 2 year production in advance, a trick Tim deployed repeatedly in many areas. MeeGo had a chance but US funds didn't want to allow a state where an EU company would rule the fastest growing market of that time and their darling MS slips into irrelevance and its trojan horse killed it off quickly.

Well, not exactly, for example if I search for LEGO I get the original but not the 10x cheaper compatible knockoffs.

They are orthogonal; preference optimization like RLHF can be done on the base model which can later be quantized, or it could be done on a new LoRA that is then converted to QLoRA.

Can you do the inverse as well? Like Amazon but only the knockoffs?

One could have anticipated XBox getting slowly destroyed by appointing a young clueless outsider as its boss.

First they tried to approve software patents during an agriculture and fisheries council session, now they are bending procedural rules to hack it in before summer vacations. Some weird form of democracy™.

Math is static, CS is dynamic. In math you describe static idealized "worlds", in CS you look at any discrete dynamic process in detail via algorithms. Many folks doing math can't understand algorithms, and many coders can't understand math. Just ask a mathematician what does A = A + 1 mean. There is some inherent impedance mismatch.

I tried to use it while biking but it's extremely confusing for bike trails (colors and parallel routes display, not highlighting real bike trails). I vastly prefer mapy.com in just showing a simple red dashed line for any official bike trail. The elevation is also displayed in 50m increments which is too little for bikes.

I find it slightly easier to do mental work after a hard workout, especially when it's cardio, but not by much.

From my experience, workout draws exactly from the same pool as mental effort, so after a tough day at work/school, there is little left for a workout and vice versa. Instead of brain spending its energy on thinking, it spends it on muscle/movement coordination.

Why are they using neural nets to model observed behavior (different parts activated) and then applying them to biological neurons that work completely differently? Real neurons communicate using precisely timed spikes and each neuron does a bunch of local computation as well.

LUKS still keeps unencrypted header on the harddrive; real men use plain dm-crypt instead! Plausible deniability compatible.

There is a difference between how things are intended to be used and how they are used. Code review has been weaponized this way even at Google.

No, the real reason for the code review is to protect the moat of senior engineers/leaders that would nitpick on minute details of code while ignoring the big picture to make sure they can gatekeep any promotions and their competition.

Do you think the current AI automated menial work and left only the fun parts? It seems like the opposite, it took any fun from coding and left the drudgery of debugging code one didn't write intact.

This looks like something from 80s sci-fi and far behind what Unitree has.

Try to run your prompts through Claude to pinpoint any ambiguous parts that can be interpreted in multiple ways, or self-contradictory sections. I typically resolve any prompt-ignoring issues with that.

Tabata is the craziest workout ever, with Tabata sprints I couldn't feel my legs 3 minutes in and after 4 minutes all I could do was to vomit while shaking on the ground. 7-minute workout with as many reps as possible (even if not in perfect form) helped more overall.

This sounds like adding way too much complexity for something that will likely be covered fully by the next gen of frontier models within a single prompt. It also makes it all opaque and difficult to trace.

Metabolic theories of mental illnesses and cancer are seriously understudied.

So a minimum of 3x RTX Pro 6000 to run 1-bit at ~76% accuracy or MacStudio 512GB RAM to run 4-bit at ~97% accuracy.

I noticed the same with me after a few years working; my solution was to take all hard mathy online MS degrees/grad certs available one after the another. Now I can understand how does the generative AI math work while designing an electric motor for my own robot. Though the latest AI can do it all better than me.

This just follows the usual tech trend that when something gets finally good enough, an unskippable package deal appears as the only way to use it.