Etern4l

Good read, meanwhile Sam is hyping his shit up: AGI is coming soon, AI agents will enter workforce in 2025. Well if the socipaths actually succeed then they will need an army of terminators to defend them from the angry US workforce or a good escape to a "safe country" plan. OpenAI Now Knows How To Build AGI, Says Altman

Did I miss something? What's being meter on desktop HW side? You could build a massive Xeon or Epyc "gold-plated" $50K workstation, and it would be more functional if you had a use case for such a machine. Perhaps what's needed is some offering in between (that doesn't melt down under load = larger socket).

People are just not that technical and need something that's foolproof. Chromebooks seem quite popular, and that's Linux under the hood, but obviously someone would need to commit the resources to making something more open along those lines without having to rely on monetising people's data.

Good news. May try it out between coldplate and IHS. Happy New Year again bro. All is well - just have been extremely busy, but hoping to be able to spend a bit more time on hobbies now. Glad to see the gang is alright.

An unintended consequence of the restrictions is that the Chinese have apparently figured out how to train models 10x more efficiently using the substandard hardware that was still allowed in. That is a game-changer which will lead to further AI advancements and proliferation. We all know the spiel, but it will be very interesting to see what Elon (sorry: Donald, the mastermind, of course), actually do about all this. Here is a hint: https://www.reuters.com/legal/trump-asks-supreme-court-pause-law-that-could-ban-tiktok-2024-12-27/ Yeah, myself and others have fallen victim of the mantra (and all the experience confirming this) that "CPUs don't die". Well, normally they don't unless the manufacturer is desperate enough to overtune everything to be able to show competitive benchmark results, severely compromising reliability. TBF I have put extended loads in the 200-275W range on the CPU, which likely contributed to the degradation. That said I've recently seen a CPU failure rate chart and there is a massive spike for 13/14th gen compared with 12th and obviously a whole range of users are affected. I'm quite constrained on time at the moment and don't anticipate an immediate need for a new rig or a massive overhaul, so would probably just cough up on a 14900KS to get the extra 25W of safe TDP.... the new 15th gen stuff doesn't actually look that bad from my perspective, but indeed doesn't really add a massive amount of value and is certainly underwhelming given the fab process improvements (and I guess one would be wise to wait at least a year before Linux support stabilises and issues are ironed out). Too bad mobos are quite expensive too, so socket/platform changes need more of a justification. In terms of overclocking fun, I would think it will get tougher and tougher since CPUs are more fragile now that they used to be given the improvement in the process, and everything will be more and more fine tuned out of the gate. Thank you it has been an annoyance and a time sink, but I am chalking it up to hobby expenses :) Certainly the first delid was memorable! Yeah, if I had to build a new machine now it would probably be AM5. I would need to check how it performs on Clear Linux though.

Wonder if they ever got audited by some independent trustworthy entity. By construction: probably not, because I imagine no such civilian entity exists in the US, would have to be the Feds.

I mean it's pretty solid advice to provide useless depressed humans with. The more jump the better now. Thankfully this is opt-in, but the issue is that of a lack of a good alternative on the mobile side.

What's the current king of the hill? I've switched to Conductonaut and have not really looked back, although issues such as the price/availability and potential impact on any RMA/resale prospects are real. What's the view on Thermalright SilverKing? https://www.thermalright.com/product/silver-king-1g/

M$s attempts to force people off of Win 10 might get more people to move to Linux. My guess though is they will drop the HW requirements closer to the dealine.

Happy New Year, hope everyone is doing well. Kind of catching up on where Intel is at... what a disaster: massive job cuts ("we have some wood to chop" (C) Pat ), Pat The AI Guy is actually out, 13/14th gen broken without much explanation. Last but not least @Mr. Fox pulled the trigger on an AMD flagship... that's all anyone really needs to know lol My 13900K seems to be on its last legs, has problems booting into Windows, doesn't run at stock voltages etc. Switched back to 12900KS and that's solid. In retrospect the right approach was to RMA when first instability/thermal issues hit. Will try to see if the latest BIOS helps, not sure it's worth even trying an RMA on a delidded CPU. Kind of reluctant to buy a 14th gen CPU, or any Intel CPU for that matter. BTW thanks MSI you dumb fcks for defaulting to 4096W PL1/PL2 limits on these CPUs when we now know that the TDP levels specified by Intel need to be observed to "ensure" (improve the odds really) reliability (125/253W in the case of 13900K) https://community.intel.com/t5/Processors/PL1-and-PL2/m-p/1649853 https://www.reddit.com/r/overclocking/comments/1axepvu/optimizing_stability_for_intel_13900k_and_14900k/ Performance based off of 300W+ power draws it's just a launch review benching gimmick as Intel was trying to hold on to the performance crown using a deprecated fab technology...

Yes, that T7000 looks very similar to the silicone glue I used just to hold things together a little. I did the baking lol - albeit internally only, as I wasn't sure if I want to do LM on the IHS or not. I'm not too worried given how phenomenal the temps are now.

Yes, Apple disappointed me greatly by teaming up with OpenAI on its "Apple Intelligence"...

Hey! :) The CPU actually behaved very well during my relatively short Windows testing, pulled 350W+ on stock BIOS settings, whereas my previous record was around 330W IIRC. Unfortunately it exhibited some crashes on Linux. It's clear Linux, which basically squeezes out another 5-20% out of the CPU, maybe that's part of it. However, these crashes never occurred during the first month or so of CPU use on stock settings, leading me to believe that there is some sort of HW issue. Another thing which started happening earlier was the system would stop seeing SSD drives - that resolved consistently after the repaste.

OK, so this was not a scientific study, but best effort, so I have tried probably 10 different pastes before TFX, SYY, CryoFuze etc. They all performed quite similarly, the difference is that the performance would degrade fairly quickly vs 7958 held up for months. Now, as far 1. Liquid metal on the die 2. Lapped the bottom of the RockIt IHS to hopefully increase pressure 3. Used a tiny amount of this Weld rubber gasket glue instead of the Super glue - rationale, more compressible In my initial attempt I did actually try to assess if LM on the IHS is of benefit, and used SYY - this also exhibited great temps. I then applied LM on the IHS and was initially gutted since the temps were pretty bad, but then figured out there wasn't enough LM and there was not enough contact between the surfaces (evidenced by one-sided patch after removing the coldplate). Very carefully applying more LM resolved the problem. All in all, my best effort before - the machine would run in the 90s undervolted and underclocked. Now, it's in the 90s stock and in the 70s undervolted and underclocked. Probably around 15C difference. Also 15C core deltas before vs maybe 5C now. Again, the caveat here is that some of that performance delta could be due to the IHS lapping and different glue.

Yes, co-founder John Schulman left recently to focus on "alignment". People sitting on a gold mine departing to try and find a way out of the mess (which clearly seems impossible at Open/MicrosoftAI, gives us a strong indication of how grave the danger really is. An excellent video. People might be duped into thinking that Neural Processors are there to run a "personal AI" for them, whereas they are really there to shift some off the processing onto unsuspecting customers' machines, to reduce the load on the servers and vastly increase the amount of data that can be collected. Is Win 11 basically unusable now, or is it possible to effectively disable all the cancerware? The second video is spot on as well, thanks. Well, the reality is these new technologies are being used by whoever can lay their hands on them. Remember that DTs original presidential election was boosted by the marketing / influencing technology developed by Thiel's company and its UK subsidiary Cambridge Analytica. Very robust regulation is required here, but fair to assume this is hardly forthcoming. Meanwhile, some good news: Linux Market Share Hits Record High

Because he is a billionaire. I would love if someone provided an example of a billionaire who can be classified as a "decent human being". They all ultimately crave more money / influence / power / popularity.

Hello, hope everyone is well. I wish I listened to that advice. While the initial PTM7950 (later changed to the 7958 paste) was durable enough and lasted for several months, I can confirm the temps are nowhere near as good as what's achievable with Conductonaut. Temps eventually degraded and the CPU started suffering from instability, so just bit the bullet and replaced with LM on both sides. Unfortunately I'm pretty sure my 13900K actually suffers from the degradation issue as it's not stable at stock speeds (even with temps under 90C), however, good news is that it's stable after downclocking and undervolting which works for me. I think the mental block problem was caused by my imagining LM as this mercury-like kind of thing that will almost surely ends up running all over the components lol Cool and stable after a couple of weeks so very happy and wanted to thank @Mr. Fox for trying to ease me into this :)

No honour among thieves I guess.

Bad Apple news today: Apple wouldn’t let Jon Stewart interview FTC Chair Lina Khan, TV host claims What does this have to do with AI? Well, turns out Jon Stewart decided to join the chorus of high profile humans speaking out about the risks and consequences of AI... and Apple didn't like that, since they are reportedly planning to jump on the LLM bandwagon. In other news: UK and US Sign Landmark Agreement On AI Safety Have not had a chance to scrutinise, but expecting max big tech sponsored BS along the lines of the below quotation: UK tech minister Michelle Donelan said it is "the defining technology challenge of our generation." "We have always been clear that ensuring the safe development of AI is a shared global issue," she said. "Only by working together can we address the technology's risks head on and harness its enormous potential to help us all live easier and healthier lives." Lastly, the artists are fighting on: Billie Eilish, Pearl Jam, 200 artists say AI poses existential threat to their livelihoods

What's the alternative though? Regarding the social and other media messaging: the split is roughly 50/50. I wonder if the omnipresent us vs them narrative is going to lead to a good outcome.... If things continue and escalate, could the worst happen? (prob not the way suggested in the trailer of that movie: Texas in alliance with California lol)

Of course, although the phrase also makes a lot of literal sense when applied in reference to hedonistic behaviour. I don't know too many people that would say that the last thing they want to do before the world ends is buy some NVidia stock lol Well, this is going more mainstream now. -- BTW I'm not sure any current LLM chatbot could pass an informed Turing test by an interviewer who is aware of LLM's shortcomings. -- Meanwhile, uninformed government bureaucrats buckle under the FOMO pressure and embed the technology into the infrastructure: NYC's Government Chatbot Is Lying About City Laws and Regulations -- More good news from our "friends" at M$ and ClosedAI: Microsoft, OpenAI Plan $100 Billlion 'Stargate' AI Supercomputer I mean this is getting completely out of hand now. Remember any time you use the current data-harvesting M$ software and services, or otherwise give them any money: you are funding that. -- Compared to the above, this next piece of minor significance and par for the course (in part because the technology is out there already): OpenAI Reveals AI Tool To Recreate Human Voices So, Closed are "responsibly" releasing a tool which can clone human voice. What could go wrong? This stuff should be criminalised ASAP. Why isn't it, America? --- I saved this real gem for last: Larry Summers, Now an OpenAI Board Member, Thinks AI Could Replace 'Almost All' Forms of Labor So, Closed are getting bolder fast: previously they would publish those reports cautiously claiming about 20% or so of the jobs would be affected, but now they hired Larry Summers (a famous and outspoken economist) to sit pretty on the board and essentially profess the end of capitalism and human civilisation - clearly an excellent business opportunity in the era of AI-pumped disaster capitalism. Stay strong and keep thinking about the future guys, while we still can do something about it.

For reference: Madoff got 150 years, although - granted for an even larger scam.

Good old Patriot Act, and the government power creep just continues. Just like the criminals: they have the motive, you bet they have the means, and technology provides a sea of opportunity. Some sort of very significant change would be needed to stop all this, not clear who could deliver that though.

Just to mention the obvious: the OS will automatically use all of it for disk caching, so any filesystem-heavy use case automatically benefits.

There is certainly a FOMO aspect to that. BTW If there is no tomorrow than the value of NVidia stock should be close to $0. I don't think any adverse scenarios are factored into the pricing. This is not particularly new, but does dismantle the "AI safety guarantee" supposedly provided by containment of advanced AI models in big datacentres which can be powered off or nuked. BTW While we have seen Geoffrey's Hinton's talks here before, they are so good I would like to share another: a short lecture where he develops and delivers the fairly natural premises behind his AI safety concerns: Worth watching the whole thing, it's not overly technical, but fastforwarded to the punch line for those time-constrained.