Yes, see above. That's exactly what the "temperature" variable sets.

No, it's technically wrong. Take an LLM, set temperature to 0, input a prompt, and it will produce exactly the same output every time. Hallucinations/confabulations are a separate question entirely, they speak to model utility/accuracy, not to whether it's deterministic or not.

Can we please stop saying that LLMs are 'nondeterministic'... It's misinformed. And it matters (because some people associate nondeterministism with some sort of magic, and others associate the word with inaccuracy -- neither of which are true). These models are clearly, architecturally, mathematically deterministic at a fundamental level, although this determinism is hidden from users. Let me explain. If you take any recent LL model (GPT, Llama, Gemini, etc...) and run it with 'temperature' set to zero, then, for a given input, it will produce exactly the same token, every single time it runs. That's not surprising, the weights are baked into the nodes of the model and calaculate the same output every time (of course excluding very rare things like uncorrected cosmic-ray bit-flips). Now, this has the side effect of introducing some undesirable behaviour (rigid responses, and looped outputs in certain contexts), so the 'temperature' parameter is used to mix things up a bit (this parameter is usually hidden ftom users in cloud models). When tokens are returrned on an inference run, there is an array of tokens provided with probabilities of being the 'right token'. The temperature value just uses a pseudorandom algorithm to return less likely tokens a proportion of the time (give the user the 2nd or 3rd most likely token some of the time instead of the most likely). To be clear, this has no effect whatsoever on the deterministic nature of the model, it's just a fudge-factor applied on the results after the token array is returned. The LLM models don't do something unpredictable or nondeterministic, they are just like any other computer algorithm run on a deterministic Turing machine. The debate around what sort of informational structure is represented in these models is interesting, but please don't lean on nondeterminism to explain your biases (positive ot negative about LLMs)... it's just incorrect.

Actually the "moderate consumption" myth has been thoroughly debunked. For more on this look up the "j curve hypothesis". It turns out that properly conducted studies (particularly in scandinavian military cohorts) establish a near linear dose-dependent effect of alcohol. That is, any amount of alcohol consumption increases the risk of many illnesses (cancers included). This is now well established, but the alcohol industry continues to push for softer messaging like, 'a glass of red wine with your meal is healthy'. We now categorically know this is not true. In some case series analyses, alcohol consumption accounts for approxinately one third of attributable cancers. In many ways, it is a similar public health threat as smoking. I do believe in a person's right to make bad choices, but I also strongly believe that people should be properly informed amd that governments have an obligation to prevent misinformation (especially when it is for-profit). Some common-sense changes that will result (eventually) from the evidence are: - limit or prohibit alcohol advertising (esp. at sporting events) - include warnings about cancer, brain damage and liver damage on labelling - consider public health when licensing venues (even without longer-term consequences, there are strong correlations between alcohol-serving venue density and violent crimes) - standardise measures, prevent free-pouring and enforce responsible serving laws which hold venues liable for serving drunk individuals -maintain strong controls on drink-driving These are all common-sense methods of harm reduction. Feel free to keep drinking ), but please don't delude yourself that it is good for your health!

You're kidding right? KVM blows every other hypervisor out of the water with its performance. It may not meet your use case but for serious emulation it smokes the competition. In our business we run hundreds of VMs with not a single machine crash and get near-native speed, especially with the improvements in virtiofs. KVM also does a great job with hardware virtualisation including PCI passthrough and can handle resource changes on-the-fly (like increasing memory to the virtual machine). As far as bridged networking, set up your bridge with standard linux tools and then just attach your VMs to the bridge. We don't use a GUI for KVM because libvirt is really easy from the command line and the config files are simple to work with. For enterprise use, it's simply the best.

Because there's no proper user separation in terminal server, resulting in one user's locking affecting all other uses (read the report). I'm not defending Chrome, but the MS architecture is not fit for purpose.

Come join us on https://soylentnews.org/

Yep, Sailfish is great, posting here from a Sailfish phone at the moment. It's much better than the malware that iOS amd Android have become. Otherwise I'd probably use LineageOS but very happy with Sailfish on Xperia X.

If you're looking for an alternative to the mini. Try an intel nuc hackintosh. I just built 4 of them and am very happy with performance (16GB RAM, EVO 850 SSD). http://www.tonymacx86.com/mave...

To be honest I haven't tested out battery drain properly but screen dimming, sleeping, keyboard dimming work fine. If you're getting battery drain what do you think the cause is? CPU not throttling maybe? I presume that should be fixable.

Texpad is awesome too, the best LaTeX editor I think.

Latest Linux mint Ubuntu edition works great on Macbookpro (2014 13" retina). Fast, portable, working with an easy install, dual boot of you want to. These are pretty affordable 2nd hand too.

True, infinity x none = none

Hell is just fine for me if it means not believing in your fairy god. The trap of religion is the true hell. Eventually I'm sure society will become civilized, the end of death sentences and the realization that crime is a function of a broken social order will be two signposts on the way.

Regarding molten salt, I assume you refer to the liquid fluorine molten salt proposals. I looked into this too but it does create a lot more complexity in separation of neutron poisons from the molten stream. I agree that in an accident the design looks better and the fact that it has self regulating qualities is good. However, the molten salt reactors don't get around the largest issue with fission power, expensive to handle waste products. Reprocessing doesn't reduce the volume of waste much (only helps to reuse the plutonium and also reduces mining to some degree). The story of the nuclear fuel cycle was of clean energy but it has left is with quite a number of very polluted sites and a huge bill in dealing with the waste sitting in casks and pools. Perhaps fast neutron burner reactors could help (molten lead looks particularly interesting) but I think that with increasing political instability and the track record of humans being poor managers of super complex systems, nuclear may just not be out friend for a couple more generations. Solar is quite capable of providing is with the energy we need (yes, base load too if we use another type of molten-salt!) and it's not prone to catastrophic failure.