FWIW I wouldnt recommend a 64GB strix halo for local LLM inference. I use a 128GB framework desktop for running qwen-flash-next which it just manages at Q4_K at a reasonable rake of 20-30 tk/s qwen27B is sloooow on it due to strix halo’s lower memory throughput than either a GPU or a integrated memory mac.
Though its possible this advice will change in the next few weeks with the release of Qwen4.
That was the (abandoned) legislation from Labour, the previous act under the Tories raised the limit on sentences by magistrates from 6 to 12 months for a year trial and then let it drop back down to 6 months. Then in 2024 it was raised permanently to 12 months by statute as the 2020 law allowed. None of that had any impact on trial by jury.
No they didn’t, I’ve seen you saying this before and it’s still wrong. In 2020 they increased amount of time a magistrate could sentence someone to from 6 months to 1 year, but the accused could still freely choose to have a jury trial.
The problem with that is that the person who slops out 50 broken shitty games is not going to disclose that they used AI, and the one man band who put genuine craft into their game but used a little bit of AI in the coding or textures will disclose it, so you have the opposite of what you want from the AI filter.
I dont understand the reasoning behind people seem to think prices for components will climb for ever and ever without generating more supply. if RAM becomes obscenely profitable for long enough others will get in on the act and produce their own to get some of those profits. The only reason people arent yet is that everyone thinks there is a bubble pop coming down the pipe (me included).
so the two options are:
Bubble pops-demand from hyperscalers vanishes, prices collapse
Bubble doesn’t pop, more supply comes online due to pent up demand, prices stabilize then come down over time
Neither involve cyberpunk dystopias of cackling supervillians monopolizing PCs
Bit of a silly article, Its premise is that demand doesnt increase cost for electricity because between 1980 and 2000 demand increased by 70% and price didnt rise (inflation adjusted). But it doesnt mention that oil prices were at their peak in 1980 and then fell by more than 70% over the next two decades.
So its hardly surprising that prices didn’t rise despite the growth in demand.
Just think, if you work hard you might be able to get up to the level of understanding arguments that has been displayed there!
Thanks for that, very useful point promoting interesting discussions.
Is a dishwasher a tool for washing dishes or automation of dish washing? If it’s a tool then I don’t see how LLMs aren’t, in both cases you do some set up, tell it to go, and inspect the result for if its acceptable. If its automation then it shows that automation can be a good thing when done well and the moral panic around it is just “its bad when the thing I like as an educated person is automated”.
Are you American by any chance? I ask because in the UK we primarily used yards before metricification which are almost the same length as metres and there wasn’t any sort of long rumbling “ohhh yards are an awkward length” they were just used. I think you feeling that feet are more suitable is probably just familiarity.
So two wrongs make a right? This person had their life horribly disrupted for reason A so an unrelated person should have their life horribly disrupted for reason B?
People should not be subject to arbitrary harm from others and we should be against all forms of it, whether that’s sexual assault, economic coercion or lack of due process.
I think a sensible place for this would be the base models, those are the ones that are trained on the entire internet in order to get a good working model that “understands” things. Then leave the fine tunes on those base model commons as available for companies and organisations to bring their own value to.