May 2, 2014

May 2, 2014 by yorksranter 0 comments

A short post

Schuller pointed out that, if the economic theories of Mises’ book Human Action really are derived by painstaking and valid deductive argument, then it should be possible to set the book out in a formal symbolic form in which all axioms, premisses, and deductions are shown formally and proven. No Austrian has ever done this……

Illegal gas fitter carried out dangerous work in mobile park home - Registered Gas Engineer
The District Judge told him that he could have killed someone and that she was relieved to find out that he no longer holds himself as competent to carry out gas work. He now runs a burger restaurant and has no connection to the plumbing and gas industry // don't think I want his burgers...
AI leaderboards are no longer useful. It's time to switch to Pareto curves.
Thort: "agentic workflows" just mean giving the AI more chances by spending more money. Like the only problem with the Shakespeare project is that we've not got enough monkeys. Also Mistral is being shonky about Mixtral - all the parameters are in memory and memory is what you're mostly paying for
The Fashion Influencers of the French Revolution - The New York Times
That something was an extremely rare and complete set of fashion plates from the Journal des Dames et des Modes, a pre-Vogue, pre-Harper’s Bazaar magazine that launched in Paris in 1797, amid the French Revolution.
Magic Pill by Johann Hari review – weighing in | Health, mind and body books | The Guardian
So I went through the references even more carefully than I normally would, to see if the studies backed up his points. Not all of them did. He argues, for example, that “people who look at social media” can end up with “profoundly distorted” body images. The endnotes cite research from 2022. But in the […]
Samsung beefing up power semiconductor capabilities for AI super cycle
A report from South Korean media outlet ChosunBiz indicated that Samsung established the Compound Semiconductor Solutions (CSS) business unit under the Device Solutions (DS) division at the end of 2023 and has been steadily increasing its elite workforce in power semiconductor design and production. This new business team, led directly by the DS division CEO, […]
Samsung sees memory biz return to profit, upbeat about foundry sales
Samsung anticipates increased demand for AI products as well as traditional server and storage solutions. To optimize its portfolio, the company intends to allocate more manufacturing to server and storage than to PC and mobile, and it is preparing for mass manufacture of HBM3E memory in response to GenAI demand
Water cooling demands to surge as Nvidia shows AI server TDP will continue rising
Air-cooled modules can also be used for cooling above 1,000 watts, however, it requires more space in the cabinet. For example, the H100 requires a 4U high cabinet, and in the B200, the height of the cabinet needs to be raised to 6U...At present, server cabinets are generally 42U, and the height of the cabinet […]
Mass Precision Strike: Designing UAV Complexes for Land Forces - mass-precision-strike-final.pdf
If a new UAV control technique is used near to a specialised counter-UAV EW asset such as the Russian Shipovnik-Aero, then the process of enemy adaptationis significantly faster – typically around two weeks...The development of AI-enabled signals analysis and EW signal development means that these timeframes for hostile adaptation against newly deployed UAV control and […]
L&Q’s house exchange scheme turned our lives into a nightmare | Property | The Guardian
It stirred itself when I first got in touch in December, and contractors were redeployed. It wasn’t until March that a technician discovered the boiler was leaking gas and capped the supply. L&Q’s scanty terms and conditions on mutual exchange state tenants accept properties “as seen”. However, that does not absolve it of basic legal […]
Apple releases eight small AI language models aimed at on-device use | Ars Technica
Apple also released the code for CoreNet, a library it used to train OpenELM—and it also included reproducible training recipes that allow the weights (neural network files) to be replicated, which is unusual for a major tech company so far.