December 30, 2009

December 30, 2009 by yorksranter 0 comments

CCC vs Wanktanks

Other people’s TDLs. I note that this year’s CCC has devoted some time to the wanktank phenomenon in Germany, and also elsewhere. Volker Bilk’s slides are here; it makes a lot of sense, but there’s quite a bit of German philosophical maundering in there, and I’d have liked to see more case studies/howtos. As far…

December 30, 2009 by yorksranter 1 comment

TDL@TYR

To-do lists are good. Grandad Bill lived by them; we found ones up-to-date for the day before he died. In that spirit, here’s a rough one for this blog in 2010: 1) Finally, end this twin-blog nightmare. I’ve had requests from multiple readers for this. I think I’ll standardise on WordPress and host it myself…

Police insfrastructure, police performance, and crime: Evidence from austerity cuts - WP202416-Police-infrastructure-police-performance-and-crime-evidence-from-austerity-cuts.pdf
I estimate a positive elasticity of violence with respect to distance of approximately 0.09, so that a 10% increase in distance increases violent crimes by roughly 1%. This effect translates in 3 additional assaults and murders in each block per year, for a total of 5,500 additional violent offenses city-wise...614 more assaults and murders are […]
jetmoe/jetmoe-8b · Hugging Face
We used a 96×H100 GPU cluster for 2 weeks, which cost ~$0.08 million // cool moddle. very tiny, too, and an MOE to boot
microsoft/rho-math-7b-v0.1 · Hugging Face
Rho-1 base models employ Selective Language Modeling (SLM) for pretraining, which selectively trains on clean and useful tokens that aligned with the desired distribution.
Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
We follow the sequence of works initiated in “Textbooks Are All You Need” [GZA+23], which utilize high quality training data to improve the performance of small language models and deviate from the standard scaling-laws. In this work we show that such method allows to reach the level of highly capable models such as GPT-3.5 or […]
The Anglosphere has an advantage on immigration
in Britain, second-generation children from the poorest Bangladeshi communities achieve better results at school than the average white British pupil, while Black Britons are more likely to attend university than their white counterparts. In the US, the children of foreign-born parents are now more likely to attend college than those with native-born parents. In France, […]
Post Office paid widow in instalments for silence, inquiry hears - BBC News
During a second day of questioning, former Post Office executive Angela van den Bogerd was shown emails where she and her colleagues discussed hiring a media lawyer after learning that Mr Griffiths was seriously ill in hospital...It also emerged that the then chief executive, Paula Vennells, questioned whether Mr Griffiths had "previous mental health issues […]
Tom Witherow on X: "New at Post Office inquiry: Angela Van den Bogerd admits she got her bonus in 2019 after "lying" to the High Court in the Bates trial. "So despite a finding in the High Court that basically you lied, you got your bonus?" Edward Henry K
and then there's perjury.
Meta value falls $190bn as investors react to plan to increase spending on AI | Meta | The Guardian
A relaxation of that restraint has rattled investors after Meta raised the upper bound of its capital expenditure guidance on Wednesday, from $37bn to $40bn...The share decline follows a record gain in market value by Meta in February, when the company added $196bn to its stock market capitalisation – a measure of a company’s worth […]
Broadcom Scales Connectivity And Performance For Advanced AI Workloads
This is why consumer AI hyperscalers have looked to work with Broadcom to develop what it refers to as an XPU—an AI accelerator that is not quite a GPU and certainly not an ASIC. Instead, it has all the foundational elements of a computational platform—memory, networking, interconnectivity and I/O. The only thing missing is the […]
What factors lie behind system assembly changes for Nvidia's next-gen servers?
Nvidia had been aiming to bring in two or more suppliers for each production stage in its H100-based servers. This strategy could not be fully implemented due to insufficient CoWoS production capacity and insufficient supply of H100 GPUs. This issue has now been resolved, with TSMC doubling CoWoS capacity and outsourcing CoWoS packaging, allowing Nvidia […]